Google Discover: Updates Tracker, Traffic Studies and Feed Changes

Google announced three changes to Search profiles on 16th Sep, again under Ibrahim Badr's byline. Profiles stay a Discover story because following one shapes the follower's feed.

The eligibility bar fell for the second time since launch. Badr wrote: "We've lowered the eligibility threshold to 10,000 followers across YouTube, Instagram, X, or TikTok."

Google's help page now lists 10,000 for each of the four platforms. At least one linked account has to clear it.

That ends TikTok's separate, higher bar. The thresholds stood at 100,000 and 300,000 at the June launch.

August cut them to 35,000 and 100,000. Now every platform is 10,000.

Multi-account support is the second change. "You can now claim and manage Search profiles for all of your sub-brands from a single login," the post says.

The help page puts a number on it. One Google account can hold up to ten profiles, one per eligible brand or identity.

The third change is cosmetic. Article thumbnails are "optimized" and headlines run longer in the profile's article view.

One item the post leaves out is a Search profile badge for publishers' own sites. Google's Search Central guide, dated 16th Sep in the documentation changelog, says: "After you've claimed your Search profile, you can encourage your audience to find your content from across the web by adding a Search profile badge to your website."

The badge links to profile.google.com/@handle, as an image or a plain text link. Google says following a profile makes its linked content "more likely to appear for your audience on Google Discover".

That badge sits beside the Preferred Sources button from August. Google's brand guidance says to use the labelled Search profile button when both appear together.

Geography has still not moved. The help page still says profiles are "currently only available in the United States", and the post says "More U.S. publishers and creators are now eligible".

The report's help page opens with a new dated note: "As of August 31, 2026, we've rolled out these insights to all websites worldwide."

The Search generative AI control carries the same date for its own rollout. Its scope includes "Generative AI features in Google Discover".

There was no announcement. Google added the note to three help pages and to its June launch post.

A 21st Aug archive snapshot of the control page still described a rollout to "a subset of website owners".

For this hub the report matters because so little else measures Discover's AI features.

The June anomaly note was Google's first written confirmation that the Discover report existed at all. Until now most properties could not open it.

What it counts is narrow and Google says so. An impression is a link to your site seen inside a generative AI feature in Discover.

It counts only once it scrolls into view, and once per result per session. It is attributed to the canonical page rather than the page a reader lands on.

It reports impressions only, by page, country and date. There are no clicks.

A property without enough impressions in Discover's AI features still sees nothing. The help page keeps that line under the new note.

The control is the other half. Excluding a site removes its links, and its use in grounding, from generative AI features in Discover.

The same exclusion covers AI Overviews and AI Mode. So a publisher cannot opt out of one surface and stay in the others.

The fuller record is on the AI Mode hub. It covers what the reports carry and how the control works.

Google announced on 20th Aug that Discover is getting a typed personalization control.

Mrinalini Loew, General Manager of the Google Search Ecosystem, wrote that "in the coming days in the Google app, you'll be able to tap any three-dot menu in your feed to tell Google the exact topics or links you want to see more (or less) of, in your words".

And: "your feed will adjust on the fly, and remember your requests".

Google's framing is that tuning the feed should become "as easy to tune your Discover feed as it is to text a friend".

The examples Google chose are the useful part. They show how fine the control is meant to be.

Both are a topic with an exclusion attached: "kitchen reno ideas (eco-friendly only)" and "videos about weekend road trips (but no camping)".

No fixed menu of interests can express a request like that. That is the difference from the feedback controls the menu already carries.

For anyone who depends on Discover traffic, this adds a reader-side variable. The feed was already the least predictable surface Google runs.

Google's interest matching decides what Discover serves, not a query. So a publisher never sees why a piece was or was not served.

A persistent typed instruction now sits on top of that, invisible from the publisher's side. Nothing in Search Console reports it.

Say a segment of readers excludes a subtopic and a story stops being served. That looks identical to a story that simply lost interest matching.

The follow and profile mechanics remain the one lever pointing the other way. There, a reader's action makes a source more likely to appear rather than less.

