Google AI Mode: Availability Tracker, Timeline and Features

Barry Schwartz reported the confirmation on 29th Oct. The spokesperson's words: "Yes, this format is part of our ongoing experiments with ads in AI Mode."

The statement continues: "Shopping Ads can appear alongside organic visual search results within carousels for commercially relevant queries. All sponsored listings are labeled as 'Sponsored'."

Search Engine Land carried the same quote the same day. Neither report names a market, a device or a start date.

The sighting is the 28th Oct row in Tests seen in the wild. Google calls it an experiment, not a rollout, so the availability table is unchanged.

Shashi Thakur, Google's VP and GM for Search Ads, announced the expansion on 28th Oct. The post names Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines and Singapore.

Its words: "Now we're scaling ads in AI Mode" in those markets. Brazil, Mexico and the United Kingdom are planned for Q4 2026, among "additional locales".

In the US, Google is "launching multiturn ads", which allow "ads in follow-up queries in AI Mode". Ads "will be clearly labeled as 'Sponsored'".

A Google spokesperson told Search Engine Roundtable that multiturn ads "show up on responses to follow-up user queries". The post does not say whether they extend beyond the US.

The post claims click-through rates "are already similar between sponsored and organic product list items within AI Mode". Its footnote dates that to US internal data for 20th Aug to 3rd Sep 2026, with no figure.

Google's Google Ads help page, read on 29th Oct, says ads in AI Overviews and AI Mode are "available in English in the United States and select English-speaking markets". Its FAQ lists the same ten countries, plus Indonesia for AI Overviews only, with Pakistan "text ads only".

The same page says existing Search, Shopping and Performance Max campaigns "are automatically eligible to serve ads within AI Overviews and AI Mode". Text and Shopping ads can appear; local inventory and vehicle ads are experiments.

The post gives no start date per country and no languages beyond the help page's "English". It does not say whether the word "test" still applies.

The availability table now lists eleven countries with ads. The May 2025 entry carries the test's history.

Google launched the preview on 28th Oct. The post, by Chips Mistry of Search and Mai Lowe of the Gemini app, calls the platform "a unified way for more app developers to bring their capabilities directly into Gemini and Search".

Developers connect "a cloud-hosted MCP server or Android MCP using the AppFunctions APIs", then register through the Google Connected Console. "Before review, developers choose whether it will go live in Search, the Gemini app, or both."

Google is "opening access in phases" and will add "more developers, markets, and Google products over the coming months". The developer page takes applications and names no market.

The post counts "more than 50 Connected Apps between Search and Gemini". Adobe, Experian, Localiza and Peloton "are live for users in Gemini", and Google says it will bring them to Search "in the coming weeks".

A Calm integration is coming "across both Search and Gemini". The post says "Search" throughout and never names AI Mode.

For AI Mode the gate is unchanged. Google's help page, read on 29th Oct, still says connected apps are "currently available in AI Mode in the US, in English".

The July entry covers the launch with Instacart, Canva and YouTube Music. The Search arrivals "in the coming weeks" are a watch item.

Alphabet reported its third quarter on 28th Oct. Sundar Pichai's edited remarks say: "Queries in AI Mode are on average three times longer than traditional searches, and one in five AI Mode searches globally use voice, images, or documents."

The Q4 2025 figure was "roughly 1 in 6" voice or image queries, per the February entry. The new figure is global and adds documents, so the two are not one series.

Pichai also said Google has "crossed over 50 billion visual searches per month". Year-over-year query growth is "the highest it's been in the last five years".

He named no AI Mode user count and no AI Overviews figure. Google is building one experience "across AI Overviews, AI Mode, and traditional Search", and is "focused on building the best Search user experience, followed by monetization".

Agentic features built for the Gemini enterprise agent are coming to Search "as a next step", with no date. The ten-country ads expansion was announced the same day.

Google Search and other revenue was $63.35 billion, up 12% from $56.57 billion, per the 8-K exhibit.

The earnings call transcript was not posted when this entry was written on 29th Oct. The update below quotes it.

Google's privacy policy changed on 16th Oct. Google's own version comparison shows the insertions about connected data.

A new paragraph says users "may choose to allow certain Google services, including services like Search or Gemini Apps" to access data from Photos, Gmail and Drive. It adds: "Search may process some of the data that you allow Search to access to personalize content, including ads."

A standing sentence gained an exception. Google shows no personalised ads based on Drive, Gmail or Google Photos content "unless you choose to make data from these services available to other Google services that support personalized ads".

A third insertion names the feature. "If you choose to allow certain services to process data from other services, including for features such as Personal Intelligence, the same data may be used to personalize your experience."

The connected-apps help page for AI Mode now carries an ads question. Ads "will not be based directly on your Gmail inbox or your Google Photos library".

But Search "may reference summaries, excerpts and inferences from these sources", and "That same info may be used to show you ads". A second answer says advertisers "will not have access to your Gmail inbox or Google Photos".

Neither answer was on the page's archived copy of 2nd Oct. The connection covers "Google Search (incl. AI Mode, Discover), Maps (incl. Ask Maps), Shopping, Flights, Hotel, Translate, and News".

Search Engine Land reports the change as a Google test of ads in AI Mode for users who connect apps. It says Google has announced no advertiser controls, targeting options or wider rollout.

The word "test" comes from that report. Google's two pages name no market, no ad format and no advertiser control.

The only date is the policy's effective date.

Ads in AI Mode were a US test from May 2025 until the ten-country expansion of 28th Oct. The availability table's Ads row now notes the connected-data clause.

Hema Budaraju, Google's VP of Product Management, announced the changes on 15th Oct. The post calls them "updates across AI Search", so they cover AI Overviews and AI Mode alike.

First, inline links. Google says it has "made continued and substantial increases to the presence of inline links" and is now showing them "higher up in responses more frequently".

No figure is given for the increase, and none for the share of answers that carry inline links.

Second, a test. For tasks with "only one or two official places on the web", Google "will be testing a new display" with those links "in a dedicated section at the top of a response".

Third, the timely-article carousels are "starting to roll out" to "additional locations and languages globally". The post names no market or language.

Fourth, the links at the end of some responses get "image thumbnails and a more clickable design".

On Preferred Sources, "the number of people who now use" the feature "has grown by more than 7x since the start of the year". There is no base count and no method.

The AI Mode-only item is a test. "This week, we started testing clickable creator and publisher names on video results that link directly to their Search profiles in AI Mode."

Search Engine Roundtable notes those links "go to Google pages, the Search profile page, first".

Google's Search profile help page, read on 16th Oct, names only Discover and Search. AI Mode is not listed.

Search profiles are open in the US, Canada and India, per our Discover page. The carousels expand on the August link carousels.

Click data stays absent. Google told Search Engine Roundtable's Barry Schwartz it will "introduce additional metrics over time" in the AI performance report.

Not said: how many more inline links, where the carousels now run, or whether clicks change. The AI Overviews side sits on our AI Overviews page.

Google announced the Universal Cart on 15th Oct. The post says it "is rolling out in the U.S. across Google Search and in the Gemini app".

Suresh Ganapathy, Senior Director of Product Management for Consumer Shopping, wrote the post. Google previewed the cart at I/O on 19th May.

AI Mode is named as an entry point. The example saves "a handbag from Steve Madden to your cart in AI Mode in Search".

The cart draws on "the 60+ billion continually refreshed product listings in Google's Shopping Graph". It sends price-drop and back-in-stock notifications.

Checkout is "powered by the Universal Commerce Protocol (UCP)". Users pay with Google Pay in the cart or transfer items "directly to the merchant's cart".

Google names Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair and "many Shopify merchants" for the simplified checkout. Merchants "will remain the seller of record".

The May preview promised Search and the Gemini app "this summer, with YouTube and Gmail to follow". The launch post names neither YouTube nor Gmail.

The post names no market beyond the US, no language, no plan and no help page. 9to5Google reports it as "US-only for now".

For merchants, the cart is the consumer side of the UCP integration hub from 6th Oct. The Gemini app side sits on our Gemini page.

Google's launch post of 13th Oct names Search. Argon is available to "Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search".

Robby Stein, VP of Product for Google Search, posted on X: "Gemini 4 Argon is now rolling out globally in Search!"

He says it "brings big improvements to tackling nuanced, multi-step tasks in AI Mode". Subscribers switch with the "+" icon, as they did for 3.8 Flash.

Argon was introduced on 30th Sep for cyber defenders only. That post promised a release "starting with paid API customers and Google AI Ultra subscribers"; the launch post adds Pro.

Search Engine Journal notes that every menu addition since Gemini 3 Pro has been a Flash model. Gemini 3.7 Flash came in Aug and 3.8 Flash in Sep.

Two things did not change. The default is still Gemini 3.5 Flash, and neither post names a language.