Two limits on what has actually shipped. This is announced for "the coming days" rather than live.

Google names only the Google app. That means the mobile feed rather than every Discover surface.

The same post carries the new embeddable Preferred Sources button. That one affects Top Stories, AI Overviews and AI Mode rather than Discover.

It also announces a topic-picking control for audio briefings in the Google News app on Android.

Google's Manage your Search profile help page gained an "Edit your handle" procedure. Search Engine Roundtable reported on 18th Aug that it had appeared in the help pages overnight.

The steps are short: click Edit profile, then click Edit next to your @handle, or "If you don't have a handle, click Edit handle", then "Select the handle you want to use".

The constraint is the part that decides whether this is useful to you.

Google's new FAQ answers it directly: "Your Search profile handle reflects a handle from your linked content platforms. To use a specific handle, you must first link a social account or content platform that uses that handle."

The procedure adds the same point as a tip: "If the handle you want isn't listed, you can link a new social account."

So this is a picker over handles you already control, not a naming field.

That matters because of how the default is assigned. Google's profile creation page says a profile's handle "is set to match your most followed linked content platform account handle", and that "if that handle is taken, the next most followed platform handle is assigned".

Until now, some publishers had their largest following on a platform with an off-brand username. They were stuck with it in Search and Discover.

Three other FAQs arrived alongside the handle section. The practical one is about waiting.

Google says it "manually verifies some changes before they appear on your profile, which may take some time. You'll get an email once the change is approved or denied", and that it reviews suggested changes to check they are "relevant to the brand or identity your profile represents".

Adding a content platform already required approval. So the route to a specific handle runs through that review too.

On dating this: Google's help pages carry no revision date. So the entry is dated to the report that identified the change.

The 14th Jul archive of the same page runs straight from pinning content to the Insights section. It has no handle procedure and none of the four new FAQs.

That puts the change between that snapshot and 18th Aug.

Availability has not moved: the creation page still says profiles are "currently only available in the United States".

Six Google authors posted arXiv 2608.15780 on 16th Aug 2026. It is accepted for the CIKM 2026 applied research track.

We read it in full after Search Engine Journal reported it on 21st Oct.

It presents SDF, "a staleness filtering system fully deployed in Google Discover, a personalized recommendation feed with hundreds of millions of daily and billions of monthly active users".

Discover is described as "a personalized feed of open-web content in the Google Search app".

The filter runs before ranking. "It prunes fundamentally stale content before it enters the computation-heavy ranking stage", and removed items "are never served".

Two filters are combined. A supersession filter judges whether a newer item has replaced an older one.

It uses a T5 model trained on LLM labels.

A decay filter predicts how an item's traffic falls over horizons from 12 hours to 14 days. Its labels come from search clicks: "We use search-click logs rather than raw page-view logs", because page views are "biased by publisher promotion".

An item's arrival time is its indexing time, "closely approximating its publication time".

The headline figure covers two years of deployment. "User-filed staleness reports (in-product user feedback) declined by 54.9% relative to the pre-deployment baseline", with "supersession by 64.0%, decay by 34.4%".

In a first-locale experiment "tuned to be engagement-neutral", human raters judged the items SDF promoted and demoted. The stale rate among promoted items was 6.91 points below the rate among demoted ones.

A two-month holdback with both filters off gives the user-side effect.

With the filters on, dismissals fell 0.16% and feed diversity rose 0.13%. Positive engagement rose 0.26%, each with a 95% confidence interval.

Serving cost fell too.

Over Q1 2026 the supersession filter saved 4.76% of serving CPU and TPU hours. The decay filter saved a further 3.98%.

The paper gives no start date for the deployment and names no locale or language. It gives no count of items removed and nothing on publisher traffic.

It says nothing about Google Search ranking, AI Overviews or AI Mode.

It gives publishers no lever and does not say whether a page's date markup feeds either filter. The authors note the decay model "misses a long tail of gradually-decaying items".

For publishers the mechanism is the point, not the figures. An article judged superseded or decayed is dropped before Discover's ranker sees it.