Google's AI Mode help page, read on 14th Oct, still describes only "Pro" and "Fast". Its Pro option is Gemini 3 Pro.

Links are present this time. Search Engine Roundtable reports Gagan Ghotra's sighting that Google "made sure links there this time around", after 3.8 Flash dropped citations in Sep.

Search Engine Journal had not seen Argon in a Pro account's menu at 12:17 pm ET. Stein's post says it is rolling out.

The practical use is the one we noted for 3.8 Flash. Running a query on two models separates a model change from a ranking change.

The same model reached the Gemini app the same day. That is a separate entry.

Adweek's Trishla Ostwal reported on 12th Oct that Google is testing ads for branded prescription drugs. The surfaces are AI Mode and AI Overviews, and Google confirmed the test to Adweek.

The spokesperson's statement reads: "We're testing standard text and Shopping ads in AI Mode and AI Overviews, which we recently expanded to include certain pharmaceutical products."

Per the spokesperson, the pharmaceutical testing "began this month" and is limited to the US. The ads appear in sections labelled "Sponsored".

Adweek names no start day. This entry carries the report's date.

Healthcare had been kept out of these surfaces. PPC Land reported on 1st Jun that Google's documentation for the Dec 2025 expansion excluded healthcare, finance and gambling.

The first healthcare test followed Google Marketing Live in May. Ginny Marvin, Google's Ads Liaison, wrote in a LinkedIn thread, per PPC Land, that "PMax, AI Max with search terms matching, Shopping, and broad match are eligible to serve ads in AI Mode (and AI Overviews)".

Marvin added a creative limit: "the first iteration of the test is limited to creatives that don't use pinning or text disclaimers".

Search Engine Roundtable's Barry Schwartz asked Google about policy changes on 13th Oct. He was told there were none, and he could not find a pharmaceutical ad himself.

The status table keeps ads at "US only" and testing. Nothing here names a second market or a general availability date.

Search Engine Land's Anu Adegbola reported a beta-labelled "Direct offer" option in the Google Ads Assets interface on 7th Oct. It sits beside Promotions, Messages and Lead forms.

Google told the outlet the asset "will only serve on campaigns using AI Max or text customization". Offers can be set at account or campaign level.

The advertiser supplies an offer name, a final URL and a short description. Google's AI decides when the offer is relevant and can write the offer text.

Incentives carry a type and a value, with optional single-use promo codes uploaded by CSV. Daily limits cap the total value or the number of claims.

In AI Mode the promotion appears under a sponsored result as a "Claim one-time code" option. Selecting it shows the code and a button to the advertiser's site.

Direct Offers began as a pilot. Google announced it on 11th Jan 2026 with Petco, Samsonite, Rugs USA and Shopify merchants among the first brands.

That post promised discounts first, then "bundles and free shipping". The asset Search Engine Land saw takes several incentive types.

This is a format sighting from a named reporter with a Google comment, not a launch post. No Google post or help page marked the asset when we looked on 8th Oct.

When this was written, ads in AI Mode were a test on Google's own account. On 28th Oct Google expanded them to ten more countries, and the availability table now says so.

Google announced Nano Banana 2.1 on X on 6th Oct as "our latest image generation and editing model".

Google says it "outperforms our previous models across the board, with notable leaps in visual design, mask-based editing, and subject consistency".

A second post in the thread lists where it goes: "Nano Banana 2.1 starts rolling out today across @GeminiApp, AI Mode in Search, @GoogleAIStudio, @FlowbyGoogle, @stitchbygoogle, Google Ads, and Gemini Enterprise Platform."

Robby Stein, VP of Product for Google Search, confirmed the Search side: "Nano Banana 2.1 is rolling out in AI Mode in Search!"

His instructions: open the Google app and "tap the" banana "icon under the Search box to get started creating".

We found no blog post or help-page change for the model on 7th Oct. The X posts are the record.

Neither post names countries, languages or plans. Google's post says "starts rolling out", so availability may lag the announcement.

AI Overviews are not in Google's list of surfaces. Search Engine Roundtable reported the AI Mode rollout and notes the same absence.

The Gemini app gets the same model. That sits on our Gemini page.

Google's Merchant Center help page now describes a "UCP integration hub". UCP is the Universal Commerce Protocol behind native checkout in AI Mode and Gemini.

The page says integrating with UCP lets a merchant "implement a checkout button on eligible product listings in AI Mode in Google Search and on Gemini".

The rollout is gated by country. "The Merchant Center integration is gradually rolling out in the U.S., with Australia and Canada to follow next year."

Merchants choose what to switch on. The page says they "can select UCP capabilities such as cart transfer to merchant site, native checkout on Google, and identity linking".

Onboarding needs the technical implementation first, then an interest form. Google says it notifies merchants it selects.

A sandbox then validates the UCP profile, cart transfer and native checkout APIs.

Search Engine Land reported the change on 6th Oct. By its reading, Google renamed the help page and removed an earlier FAQ about UCP checkout for "select U.S. merchants".

The help page carries no date, and we found no Google post. This entry is dated to Search Engine Land's report.

For a US retailer, the hub is where September's auto-enrolment becomes a choice. Cart transfer keeps the sale on the merchant's site; native checkout closes it on Google.

Google's merchant loyalty program help page now names two more surfaces. It reads: "Loyalty benefits are also eligible to be shown on AI Mode and Gemini Apps (US only)."

The benefits are the ones merchants submit through Merchant Center: member prices, free shipping and redeemable points. The page also lists free listings, Shopping ads and local inventory ads as places they show.

Barry Schwartz reported the edit at Search Engine Roundtable on 2nd Oct, with a before-and-after comparison. By his comparison, the AI Mode and Gemini wording was not there before.

We could not check that ourselves, because the help page carries no date and the Internet Archive holds no copy. This entry is dated to his report.

The page applies to merchants in 13 countries, with a separate points programme in Japan. Roundtable says India, Italy, the Netherlands, South Korea and Spain are the additions.

Google matches shoppers to memberships through Loyalty Customer Match. Merchants upload member lists through the Merchant API, or let Google reuse their Customer Match lists from Google Ads.

The page says the option is open to "all merchants, including non-advertisers". It adds that "experiences may differ on desktop and mobile devices".

The page does not say how a member price appears inside an AI Mode answer. We found no Google post about the change.

For retailers, the feed behind free listings now reaches AI Mode as well. It joins native checkout, which reached AI Mode and Gemini through Merchant Center in September.

Google Search VP Robby Stein announced the rollout on X on 28th Sep. He calls the feature "info monitoring capabilities in AI Mode".

It is "now rolling out to all users in Search globally", he wrote. He points people to the Google app to try it.

You tell AI Mode what to watch. Search then keeps checking "sites, forums and social posts", Google's real-time data and the Shopping Graph.

Stein puts the Shopping Graph at "60B+ products". Search will also suggest monitoring tasks while you search.

His three examples are local and commercial. They are new restaurants nearby, holiday activities for children, and back-in-stock and price-drop alerts.

We logged this feature in June as information agents. It launched for AI Ultra subscribers, with AI Pro access promised for the summer.

Stein now says "Ultra and Pro subs have been using" it. We found no Google post that dates the start of Pro access.

The post leaves gaps. It lists no languages, and it does not mention links inside the updates.

Stein's June post had promised links to the web, Search Engine Journal notes. SEJ adds that neither post says how Search picks its sources.

For site owners, monitoring shows pages to people who did not run a fresh search. Local businesses and retailers are the examples Google chose.

The Regional Court of Aachen granted a temporary injunction against Google Ireland on 24th Sep (case 8 O 268/26). It bars Google from repeating the false claims through AI Mode, the legal news service beck-aktuell reports.

The doctor had been chief physician at a clinic and left it in summer 2026. A search for his name and the word "Freistellung" returned an AI Mode answer, his lawyers at Höcker write.

The answer tied him to the sexual offence of a former senior employee of the same clinic. The articles AI Mode linked as sources do not name the doctor, the firm says.

He asked Google in late August to remove the answers. Google replied that it had reviewed the complaint and would take no action, per beck-aktuell.

The court held that Google provides the AI search and is therefore "auch für die insoweit generierten Antworten verantwortlich", also responsible for the answers it generates. The quote is from Höcker's account of the order.

It followed the Munich ruling of 28th May. The answers are "eigenständige, neue und inhaltliche Äußerungen", the court said: Google's own new statements, not relayed third-party content.

The court added that Google could have checked its own AI against the doctor's objections and found the statements false.

Three caveats. The order was issued without hearing Google and is not final, The Decoder reports, and Google can contest it.

Google has not commented on the case in anything we read. The doctor is not named, and the order itself is not published.

The Decoder puts the penalty for a breach at up to 250,000 euros per violation. Neither German source we read states a figure.

Aachen is the second German court to hold Google liable for a generated answer, and the first for AI Mode. The Munich case concerned AI Overviews about two publishers.