Search Engine Journal's report adds advice on freshness and interlinking. That advice is not in the paper.

Google's data anomalies page logs a fourth Discover data loss of 2026: "A logging error caused a decrease in clicks and impressions on the Discover performance report for data on August 13, 2026."

As with the 7th-8th May, 21st May and 24th Jun errors, the note says the issue "affects data logging only". No backfill is mentioned.

The note repeats the Generative-AI-in-Discover line that first appeared in June: "For properties with access to Generative AI features in Discover, this error also caused a decrease in reported impressions."

The date is the part to hold onto. A separate error began the same day on the Search side.

It cut impressions in the Generative AI performance report in Search, and that one ran until the 17th.

So for any publisher measuring both surfaces, 13th Aug is a bad day in two reports at once. The Search-side gap runs five days wide.

Google has not said whether the two are related. The anomaly page lists them as separate notes.

A dating note: Google's help pages carry no revision dates. So this is dated to the data it describes rather than to when Google wrote the note.

We read the page on 24th Aug 2026.

The bar for claiming a Search profile has fallen by roughly two thirds. That brings the feature within reach of mid-sized publishers and creators for the first time.

Robby Stein, VP of Product for Google Search, posted on 12th Aug: "We're lowering the follower thresholds so more creators and publishers can set them up", adding that Google will "also expand them to more countries soon".

The numbers sit on Google's own help page. It now asks for 35,000 followers or subscribers on YouTube, Instagram or X, or 100,000 on TikTok.

At least one linked platform has to clear the bar. An archived copy from 8th Aug shows the bar it replaced: 100,000 on the first three, 300,000 on TikTok.

Our launch entry carried those old figures, which had stood since June. Nothing else in the eligibility list moved.

You still need to be 18 or over, with content meeting Google's policies. Google may already have generated a profile for you if you have a knowledge panel.

This belongs on a Discover page rather than a Search one because of what a profile does. Following a publisher from its profile makes that publisher's content likelier to appear in the follower's feed.

Creating a profile, Google says, "doesn't directly affect your content's ranking".

Geography has not changed yet. The help page still says profiles are "currently only available in the United States", so "more countries soon" carries no date and no market list.

The 1492.vision researchers surfaced the 54-publisher pilot in May. They posted the new thresholds the same evening Stein did, ahead of any trade coverage.

Google introduced platform properties in a Search Central post on 7th Jul. They are a Search Console property type for accounts rather than websites.

Google described them as a way for creators to understand "how their social and video posts perform on Google Search and Discover". It said the property type would arrive gradually.

On 29th Jul Google made them globally available and published an analysis guide alongside. That post names the surfaces precisely: Instagram, TikTok, X and YouTube posts as they "perform on Google Search, Discover, and Google News".

The Discover specifics sit in the help pages rather than in either announcement. Platform properties show "reports for your content performance on Google Search, as well as News and Discover if your content appears there", and Google adds that "Discover and Google News reports only appear if your content receives traffic from those surfaces".

Why this lands on the Discover beat: social posts formally entered the feed in September 2025. Since then, the measured result has been publishers losing feed space to X and YouTube.

Until now the accounts filling that space had no way to see their own Discover numbers. Neither did anybody else.

Read the scope carefully before assuming parity with a website property. Discover carries no queries, so the query-level reporting Google describes is a Search feature.

The Insights report is documented in terms of "Google Search" alone.

The new analysis guide offers no Discover-specific guidance at all.

One thread back into this hub's own timeline. Have you already set up a Search profile?

Then Google says you should have access to that same platform's data in Search Console.

Press Gazette reported that AI summaries grouping publishers beneath one AI overview have reached the UK Discover feed. Deployment is described as sporadic so far.

The example that made the point came from Reach's SEO and Discover director Nicola Agius. Fourteen publishers covering one Jennifer Lopez story were bundled under a single AI card.

The same reporting sketched how the UK feed's make-up has changed. Social media posts now make up almost a quarter of the feed.