The claimant's firm first reported the order on 7th Oct. We added this entry on 22nd Oct.

On 22nd Sep, Google Ads consultant Menachem Ani posted a Google email headed "Your products are now eligible for native checkout".

Search Engine Roundtable transcribed it. The email says the merchant's Shopify store was matched to its Merchant Center account.

That match enables native checkout in AI Mode and Gemini.

The key line is the default. "Eligible products are automatically included. There's nothing for you to set up."

The email names the way out: "you can turn off native checkout in your Shopify admin under Sales channels > Agentic".

Shopify's help page matches. "Agentic storefronts is active by default for eligible stores."

For Google AI Mode and Gemini, "if you have direct checkout activated, then customers can complete their purchases in the channel's Shopify-powered direct checkout without having to leave the AI channel".

ChatGPT is different on the same page. Shopify calls it "a discovery-focused referrer platform", with purchases finished on the merchant's own checkout.

Search Engine Land lists the eligibility conditions. The store must be US-based and sell to US customers.

It needs a valid Merchant Center account with eligible products in it.

The checkout runs on the Universal Commerce Protocol, which Google announced on 11th Jan 2026. Google's 16th Sep shopping post says UCP "already enables direct checkout for hundreds of thousands of brands and retailers".

Merchants keep the order. The email says "You manage and retain full ownership of your customer relationships, data, and sales."

Per Search Engine Land's reading of Shopify, some features do not work in direct checkout. They include some checkout blocks, product bundles, custom pixels and client-side Google Analytics tracking.

Google has published no blog post about the enrolment itself. The email and Shopify's help page are the record.

For a store owner, the practical change is where the sale closes. A shopper in AI Mode can now buy without reaching the merchant's site, unless the merchant turns it off.

Google announced Gemini 3.8 Live on 15th Sep. The post lists Search Live under "For everyone" in the rollout, starting that day.

Rajan Patel, Google's VP of Engineering for Search, posted on X the same day. "3.8 Live is now powering real-time conversations in Search Live."

He lists three changes: "more helpful responses, complete with web links to dive deeper", switching languages mid-conversation, and "more natural, free-flowing interactions".

His post ends "Available now for everyone globally!" The blog post gives no country list of its own.

Search Live is the voice and camera layer inside AI Mode. It reached every AI Mode country and language in March; our earlier entry covers that.

Google's post promises "step-by-step, real-time troubleshooting help powered by Gemini 3.8 Live" inside Search Live.

For site owners, the web links line is the one to watch. Google has not said how a spoken answer chooses which links to show.

The larger model, 3.8 Live Extended Thinking, goes to the Gemini app instead. That is on our Gemini page.

Laurie Richardson, Google's Vice President of Trust and Safety, wrote on 9th Sep that Google is "rolling out interactive tools across Search, including in AI Mode and AI Overviews, and in the Gemini app to connect you with election information".

The tools cover "polling locations, registration deadlines and other official data provided by state and local governments and Democracy Works". They also carry "real-time voting results from The Associated Press".

Democracy Works said the same day the information "will be available to voters on Google Search, including in AI Mode and AI Overviews". It adds that the data "will roll out to the Gemini App soon".

The two posts differ in tense. Google says the tools are rolling out; its partner says they will be available.

Neither post says when the tools went live, what an answer looks like, or how citations change. Neither names a market beyond the US midterms.

The post also covers SynthID checks in Search, limits on misleading election images and election-ad verification. Those are outside this tracker.

We found the post on 2nd Nov 2026, the day before the election, and date this entry to it. Claude announced a Democracy Works integration on 14th Oct.

Denise Ho, Senior Director of Product Management for Search, announced the features on 9th Sep.

The AI Mode part is account linking. "You can link your Yahoo Fantasy or Sleeper account with Search and ask for customized insights through AI Mode."

Once linked, Search "can bring in the context from your live roster and all of your leagues". Google names start/sit recommendations, waiver targets, draft assessments and a weekly league recap.

The gate is stated once. "The ability to connect fantasy football apps with Search is now available in the U.S. in English."

These are the first apps added since Instacart, Canva and YouTube Music launched on 16th Jul. Google promised more partners then and has now named two.

The post does not use the connected-apps label. It describes the same flow, an option to "securely link" an account inside an AI Mode answer.

The rest of the post is classic Search. A Live Game Feed for professional football shows play-by-play updates, video highlights and "AI-powered insights", on mobile in the US in English.

College teams and more countries are promised "later this month". A new matchup carousel and league-leader stats roll out globally on mobile.

Nothing in the post names AI Overviews, a model or a market beyond the US. The availability table below adds the two apps to the connected apps row and keeps it US only.

Google's help page for the report, About AI performance insights, now states availability in the present tense.

The report is "currently available for English-language queries for Merchant Center accounts in Australia, Canada, India, New Zealand, and the United States".

When we read Google's announcement page on 1st Aug it said the report "will be rolled out" to those five countries "in the coming months". The May entry therefore called it announced, not shipped.

The help page is undated, so this entry takes the date of the first sighting we could verify. Brodie Clark posted at 11:45 UTC on 8th Sep that he was "now seeing it for a broader set of client accounts".

He describes three sections added since his July screenshots: AI Search intent, AI Search terms and AI attributes. A "Products showing" column counts how many of a domain's products appear in AI searches.

The help page matches. It lists four metrics: share of voice, competitors' average share, frequency and products showing.

The trend cards are top terms, popular attributes and top search intents. Each links to a full list.

Share of voice now has a definition: "your number of AI impressions divided by the total amount of impressions across your defined competitors in Merchant Center and yours". The May announcement had none.

Two Google pages now disagree.

The new page names AI Mode and AI Overviews only. The announcement page still names the Gemini app and still says "coming months".

Two limits are stated. The traffic filter "is strictly limited to organic AI traffic", so paid ads are excluded, and the report still counts share and products rather than clicks.

Search Engine Roundtable reported the expansion on 9th Sep.

Anthony Higman posted on LinkedIn at 20:24 UTC on 3rd Sep. Google was "serving regular pure search campaigns with restrictive match types in AI Mode".

He said he had confirmed it "through multiple campaigns and tests on our side". Other advertisers replied that they were seeing the same.

Ginny Marvin, Google's Ads Liaison, answered in the comments on 4th Sep. "Exact and phrase match keywords are eligible to serve text ads in AI Mode."

She added a limit. "Note however that this is limited to cases where there is explicit and direct user intent."

Her comment also points advertisers elsewhere. "AI Max and PMax for Search are designed to capture complex conversational intent and access our latest ad formats (like Highlighted Answers)."

Search Engine Roundtable asked her directly the same day. Her answer: "This is a small experiment that recently kicked off."

That is the whole of what Google has said. No market, no share of queries and no end date are given.

The 2025 test drew ads from existing Search, Shopping and Performance Max campaigns using broad match. Exact and phrase match keywords are new to this surface.

For a search advertiser the difference is practical. Inside the experiment, a keyword-controlled campaign reaches AI Mode without moving budget into AI Max or Performance Max.

One question under the post went unanswered. An advertiser asked whether AI Mode traffic gets its own reporting row.

Google had not replied when we read the thread on 5th Sep.

The status table below keeps ads at "US only" and testing. Nothing in the exchange names a market.

Two days after Gemini 3.8 Flash reached AI Mode, Google has confirmed a citation problem with it.

Gagan Ghotra posted at 23:51 UTC on 2nd Sep that he was testing the model on informational queries. For "a lot of TOFU level queries it's not showing any citation".

His screenshots show AI Mode answers with no links. Glenn Gabe posted further examples.

Search Engine Roundtable collected them on 3rd Sep and asked Google.

Robby Stein, VP of Product for Google Search, replied on X at 00:05 UTC on 4th Sep. "This isn't working as intended, and we'll roll out a fix soon."

That is the whole statement. It does not say what the fault is, how many queries it touches, or when the fix lands.

Two limits on what the sightings show. The examples are top-of-funnel queries chosen by SEO testers, not a measured sample.

And the model is a menu option for AI Pro and Ultra subscribers. The default, Gemini 3.5 Flash, is not part of the reports.

The point for site owners is the confirmation itself. Google treats an AI Mode answer with no sources as a defect, not as a design choice.

We will record the fix when Google says it has shipped. Until then, a 3.8 Flash answer without links is a known bug.

Google introduced Gemini 3.8 Flash on 2nd Sep. This time the launch post names Search.

The model "is available to Google AI Pro and Ultra subscribers across the Gemini app, AI Mode in Google Search and Gemini in Google Sheets".

Robby Stein, VP of Product for Google Search, posted on X: "Gemini 3.8 Flash has landed in Search".

He says it is "available today in AI Mode for Google AI Pro & Ultra subs around the world". Subscribers pick it from the model menu behind the "+" icon.

Nick Fox of Google Search put the cadence in one line: "Less than three weeks after 3.7".

Google describes the model as its "most intelligent workhorse model". It claims gains over 3.7 Flash on coding, agentic tasks and multi-step reasoning.

Three things did not change. AI Mode's default is still Gemini 3.5 Flash, and Google has not said whether that moves.

The gate is the same paid one as for 3.7 Flash. A reply asking for 3.7 Flash on the free tier drew "looking to bring these capabilities to more people as we can" from Stein.

And Google's post makes no claim about AI Overviews. Neither post names a language, so we cannot say whether the English-only limit on 3.7 Flash applies.

Search Engine Roundtable reported both posts the same day.

The practical use is the one we noted for 3.7 Flash. Running a query on two models separates a model change from a ranking change.

Update, 4th Sep: testers found the model answering some queries with no citations. Google confirmed it is a bug and promised a fix.

Google recorded the change as a dated note rather than a post. The control's help page now opens "As of August 31, 2026, we've rolled out this control to all websites worldwide", and the Search and Discover report pages carry the same sentence for "these insights".

The June launch post gained the note in place. No new blog post, dashboard row or help-pages changelog item came with it.

Search Engine Roundtable was the first outlet to notice, on 1st Sep.

The before-and-after is dated by the archive. A 21st Aug snapshot of the control page still reads "We're rolling out this control to a subset of website owners, allowing for thorough testing before rolling it out further".

A snapshot of the Search report page at 02:35 UTC on 31st Aug still carried its subset note. So the edit landed later that day.

Every property now gets the report described in June. It shows impressions for AI Overviews and AI Mode, broken down by page, country, device and date.

A separate report covers generative AI features in Discover. There are still no clicks, no queries and no position, and Search Labs experiments stay excluded.

Two qualifications stand in Google's own text. A property that has not received enough impressions in AI features sees no report.

And the help pages keep the line "Not all properties have access to the report, as we're rolling out over time" directly under the new worldwide note.

The control sits under Settings, as Search generative AI. Its three states are include (the default), exclude, and inherit from a parent property.

Exclude removes a site's links and its grounding use from AI Overviews, AI Mode and Discover's AI features. That generally takes one to two days.

Google repeats that the control "isn't used as a ranking or inclusion signal affecting other parts of Search". It does not touch model training, which remains Google-Extended's job.

Nor does it override Merchant Center or Google Ads participation.

For site owners the practical change is simple. Since June this page has described an all-or-nothing choice.

The only way out of AI features was out of Search. That choice now has an alternative open to every property.

What has not changed is the measurement gap. AI Mode clicks still land inside the Web totals of the main performance report.

And the new report cannot show what an AI impression is worth.

This is a sighting rather than an announcement, and it is written as one. Google has published nothing and has not been asked on the record.

There is no measurement of how widely it renders.

On 28th Aug, the SERP Alert tracker run by Brodie Clark posted screenshots on X: "The Shopping ads carousel is now showing inline within the response for the first time, alongside an additional ad unit at the bottom."

The tracker's own reading was that this is "a big change to ads within Google's AI Mode", and the phrase to hold onto is "for the first time". Google has tested ads in US AI Mode since May 2025.

Until now they sat below the answer or attached to its end.

Two days later the same account reported a second format: "Now seeing the singular ad unit (easily confused for an organic result) alongside the carousel ad unit at the bottom."

That parenthetical is the part a marketer should read twice. A single sponsored result styled like an AI Mode citation is a different proposition from a labelled carousel.

That holds both for the advertiser buying it and for the publisher sitting next to it.

Search Engine Roundtable carried both observations on 31st Aug with the screenshots. It reached the same summary we would: Google is finding more places to put ads inside AI Mode.

What this does not establish is scope. One tracker on some queries is not a rollout.

At the time, Google had never described ads in AI Mode as anything other than a US test. On 28th Oct 2026 it expanded them to ten more countries.

The reason it belongs on this page anyway is where the ad now sits. An ad below an answer competes with the answer for attention.

An ad inside the answer is part of the thing the reader came to read.

Anyone tracking AI Mode ad placement has a dated observation to check against their own queries. That is the only way this gets settled until Google says something.

Google announced three travel upgrades to AI Mode in one post on 27th Aug. James Byers, Group Product Manager for Search, wrote it.

The first brings Google Flights price tracking into the conversation. Describe where and when you want to fly, and AI Mode shows options "with the latest prices from more than 300 partner airlines and travel sites".

Asking it to track those prices sets up an email alert.

Google says flight price tracking is live "in more than 180 countries and territories". The footnote carries the catch: it is available in all supported AI Mode locations and languages "excluding EEA countries and territories".

So the newest travel feature repeats a familiar pattern on this page. EEA users get the surface but not the feature, with no reason stated.

The second capability shows award costs in points or miles for flights and hotels, using rates "directly from our partners". This one is globally available, though the same footnote block says some depicted features will not reach EEA users.

The launch partners are Alaska Airlines and Hawaiian Airlines, American Airlines, Choice Hotels International, Hilton, and Wyndham Hotels & Resorts. Accor, Flying Blue, Hyatt, LATAM Airlines and Lufthansa Group are promised "in the coming weeks".

The third is the one that changes what a search result can complete: hotel booking inside AI Mode. Ask for a hotel and get a visual list with guest reviews.

Then choose "Continue on Google", which appears beside the integrated partners' own booking options. Pick a room type, review the cancellation policy and pay through Google Pay.

Google states the commercial boundary plainly: "the hotel or booking platform will act as the merchant of record and handle any customer service."

Hotel booking "has started rolling out in the U.S. in English" and reaches ten partners over the coming weeks. The first are Booking.com, Choice Hotels International, Expedia, Hilton and Hotels.com.

The others are IHG Hotels & Resorts, Marriott International, Priceline, Trip.com and Wyndham Hotels & Resorts.

This is the third agentic transaction surface in AI Mode. It follows restaurant reservations in thirteen markets and the event-ticket and appointment flows.

It is also the first where the stay is paid for without leaving Search.

For site owners the partner list is the point. A hotel chain or travel site outside it is not in the transaction path.

The "Continue on Google" option sits next to partner booking links rather than replacing them.

The post says nothing about how partners are chosen or whether any commission changes hands. Nor does it say when booking reaches markets beyond the US.

Google has brought link carousels for developing topics to AI Mode. AI Overviews has carried the format since July.

That is when the version Google announced in May went fully live there.

The announcement came from Robby Stein, Vice President of Product for Google Search. He posted it on X on the evening of 25th Aug, with no blog post and no help-page change.

His description of the trigger and the contents: "Ask about something that's trending and you'll see links to relevant articles, webpages (incl your Preferred Sources) front & center."

Stein framed the goal as making it "easy to connect with original coverage and a range of perspectives", and said the format was already in AI Overviews.

For publishers this is a rare change to an AI surface that adds link slots rather than removing them. That is why it belongs on this page even though it arrives as a single social post.

The mention of Preferred Sources connects it to the control Google shipped an embeddable button for five days earlier. A reader who has picked a publication can see it surfaced in these carousels.

What the post does not say is the part to hold onto.

It gives no countries, no languages and no percentage of queries that qualify as trending or developing. Google has published nothing else on it, so the scope of the rollout is unverified.

And there is no reporting surface that would separate a carousel click from any other AI Mode click.

Search Engine Roundtable wrote it up the next morning and dated the change to the day before. It noted the format was welcomed unevenly by publishers who would rather have the click than the citation.

Note that its write-up places the AI Overviews version in June. That is when the carousel was first spotted in testing rather than when it shipped.

Google published a back-to-school post on 19th Aug that reads as five study features. It lands as the largest set of AI Mode availability changes since France.

Awaneesh Verma, a Distinguished Product Manager on Search, wrote it. The availability sentences buried in each section are the part that matters here.

The headline for this hub is the first one. On interactive visuals, Google says: "Generative UI capabilities have launched globally in English in AI Mode and have started rolling out as well in AI Overviews."

Generative UI lets Search build a custom tool or simulation for the question rather than describe it. It arrived with Gemini 3 on launch day in November 2025 for AI Pro and Ultra subscribers in the US.

At I/O in May Google promised it "free for everyone" over the summer. That promise now looks kept in substance.

It is the third of the three I/O summer commitments to land.

The plan requirement is the part to read carefully, because Google's own two statements do not line up.

The AI Mode help page now describes the interactive-visual experience separately from the model tiers. It says the experience "is currently available to signed-in users in English", naming no subscription.

A few paragraphs above, the same page still lists generative UI among the things Gemini 3 Pro "unlocks".

And the page documents Gemini 3 Pro in AI Mode as English-only. It is for Google AI Plus, Pro and Ultra subscribers aged 18 and over.

So being signed in and using English is what both statements support. Google does not state in one place whether the richest version still sits behind a paid plan.