Roughly 64% more publishers were appearing in Discover than a year earlier. That spreads the same feed across more sources.

The US version of this story is a year further along. The measured outcome there is a majority-AI feed whose summary cards default to YouTube.

That is the reference for what UK publishers should expect if deployment widens.

Google's Data Anomalies page logged a third Discover data loss inside seven weeks. Clicks and impressions for 24th Jun were under-reported, following the 7th-8th May and 21st May errors.

Each note says the issue "affects data logging only". None of the data is backfilled.

Many publishers' dominant Google channel reports through this one surface. For them, documented gaps in the only impression record matter on their own.

PPC Land's coverage of the May error walks through what that means in practice.

The June entry carried something bigger in passing: "For properties with access to Generative AI features in Discover, this error also caused a decrease in reported impressions."

That is Google's first written confirmation that Search Console reports Generative AI impressions inside Discover, for some properties. It is a sibling of the Generative AI reports that reached Search in June.

A dating note: Google's help pages carry no revision dates. So these facts are dated to our reading of the page rather than to when Google wrote them.

Google launched Search profiles in the US, announced by Ibrahim Badr, Product Manager for Search: "We're launching Search profiles, a new way for publishers and creators to shape their presence on Search."

The Discover mechanics are the reason it is on this page. Profiles are reachable from knowledge panels, a direct URL and by tapping a publisher's name in Discover.

Following from a profile makes that source's content more likely to appear in the follower's feed.

The gates and furnishings at launch, per PPC Land:

• Follower thresholds of 100,000 on YouTube, Instagram or X, or 300,000 on TikTok, cut to 35,000 and 100,000 in August, then to 10,000 on every platform in September.
• Up to eight pinned featured posts or videos from the past year, plus eight web links.
• An Insights beta powered by Search Console data.

The launch also explained a spring mystery. PPC Land had reported in May, citing 1492.vision researchers, that 54 US publishers held exclusive profile controls since March.

That pilot was this product's testing phase.

In Reach's full-year results, covered by Press Gazette, chief executive Piers North "told investors Google Discover traffic to Reach sites was down 46% in the second half of the year as the platform prioritises more user-generated content and video".

The knock-on numbers cover the Mirror, Express, Star and dozens of regional brands. Page views were up 6% in the first half, then down 8%, mainly put down to Google changes.

Full-year digital revenue slipped 0.9% to 128.9 million pounds.

North described the decline as general degradation rather than a cliff edge. On the February update's local emphasis: "There's some positive signs. They talk a lot about the importance of local relevance. We're yet to see that flow through."

Company-reported financials are the hardest currency on this beat. That is why one publisher's results sit alongside the panel studies: they are the same story measured two ways.

NewzDash compared its DiscoverPulse panel for 25th to 31st Jan against 8th to 14th Feb. The panel covers the top 1,000 domains and top 1,000 articles in the US, California and New York.

The clearest finding cuts against the obvious reading of Google's goals. Unique domains in the US top 1,000 articles fell from 172 to 158.

California fell from 187 to 177. The share of visibility held by the top ten rose.

More topics covered, reaching readers through fewer publishers.

New York moved the other way, 192 to 194 domains. That needs saying, because it stops the consolidation finding from being read as universal.

Regional personalization was the strongest confirmed effect. California local articles in the top 100 rose from 10 to 16.

They carried titles such as SFGate and the LA Times. Those were absent from the national top 100 in the same window.

On clickbait the study grades Google's goal "Directional" and says that "clickbait reduction" cannot be proven from headline markers alone.

What it does show is particular patterns losing ground. Yahoo fell from 11 articles to 6 in the US top 1,000, and from several top-100 placements to none.

NewzDash records Yahoo recovering most of that visibility by 14th Feb.

X.com posts from institutional accounts climbed from 3 to 13 in the US top 100. NewzDash is explicit that this data "cannot prove or disprove cannibalization" of publisher clicks, calling the comparison a directional sanity check.

Three caveats travel with the numbers. NewzDash sells Discover monitoring tools, a disclosure Search Engine Journal makes in its own write-up.