Google also tells users to ask for it explicitly ("a simulation, interactive visual, diagram, calculator, interactive model or something similar"), so it is not a thing AI Mode volunteers.

The second change is a product moving in rather than out. Notebooks from Gemini Notebook, which Google's post calls the new name for NotebookLM, now appear inside AI Mode.

That means a notebook per class or project, with class slides, syllabi and web articles as sources. Previous AI Mode threads can be added as sources too.

Existing notebooks from Gemini Notebook or the Gemini app "will appear automatically in Search, syncing the sources and instructions for each notebook across Google products". Google says notebooks in AI Mode are "rolling out over the coming days across more than 180 countries in English, with local language support to follow in the weeks ahead".

Three more, each with its own availability line.

Practice quizzes are "now globally available in English free of charge, in both AI Overviews and AI Mode", covering any subject plus the ACT, AP, ENEM, GRE, JEE, LSAT, MCAT, NEET and SAT.

The test-prep material comes from named partners rather than from the open web. They are The Princeton Review, Careers360, PhysicsWallah and Akira Enem.

Custom file creation turns uploaded files or an AI Mode thread into a document, slide deck, spreadsheet or text file. It "has rolled out globally in English and will come soon as well to AI Overviews".

And Lens in Search gains an interactive learning mode "over the coming weeks", globally in English.

A photo of your working returns an AI Overview that explains the step and points at the mistake. Follow-ups continue in AI Mode.

Read together, four of the five features put the finished artifact inside Search. Those are the simulation, the quiz, the file and the coached explanation.

The quizzes are the clearest example of a different sourcing model. The authority comes from a commercial licensing deal rather than from a cited page.

No publisher can be crawled into that slot. The availability table now carries all four as rows.

The strongest evidence yet about what AI Mode does to traffic comes from an experiment that made people use it.

Stephanie T. Wang of the University of Pennsylvania and Danaé Metaxa are two of the authors. The other three are Jeffrey Gleason, Yakov Bart and Christo Wilson of Northeastern.

They ran a preregistered field experiment and posted it to arXiv on 18th Aug.

The design is what makes it unusual. They recruited US Google users through Prolific and a Northeastern work-study program in mid-March 2026.

They screened for people who use Chrome and Google as their primaries, and installed a browser extension.

Everyone spent three days on unmodified Google. Then each was randomly assigned for seven days to one of three conditions.

In AI Mode Search, every query was redirected into AI Mode. In No AI Search, AI Overviews were hidden and AI Mode redirected away.

The third condition was unmodified Google as the control.

Of 1,444 people enrolled, 1,100 searched at least once and 956 finished the exit survey. During the baseline period participants averaged 9.2 searches a day.

36% of their searches produced an AI Overview, and they used AI Mode voluntarily for 0.6% of them.

Assignment to AI Mode cut click-through to the web by 18.8 percentage points (95% CI -22.2 to -15.3). It also cut daily search sessions by 0.92, which the authors had predicted would go the other way.

The drop was steepest among the heaviest Google users, at 2.01 fewer sessions a day.

Time inside a session rose by 0.43 minutes. The fraction of users clicking through to news sites fell 12.5 points.

The share clicking through to Reddit fell 21.2 points, and to Wikipedia 9.9 points.

Ad clicks fell 42.7 points, but that one is an artifact of the date rather than a finding. AI Mode carried no ads when the experiment ran.

The perception results are the half that has no precedent in the panel studies. Assignment to AI Mode reduced trust in information found on Google by 0.34 points on a seven-point scale.

It also reduced perceived usefulness, satisfaction, agency and relevance by 0.42 to 0.73 standard deviations, all at p<0.001.

It raised the fraction of users who searched on Bing, DuckDuckGo or Yahoo by 11.2 points. It also raised stated intent to switch to Bing by 1.25 points on a seven-point scale.

In free-text answers, negative impressions of AI Mode outnumbered positive ones, 33.6% against 29.3% of 309 responses. The most common specific complaint was loss of control, at 17.6%.

Difficulty reaching a particular website came next at 15.3%, then too few or too similar sources at 13.4%.

Three limits belong with the numbers, and all three are the authors' own. The AI Mode condition is forced full adoption, against a real-world baseline of 0.6% of searches.

So it measures what an AI-Mode-only Google would do rather than what today's opt-in AI Mode is doing.

The exposure was seven days, too short to see whether people adapt. And the sample skews young, highly educated and left-leaning.

The AI Overviews half of the same experiment has its own problem. Google changed the markup of AI Overviews mid-study and broke the extension that was hiding them.

That arm is covered on the AI Overviews hub, where the click effect belongs.

For anyone reading this hub to plan, the useful part is the direction of the tension. Google has said repeatedly that AI Mode is not the default search experience.

This is the first causal estimate of what happens if it becomes one. The answer is less traffic to the web, and users who like Google less and look at Bing more.

The report that carries AI Mode impressions under-counted them for five days in August. Google has confirmed it, and the fault is in the logging rather than in the surface.

From its data anomalies page: "A logging error caused a decrease in impressions on the Generative AI performance report in Search for data from August 13 - August 17, 2026. This issue affects data logging only."

That note first said the error started on 13th Aug and was "ongoing". Google has since closed the window at five days and dropped the word.

There was no announcement and no revision date on the page.

John Mueller added on Bluesky that Google is "working on resolving it", that it is "just a logging issue and not representative of visibility changes in Search", and that an annotation will appear in Search Console "in the near future".

The timing is awkward for this hub in particular. The Gemini 3.7 Flash switch of 14th Aug sits inside the window.

So anyone measuring what it did to their own AI Mode impressions is reading corrupted days.

Google's note covers exactly one report. Site owners separately reported a clicks decline in the ordinary Search performance report from around 12th Aug.

Search Engine Roundtable collected those reports, and Google has said nothing about that.

AI Mode queries do reach the ordinary report, under the counting rules.

Nothing has been said about a backfill. The four Discover logging errors this year were never backfilled, which is the only precedent to go on.

A dating note: Google's help pages carry no revision dates. So the closed window above is dated to our reading of the page on 24th Aug 2026.

Robby Stein, VP of Product for Google Search, announced it on X: "We're bringing Gemini 3.7 Flash to Search!"

He said the model is "better at following instructions + understanding your intent, so you get even more helpful responses", and that it is "rolling out today globally in AI Mode for Google AI Pro & Ultra subs in English". Subscribers select it with the "+" icon in the "Ask anything" bar.

Google had introduced Gemini 3.7 Flash the day before as its "most intelligent workhorse model yet for coding and agents". That announcement did not mention Search anywhere.

Barry Schwartz of Search Engine Land says he asked Google about it, and Stein posted the next day.

It is an option rather than a replacement. AI Mode still runs on Gemini 3.5 Flash by default, as it has since I/O in May.

Google has not said whether 3.7 Flash will take that slot. Nor has it said whether the new model feeds the automatic routing Google started building into AI Mode last November.

The gate is the usual one. A paid Google AI plan is required, so the free tier does not get the new model.

The rollout is English only for now, even though it is global.

Google's own help pages have not caught up. The AI Mode help page still describes only "Pro" and "Fast" as menu choices.

It does not mention 3.7 Flash at all, checked on 16th Aug.

Search Engine Journal reports that the live menu lists it under a "Gemini 3 models" heading beside "Auto" and "Pro". So the labels a reader sees and the labels Google documents are currently different.

For anyone tracking AI Mode citations, the practical use is comparison. Running the same query on more than one model now separates a model change from a ranking change.

That was guesswork while the surface offered a single answer.

Khushal Bherwani posted a screenshot of the Google home page carrying three new buttons: "Create images", "Ask about files" and "Brainstorm". Each one, he noted, takes you to AI Mode.

Barry Schwartz covered it at Search Engine Roundtable the same morning. The buttons sit below the big search box, where the Google Search and I'm Feeling Lucky buttons live.

They are additional to the AI Mode button, the plus sign, Lens and voice search that are already there.

Google confirmed it hours later. Robby Stein, VP of Product for Google Search, replied on X: "This is a small test on desktop to help people find new things they can do with Search."

He added that there is "No impact to how the core search box works", and that people should type their question and hit enter as usual.

None of the three features is new. All of them already existed inside the Google app and behind the plus icon in the search box.

What is new is where they are. They sit on the page a desktop searcher lands on before typing anything.

And they point at AI Mode rather than at the results page.

That is why it is on this page at all. Almost every AI Mode doorway Google has added this year has been inside something else.

That something was a browser, a phone app or a voice surface.

This one is on the home page itself. The three labels tell a reader what AI Mode is for before they have used it.

It changes nothing in the availability table. Google called it a test, named no countries, gave no timeline and has said nothing about a wider rollout.

So the platforms and default-experience rows are untouched.

Stein's second sentence is a reminder that the ordinary search box still behaves as it always has.

One detail separates this from the redesign it resembles. Schwartz notes at Search Engine Land that none of these buttons were in Google's I/O demo.

That demo in May showed an intelligent search box. So this is not that redesign arriving.

It is a separate experiment in the same direction.

Two months after the Generative AI performance reports arrived, the rules deserve a read. They decide what an AI Mode impression actually means.

They sit in Google's Search Console help pages rather than in any announcement.

The rule that matters most is about conversations. Google's help page says that "if a user asks a follow-up question within AI Mode, they are essentially performing a new query", and that "all impression, position, and click data in the new response are counted as coming from this new user query".

One AI Mode session of three or four turns can generate several queries' worth of impressions for the same page.

Google's own framing is that an AI Mode search runs about three times longer than a traditional one. That makes the multiplier a feature of the surface rather than an edge case.

The other two AI Mode rules are simpler. Clicking a link to an external page in AI Mode counts as a click.

Position, Google says, "follows the same methodology as a Google Search results page".

What the Generative AI report itself carries is narrower than that. It reports AI Overviews and AI Mode impressions with no queries, no clicks, no click-through rate and no average position.

It also excludes Search Labs experiments.

When this was written Google was still "rolling out this report to a subset of website owners". The report reached every property worldwide on 31st Aug 2026.

There is also an aggregation trap in it.

Chart data is aggregated by property, so in Google's example, "if two results from the same site appeared in a generative AI search results feature, they count as a single impression". The page table aggregates by URL instead.

Page-level impressions will therefore not always add up to the chart total. The difference is the dimension rather than an error.

Queries are the gap that has not moved. AI Mode queries are absent from the Generative AI report.

The only place they appear is the main performance report, inside the Web search type totals. That is exactly as they have since June 2025, and still with no AI Mode filter.

A site owner can see the conversational phrasing arriving. They cannot separate it out.

None of this closes the clicks gap. Impressions counted three times in one conversation, with no click figure beside them, measure how often Google used a page.

They do not measure what that use was worth.

Carolyn Shelby, writing for Search Engine Journal, put the same warning more bluntly. Do not pour generative and ordinary organic impressions into one bucket, and do not compute a blended click-through rate.

A note on the date: Google's help pages carry no revision date. So this entry is dated to the analysis that set the rules out, not to a help-page change.

Productrise tracks organic product visibility in Google Shopping. It ran the same shopping queries on standard search and on AI Mode.

It did so on the same day for 21 days in July 2026. Its study drew on over 2 million product listings across more than 100,000 result pages and AI Mode responses.

The headline gap of 95% is two effects multiplied together.

The first is whether a shopping search returns any products at all.

Standard search did on about 88% of the tracked queries. AI Mode did on about 23%.

The second is how many appear when they do appear. Standard search averaged 22.5 products per results page against 4.3 in AI Mode.

Productrise describes that as roughly 5.3 times fewer slots to compete for.

The more useful number is the overlap. Only about 0.8% of products seen in standard search also showed in AI Mode for the same query that day.

Productrise reads that as the two surfaces picking "largely different products, so a strong position on one does not transfer to the other".

That points the same way as a different measurement on this page. SISTRIX tracked citations rather than products.

It found AI Mode and AI Overviews citing different domains 83% of the time on the same prompt.

Visibility on one Google surface is turning out to say very little about the next.

The scope is narrow and the study says so. It covers the United States and the United Kingdom only, "where Google's Shopping Graph feeds directly into AI Mode answers".

Productrise left other markets out because they do not return rich product results in AI Mode yet.

Two more limits apply. The queries are the product-centric ones Productrise already tracks, not a cross-section of everything people search for.

And the study typed ordinary keyword queries rather than the longer conversational prompts AI Mode invites. The reasoning was that AI Mode opens from the normal search box.

It counts listings, not clicks. Nothing in it measures what a shopper did next.

The payoff its author suggests is stated as expectation. Hugo Huijer, who founded Productrise, says click-through "should be considerably higher for the products that do make the cut".

Callum Lockwood of Re:Signal, quoted in the same release, cautions the other way. Fewer products in one response "doesn't necessarily mean fewer opportunities for brands to be recommended over the course of an entire shopping journey".

Google's own version of this measurement was announced in May and went live in five countries by 8th Sep.

It shows a retailer's share of voice across AI Mode and AI Overviews. It reports share and product counts, not clicks.

Google launched AI Mode ("Mode IA") and AI Overviews ("Aperçus IA") in France on 22nd Jul. The post was by Sébastien Missoffe, vice president and managing director of Google France.

AI Mode arrived complete rather than staged. It came to desktop and mobile web plus the Google app on Android and iOS.

Search Live voice and camera input came too, along with query fan-out and the new intelligent search box. Gemini 3.5 was credited for the reasoning and multimodal understanding.

Google's French post repeats the framing this page tracks. An AI Mode search runs about three times longer than a traditional one.

It also repeats the same division of labor: an overview for the gist and AI Mode for the follow-ups.

France was the last major market without AI Mode. It was excluded from the October 2025 European wave over the press neighboring-rights standoff.

The launch lands two months inside the 23rd Sep date Google committed to in June.

The publisher terms that made it possible are the part to watch. Google has neighboring-rights contracts with about 450 French publishers and press agencies.

It told them AI answers would count toward those payments.

Its launch post also points site owners at the Search Console opt-out from generative AI features. It notes that the control is available in France.

And it repeats that the control is not a ranking signal outside those features.

No collective agreement has been announced and negotiations continue. The fuller publisher picture is in our AI Overviews coverage of the launch.

Google announced three new Gemini models on 21st Jul: 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber. One line in the post concerns Search: "3.5 Flash-Lite is also rolling out in Google Search."

Google describes Flash-Lite as its fastest and most cost-effective 3.5-class model, at roughly 350 output tokens per second. It is built for low-latency, high-throughput agentic tasks.

What that means for Search is not spelled out, and this is where the speculation starts. The post gives no detail on what Flash-Lite does inside Search and never mentions AI Overviews or AI Mode.

What we can state holds steady. AI Mode's answers run on Gemini 3.5 Flash, which became the global default at I/O 2026.

Google has not said that changed.

Search Engine Land reports Google tied Flash-Lite to "agentic search" and speculates it "might be used for Google AI Overviews and Google AI Mode as well". It presents that as a possibility rather than a confirmed change.

We are logging the model reaching Search and will update this entry if Google confirms where Flash-Lite is actually used.

Google started letting people connect apps to AI Mode, moving it from answering questions toward completing tasks.

Connect Instacart and AI Mode can add ingredients straight into your Instacart shopping cart. Connect Canva and you can ask it to show template options for something like a flyer.

Connect YouTube Music and it can curate a playlist and save it to the app.

Google framed it as "starting to roll out this week in the U.S." and said "we're working with a range of partners and look forward to launching with more apps soon".

It is Google's answer to the app connections ChatGPT and Claude already offer, per TechCrunch. It now sits inside the search box more than 9 in 10 searches run through.

Google pairs it with Personal Intelligence, the opt-in feature that connects Gmail and Photos. With both switched on Search can give "even more tailored responses".

Google did not say whether connecting apps needs a subscription, and it is US-only for now. So there is no new SEO surface to act on yet.

We track it here as AI Mode keeps absorbing tasks that were never search queries.

Google Search VP of Product Robby Stein announced on X that Personal Intelligence in Search "now connects to Google Calendar", per Search Engine Journal. AI Mode can "add invites or other meetings to your Calendar directly", and its answers can take account of plans that are already in there.

That makes Calendar different from the two apps Personal Intelligence already used. Gmail and Google Photos hand AI Mode information to read.

Calendar is the first connection it can also write to.

Ask for dinner options and the answer can work around a meeting you have already booked. Mention an invitation and AI Mode can create the entry without you opening Calendar, as Digital Trends describes it.

Stein said it is available in the US with more countries to come. Google has given no date for them.

Google previewed a Calendar connection at I/O in May without a timeline. That was the same announcement that took Personal Intelligence to nearly 200 countries and 98 languages with no subscription.

Google's AI Mode help page now lists Gmail and Google Calendar together as the Workspace apps Personal Intelligence connects to.

The part to watch is what personalization does to measurement. Both outlets point to iPullRank testing.

It found connecting Gmail changed which brands appeared in AI Mode answers to identical prompts.

Calendar adds timing to that mix. Two people can ask the same question and get different answers because their evenings differ.

There is still no way to see any of it from the outside. Search Console's Generative AI reports carry impressions only, with no queries and no clicks.

Common Sense Media's Youth AI Safety Institute rated Google Search's AI features "Unacceptable Risk", its lowest tier, for children.

The test ran 2,624 interactions in seven test plans. AI Mode and AI Overviews failed seven of the Institute's eight AI principles and all five severe-harm "Red Lines".

The assessment's headline: "Google Search's AI answers are not safe enough to be kids' default answer machine."