The panel is a sample, useful for identifying "changes in patterns and composition, not to claim absolute totals for the full Discover ecosystem".

And the study calls its findings "directional, not final" because the rollout overlapped an active news cycle.

Search Engine Journal adds the specifics. The post-update window covered the Super Bowl, the Winter Olympics and the ICC Men's T20 World Cup.

Any of those can lift news and sports visibility on its own.

Announced by John Mueller on the Search Central blog: "Today we've released the February 2026 Discover core update. This is a broad update to our systems that surface articles in Discover."

The post states three goals. The first is more locally relevant content from websites based in the user's country.

The others: "Reducing sensational content and clickbait in Discover", and more in-depth, original and timely content from sites with expertise, identified, in the post's words, "on a topic-by-topic basis".

The dashboard facts: rollout began 5th Feb at 17:00 UTC and completed 27th Feb. That is 21 days and 17 hours, for English-language users in the US.

Google wrote that "we're releasing this update to English language users in the US, and will expand it to all countries and languages in the months ahead".

Its historical weight needs its bound stated precisely. This is the only Discover-named update on Google's ranking-update history, which reaches back to November 2021.

Google had never before given Discover its own dashboard entry.

Our full write-up is on the Google updates hub: the February 2026 Discover core update. It has the rollout timeline and sources.

Google's statement to The Verge, carried in full by 9to5Google, settled the status of AI-generated headlines in the feed: "We launched a new feature last year in Discover to help people explore topics that are covered by multiple creators and websites."

The key sentences for publishers: "The overview headline reflects information across a range of sites, and is not a rewrite of an individual article headline. This feature performs well for user satisfaction, and we continue to experiment with the UI to help people click through and explore content on the web."

The same 9to5Google piece documented the failure mode.

A Tom's Hardware story surfaced under a generated headline about free-GPU and Amazon scams. The article itself did not support that headline.

Google's own guidelines police this surface for misleading previews, and publishers noted the asymmetry. Their titles must not promise what the page does not deliver, while the feed's own generated headlines sometimes do.

The figures come from Nic Newman's Journalism and Technology Trends and Predictions 2026 for the Reuters Institute. They rest on Chartbeat data spanning 2,576 sites worldwide.

Press Gazette's write-up carries the Discover numbers in text: "referrals to more than 2,500 publisher websites from Google Discover, a feed served to users on Google's native mobile apps and within its Android operating system, were down 21% year on year".

And: "In the US only, referrals from organic Google search were down by 38% year on year and Google Discover was down 29%."

Two labeling rules keep these numbers honest. The same panel measured SEARCH referrals down by a third globally and 38% in the US.

Search is a different surface. It must never share a label with the Discover figures.

And the longer window is a different measure again: since May 2023, Discover was down 18% while search fell 21%.

Read together with the share studies, the picture is a channel that grew to dominance and then started shrinking. Publishers' biggest Google pipe now carries less each year.

The study of how common AI is in the feed comes from Marfeel's Discover Monitoring, detailed by PPC Land.

As of December 2025, 51% of Discover feed positions in the US, Brazil and Mexico were AI Summaries.

The position breakdown is the sharper finding. In the US, AI Summaries were 21.6% of positions one to five and 13.9% of six to ten.

They were 56% of eleven to twenty, and 82.7% of everything after position twenty.

The deeper a user scrolls, the less the feed is publisher cards at all.

Where the taps go: 77% of US AI Summaries defaulted to an inline YouTube play. Only 23% linked to publisher sites.

In Brazil and Mexico, every measured AI Summary exit went to YouTube.

Marfeel's Xavi Beumala, quoted by Press Gazette: "Google Discover is no longer a publisher-first surface. It's becoming an AI platform with YouTube and X absorbing real estate that once went to newsrooms."

The scope: three test markets, not global. Marfeel's own methodology page does not state a sample size in either write-up we have read.

So the figures carry their attribution chain: Marfeel Discover Monitoring, December 2025, via PPC Land and Press Gazette.