AI Mode, the conversational tab, did better than the automatic AI Overview on most measures. It routed 77% of crisis prompts to an appropriate resource against AI Overviews' 58%.

But it still failed all five severe-harm Red Lines, and it completed 100% of the homework assignments testers gave it.

As PBS reported, neither feature can be switched off by parents or schools, unlike the standalone Gemini app. Google said its AI features have "strong quality and safety guardrails" and called the testing flawed for using "a narrow set of ambiguous and contrived queries".

Google Search VP of Product Robby Stein announced the change on X on 30th Jun. AI Mode now shows "prominent links at the top of responses with useful details and images" for relevant recipe queries.

Those include the creator name, recipe ratings and the number of ingredients, per Search Engine Journal.

"We've heard feedback on recipe results in AI Mode, and we're making updates to better connect people with recipe creators on the web," Stein wrote.

It builds on the recipe panel Google added in March. That let people tap a dish to open links to recipe sites.

Creators are not sold yet. Android Authority's coverage found food writers still wary of how AI Mode handles their work.

One describes full AI recipes as ranging "from verbatim plagiarism to wildly inaccurate representations". More prominent links only help if people click them.

We will watch whether the change registers as impressions in Search Console's Generative AI reports.

Google told French publishers that AI Mode and AI Overviews will launch in France by 23rd Sep 2026. The letter was reported via Ouest-France.

France is the last major market without either feature. It was excluded from the October 2025 European rollout over the press neighboring-rights standoff.

The terms are the story: Google committed to a publisher opt-out, transparency about AI-driven traffic, and compensation under neighboring rights. That goes further than what publishers get anywhere else except the UK, where the CMA ordered an opt-out in June.

It sets a precedent every other press market can point at.

Google began rolling out information agents in AI Mode on 12th Jun. They are the first of the Search agents announced at I/O 2026 to reach users.

Ask AI Mode to "keep me updated on" or "alert me when" something and the agent keeps working after you close the tab. It watches blogs, news sites, social posts and real-time finance, shopping and sports data.

Then it sends an update with links when something new appears, per 9to5Google's rollout coverage.

Availability is the caveat. Information agents are exclusive to Google AI Ultra subscribers at launch, across every AI Mode language and market.

AI Pro access is to follow later in the summer.

Update: on 28th Sep Google began opening the feature to all users.

For SEO this is a new surface to be cited on without being searched for. The agent decides which pages are worth sending.

We will watch whether agent-driven impressions show up in Search Console's Generative AI reports.

A Chrome Canary experiment that routed address-bar searches into AI Mode set off a wave of "AI Mode is becoming the default" reports. Google Search VP Rajan Patel shut the speculation down on the record: "This is an error, and we do not plan to make AI mode the default behavior for Chrome search."

That matches Google's line at I/O 2026. AI Mode is a tab and an entry point, not the default search experience.

That holds even as AI Overviews (which do show by default) hand off into it. We track this distinction in the availability table.

The reason is that "is AI Mode the default now?" is the most common misreading of Google's announcements.

Google announced Search Generative AI performance reports for Search Console: dedicated reporting for AI Overviews, AI Mode and Discover AI. The catch: the reports cover impressions, not clicks.

So publishers can finally see how often they appear in AI surfaces. But they still cannot measure what that appearance does to their traffic.

On scope Google said only that the reports go to "a subset of website owners". It named no country in the post or in the help pages.

Trade coverage reported the first properties to get them were UK ones. Search Engine Journal was still describing it three weeks later as limited testing with a select group of UK sites.

Sightings elsewhere followed quickly. So treat the UK framing as where it was first observed rather than as a stated rollout order.

The same release added a control to block content from Google's AI features. It is the first explicit AI opt-out in the product.

Until now AI Mode traffic has been invisible as such. Since June 2025 it has been lumped into the "Web" search type with no filter, as the SEO impact section explains.

The rollout reached every property worldwide on 31st Aug 2026.

A week before Search Console's generative AI reports, Google announced the retailer equivalent: AI performance insights in Merchant Center.

It covers shopping journeys that start in AI Mode, AI Overviews or the Gemini app.

Four components. Share of voice benchmarks a brand "against brands similar to yours on shopping journeys starting from AI Mode, AI Overviews, or the Gemini app".

Shopping funnel performance splits that across discovery, evaluation and purchase. Product term insights show popular product terms and a brand's share of them.

Product attributes insights score attribute completeness so a merchant can see what its feed is missing.

For anyone tracking what AI Mode does to their business, this is the second Google-owned measurement surface in a fortnight. It is the first aimed at retailers rather than site owners.

It also has the same hole. Search Console reports impressions, Merchant Center reports share, and neither reports clicks from an AI answer.

Treat it as announced rather than shipped. Google said it "will be rolled out to the U.S., Canada, Australia, India, and New Zealand in the coming months", with no date and no word on general availability.

It has published no methodology for how share of voice is calculated.

PPC Land covered the announcement the same day.

Update, 10th Sep 2026: the report is now live in all five countries, with new search-term and intent sections.

At I/O 2026 Sundar Pichai said AI Mode had "surpassed 1 billion monthly active users" within a year, with queries more than doubling every quarter.

Gemini 3.5 Flash became the default model globally. AI Mode and AI Overviews merged into a single flow worldwide.

Google also shipped what it called the biggest upgrade to the search box in more than 25 years.

The roadmap: Search agents for Pro and Ultra subscribers in the summer, and generative UI free for everyone "this summer". The roadmap also had expanded agentic business-calling, and Personal Intelligence reaching about 200 countries and 98 languages without a subscription.

Google was explicit that AI Mode is still not the default search experience.

This is the clearest causal evidence yet that AI Mode and AI Overviews do different things to a content platform. And they point opposite ways.

It comes from a working paper by Peibo Zhang, Ruomeng Cui and Dennis J. Zhang. It was posted to arXiv and is not yet peer reviewed.

The design uses Google's own content policy as a natural experiment. Safe-for-work subreddits can be cited in both AI Overviews and AI Mode.

Not-safe-for-work subreddits are indexed in organic search but barred from both. That gives a treatment group and a control group inside one platform.

The AI Overviews baseline is covered in full on the AI Overviews hub. Daily comments rose 12.0% and daily commenting users 12.4% against the control after the August 2024 rollout.

The gain was concentrated in advice and opinion communities rather than factual ones.

The explanation the authors give for that gain is what makes the AI Mode result follow. A static summary cannot reproduce a discussion thread.

So for opinion and advice it worked as a discovery route into one rather than a replacement for it.

Then the interface changed. The authors extended the panel to December 2025, with the 21st Aug 2025 global rollout as a second treatment date.

The premium on experience-based communities falls 59% for commenting users and turns slightly negative for comments.

The reading is that a conversational interface delivers the depth a static summary could not. So the reason to click through into the discussion goes away.

Where AI Overviews complemented the platform, AI Mode substitutes for it.

The authors test the obvious alternative, that users simply grew to trust AI answers over time. They argue against it three ways.

A trust story predicts the decline should land on factual content, where the summary already answers the question. It lands on opinion and advice instead.

They also find a discrete break at the AI Mode launch rather than a slow drift. And their day fixed effects absorb any trend in AI trust shared across communities.

Two limits belong with this. The authors flag the short post-AI-Mode window themselves, so it measures attenuation in progress rather than a settled end state.

They cannot say whether it stabilises or runs on towards full substitution.

And it counts comments and commenters inside Reddit, not clicks, referral sessions or publisher traffic. It is not a measurement of what AI Mode does to your site.

That is why it carries no row in any statistics table here.

SISTRIX tracked 82,619 prompts once a week from 17th Dec 2025 to 8th Apr 2026. It covered AI Mode, AI Overviews and ChatGPT Search in six countries.

AI Mode churns hard: each answer cites 14 to 16 domains and 56% of them are new every week.

The churn is not evenly spread. In 86.5% of prompts there is a stable core of one to five domains that persist for months.

Meanwhile 89% of the domains outside that core rotate weekly.

So the useful question for a site owner is not whether it appears in AI Mode. It is whether the site sits in the core or in the rotating remainder.

The pattern barely varies by market. Weekly churn ran between 54% and 59% in Germany, the US, the UK, Italy, Spain and France.

Answers cited 11.8 to 16.1 domains, and the churn showed no sign of settling over the 17 weeks.

YouTube and Amazon turn up in the core often, which is no surprise. More usefully, in 83% of prompts at least one specialist domain sits in the core alongside them.

So the core is not closed to smaller sites.

The contrast with AI Overviews is the finding that matters most for anyone treating the two surfaces as one thing. AI Overviews are frozen for just over half of prompts.

On the same prompt the two products cite different domains 83% of the time, a Jaccard overlap of 0.17.

ChatGPT Search churns hardest at 74% a week, and from a much shorter list. It cites 3 to 4 domains per answer against AI Mode's 14 to 16.

The window closed six weeks before I/O 2026 merged AI Overviews and AI Mode. So it measures them as the separate surfaces they then were.

Content type predicted durability better than anything else measured. News articles held a place in a stable core only 1.4% of the time, while video was highest at 24%.

This is SISTRIX's own AI research index, so it is vendor panel data rather than ground truth. The study calls its own figures conservative: measured per URL instead of per domain, weekly drift runs about 15% higher.

PPC Land covered it the day it was published.

Robby Stein, VP of Product for Search, and Mike Torres, VP of Product for Chrome, announced that "when you're using AI Mode on Chrome desktop, clicking a link opens the webpage side-by-side with AI Mode".

Google's stated purpose is to make it easier to "visit relevant websites, compare details and ask follow-up questions while still maintaining the context of your search".

The same post added context from tabs. A plus menu sits on the New Tab page and in AI Mode.

It pulls recent tabs, images and files into a query. Files include PDFs, and it works on Chrome desktop and mobile, several items at once.

Canvas and image creation sit in that menu too.

Everything announced that day was live in the US only, with Google saying it would "expand soon to more places around the world". TechCrunch reported the same scope.

What makes it matter here is what a click becomes. A visit that no longer costs the reader their conversation is a visit they have less reason to abandon.

That is the mechanism behind Google's standing claim that its AI features spread traffic across more sites.

The other reading is that the page now renders inside a layout Google controls. Google has published nothing about what a side-by-side visit looks like from the publisher's side.

Search Console still reports impressions rather than clicks for AI surfaces. So a site owner has no way to isolate it.

Google announced that the Google app for desktop is "now available for Windows users globally in English". The post was by Vinay Mahagaokar, a group product manager on Search.

"With AI Mode built right in, you can ask whatever's on your mind and get helpful AI-powered responses with links to the web."

The app had been a Search Labs experiment since September 2025, US only and capacity limited. 9to5Google reported it is now on Windows 10 and later "around the world after previously requiring Google Labs sign-up".

Pressing Alt and Space opens a floating search box over whatever is on screen. It searches the web, Google Drive, installed apps and files on the computer.

It takes a shared window or whole screen for contextual questions, and uses Lens to select things on screen.

Google's help page puts the requirements at Windows 10 or later, English and age 13 or over, and adds that "AI Mode is not yet supported for all accounts, countries, and languages". The app's global reach is therefore not the same thing as AI Mode's.

What puts it on this page is the surface. Every other way into AI Mode is a browser tab, the Google app on a phone, or something inside Chrome.

This one is an operating-system-level shortcut that opens into AI Mode without a browser at all.

Google expanded agentic restaurant booking to eight new markets. They are Australia, Canada, Hong Kong, India, New Zealand, Singapore, South Africa and the United Kingdom.

Describe the table you want ("a table for two at a dog-friendly Italian restaurant in Shoreditch for Saturday at 7 p.m.") and AI Mode checks availability across reservation platforms. It returns bookable options with direct links.

UK partners include OpenTable, TheFork, SevenRooms, ResDiary, Mozrest, Foodhub, Dojo and DesignMyNight.

Rose Yao, VP of Product Management, announced the expansion on X as taking agentic restaurant booking in Search global.

The gates came off with the geography: no Labs opt-in or subscription is required in the new markets. The feature had launched in August 2025 for US Google AI Ultra subscribers only.

It reached all opted-in US Labs users that November. So this was the step that turned a premium US experiment into a plain AI Mode capability.

Search Live expanded globally to all 200+ AI Mode countries and languages, upgraded to the Gemini 3.1 Flash Live model.

Search Live is AI Mode's real-time layer: a spoken conversation with Search that can also watch through your camera. It launched in the US in mid-2025 and had been limited to the US and India before this expansion.

Canvas reached general availability for everyone in the US. Canvas was introduced through Labs in July 2025.

It opens a side panel where AI Mode drafts and iterates on documents, code and simple apps across a session. It is powered by Gemini 3.

It is the clearest example of AI Mode absorbing assistant-style work that would never have been a classic search query.

Google Search SVP Nick Fox announced 53 new AI Mode languages, the largest single language addition since launch. It takes AI Mode to just under 100 languages and covers roughly a billion more people.

The path to here: English only at launch (March 2025), then five languages including Hindi, Japanese and Korean (September 2025). Spanish followed (September 2025), then 35+ more in the big October 2025 wave, and now 53 at once.

Alphabet's Q4 2025 earnings gave the deepest usage picture yet. Daily AI Mode queries per US user have doubled since launch.

AI Mode queries are about three times longer than traditional searches. Roughly 1 in 6 arrive as voice or image rather than text.

Monetization remains in "early stages", with a Direct Offers pilot testing merchant offers inside answers.

Sundar Pichai also said directly that AI Mode is not cannibalizing core Search. Publishers have no independent data to test that, which is the recurring theme of AI Mode's SEO story.

Alongside making Gemini 3 the global default for AI Overviews, Google shipped a mobile handoff. Continuing from an AI Overview opens an AI Mode conversation that already carries the query and answer context.

Google's framing: "one fluid experience... a quick snapshot when you need it, and deeper conversation when you want it."

This was the moment the two products stopped being separate features. The full merger followed at I/O 2026 in May.

Google launched Personal Intelligence for AI Mode: an opt-in Gmail and Google Photos connection that personalizes answers with your context. It started with US English AI Pro and Ultra subscribers.

By I/O 2026 Google said it would reach about 200 countries and 98 languages without requiring a subscription.

Google made Gemini 3 Flash the default AI Mode model globally, pitching it as "frontier intelligence built for speed" with faster reasoning on complex questions.

Gemini 3 had reached AI Mode as a subscriber option a month earlier. Now every AI Mode user worldwide was getting a Gemini 3 model by default.

The same update opened Gemini 3 Pro to everyone in the US through the "Thinking with 3 Pro" picker. It also expanded Nano Banana Pro image generation to more US users.

AI Pro and Ultra subscribers kept higher usage limits. The default changed again at I/O 2026, when Gemini 3.5 Flash took over globally.

Ahrefs captured an AI Mode response and an AI Overview for the same query and compared the pair. It did so across 540,000 query pairs for citations and 730,000 for content.

The data was US data from September 2025, from its Brand Radar index.

The two surfaces cited the same URL 13.7% of the time. That rose only to 16.3% across the top three citations in each response.

From AI Mode's side the differences are consistent. It cites sources in 97% of responses, where AI Overviews return none at all in 11% of theirs.

It names 2.5 times as many people and brand entities, 3.3 per response against 1.3. And it leans harder on reference material.

Wikipedia supplies 28.9% of its citations against 18.1% of AI Overviews', and Quora appears 3.5 times as often.

The answers agreed even where the sources did not. Response pairs scored 86% semantic similarity on average, with 89.7% above 0.8 on a scale where 1.0 means identical meaning.

Despina Gavoyannis and Xibeijia Guan put it plainly: "9 out of 10 times, AI Mode and AI Overview agreed on what to say".

One number needs keeping straight. The study also reports a Jaccard figure of 0.16, and that one measures words rather than sources.

Only 16% of the unique words overlapped between the two responses. They opened with the same first sentence 2.51% of the time.

Each surface researches and writes its own answer.

For a site owner the practical line is that being cited does not carry across. Being cited by one surface says little about the other.

A brand mentioned in an AI Overview has a 61% chance of appearing in AI Mode's longer response, Ahrefs found.

The study states its own limit. It compares single generations.

Ahrefs' earlier work found 45% of AI Overview citations changing between generations of the same answer. So the pool each surface draws on may overlap more than one snapshot shows.

The panel is Ahrefs' own index rather than ground truth.

The data is from September 2025, eight months before Google merged the two surfaces at I/O 2026. So it describes them as separate products.

SISTRIX measured the same divergence on a different panel in early 2026. It found the two citing different domains 83% of the time, and AI Mode swapping 56% of its domains weekly.

So the split held across two panels, two methods and five months.

Google expanded Gemini 3 Pro in AI Mode to nearly 120 countries and territories in English. It was for Google AI Pro and Ultra subscribers, two weeks after its US debut.

Nano Banana Pro expanded to more countries alongside it. That is the Gemini 3 Pro image model that can turn answers into visuals and infographics.

For the first time, a new Gemini model reached Search on its launch day. Gemini 3 arrived in AI Mode the day it was announced, for US Google AI Pro and Ultra subscribers.

It came through a new "Thinking" option in the model picker, with a broader US rollout to follow.

Gemini 3 also brought generative UI to AI Mode: dynamic visual layouts, interactive tools and simulations built for the question. Examples are a physics simulation or a custom loan calculator.

Complex questions route to the frontier model while faster models handle simpler ones. And an upgraded query technique made the web searches AI Mode runs behind an answer smarter.