Search Console and Googlebot changes

A data gap is a day missing from a Search Console report, or a wrong status shown for an hour. It appears for every property at the same time.

That is the test: if your report changed and nobody else's did, it is your site. If everyone's did, it is the report.

September 2026 had three. The indexing "short blip" of 8th Sep, confirmed by John Mueller as a reporting bug, came first.

Then June's data went missing from the page indexing report on 11th Sep, and Crawl Stats lost 15th Sep.

November 2026 added a fourth. A logging error cut impressions on the Generative AI performance report for 9th to 11th Nov, Google's anomalies note says.

The November 2026 core update began inside that window. The report therefore cannot date the update's first effect on AI features.

Google has restored every gap in this record. Do not change your site in response to one, and annotate the date if you report on the data.

In August 2026 Google said its JSON-LD extraction now applies "a single pass of HTML unescaping", to bring the parser "up to JSON and other standards".

The practical effect, in Google's words: double-escaped entities "will no longer be unrolled". A product name written as "Smith & Sons" inside JSON-LD used to be corrected; now it is read as written.

Gary Illyes pointed to RFC 8259 section 7 for proper escaping. Use standard JSON escapes, or a Unicode escape for the ampersand.

Illyes added on 25th Aug that Google's crawlers do not parse JSON at all. They download, and indexing parses, so the change lives in indexing and very likely applies to Merchant Center too.

Check with the Rich Results Test. A page that validated before August and fails now, with an entity in the message, is the signature.

From about 13th Sep 2026, several tools that collect Google results reported being blocked more often. Derek Perkins of Nozzle told Search Engine Roundtable of an 80% drop in retrievable data, and the same at DataForSEO.

Sistrix's status page said on 16th Sep that Google "has made further changes to the way search results are delivered" and that its collection was "running at a reduced rate". Steve Paine of Sistrix said Google had been changing delivery "for about a year now, currently at a very high pace".

On 7th Oct a Sistrix support page called the divergent results "an industry-wide problem". Sistrix publishes only results that pass extra checks, has switched off on-demand updates, and gives no share.

On 10th Oct Nozzle published a study of 27.5 million SERPs. Since 13th Sep, 8.0% came back "poisoned": normal-looking pages missing most of the keyword's usual sites.

Nozzle's page, revised by 17th Oct, names the denominator: 17.8 million SERPs judged against each keyword's history. The 12th Oct copy gave only the 27.5 million.

Such pages were a quarter of all SERPs from 27th to 29th Sep. The peak was 28.6% on the 29th.

They ran at 8 to 11% a day from 8th to 13th Oct. On 14th Oct the share fell to about 3%.

It was 6.6% on 17th Oct and 7.0% on 18th Oct.

Nozzle's 19th Oct update carries data through 18th Oct. It puts the share since 13th Sep at 8.4% of 27.0 million judged SERPs.

Nozzle now says 68.9% of poisoned SERPs had the usual sites back within three days, and 92.7% within two weeks. The 10th Oct version said 77.9%.

Its advice stands: a one-day drop in late September is more likely this than a ranking change. On 19th Oct it called the pattern "too early to call it over".

The 21st Oct update, with data through that day, drops that line. Of 29.9 million SERPs analysed, 8.7% of page-1 results since 13th Sep came back poisoned.

A new count includes page 2. Of SERPs judged on both pages, 22.6% had either page poisoned, against 0.28% before, peaking at 43.5% on 30th Sep.

The page-1 share rose again after the mid-October lull. It was 11.4% on 19th Oct, 9.0% on the 20th and 15.6% on the 21st.

Nozzle reads the pattern as a response to AI scraping, with rank trackers "caught in it".

Update, 4th Nov: Nozzle's 1st Nov revision says the pattern "climbed to its worst yet at the end of the month". Page 1 or page 2 was poisoned on 64.0% of SERPs judged on both pages on 31st Oct.

Page 1 alone peaked at 34.8% on 27th Oct and runs at 14.3% since 13th Sep. Only 46.9% of poisoned SERPs now have their usual sites back within three days.

Update, 8th Nov: Nozzle's 3rd Nov revision says the poisoning "hit its worst level yet in early November". Page 1 or 2 was poisoned on 68.6% on 2nd Nov and on 27.6% since 13th Sep.

Page 1 alone runs at 15.0%. Recovery within three days fell to 45.1%, and the 3rd Nov row has the rest.

A second Nozzle study, dated 22nd Oct, counts a different failure. Since 16th Sep, page 1 or 2 answered one word of the query on 13.2% of SERPs.

The peak was 39.9% on 28th Sep.

Nozzle's 3rd Nov revision gives 13.1% since 16th Sep. It says the weekly rate was "back to 10.7% in early November", up from 7.6% in late October.

SE Ranking's 3rd Nov study measures the same one-word pattern on 10,000 US desktop keywords. It affected 57.3% of them on 17th Sep and 67.2% by 24th Sep, nearly always from position 9 down.

Keyword.com's 16th Oct study adapts Nozzle's definition to its own customer data. It finds 13.1% of SERPs poisoned from 13th Sep to 14th Oct, against 0.3% in August.

Every figure in this section comes from a vendor that sells rank tracking, each on its own sample and definition.

DemandSphere reports the same pattern at a different scale. Its 7th Oct post puts spurious pages at about 11% of its daily SERPs, a figure it says "could change rapidly".

Ray Grieselhuber dates the heaviest blocking to "right around September 11th", two days before the reports above. He writes that Google's target is the AI labs, with rank trackers "caught in the crossfire".

The two vendors measure different keyword sets. The 7th Oct post did not say how it counts a spurious page, so 8% and 11% were two estimates.

Update, 27th Oct: DemandSphere's 23rd Oct post gives a method and a new figure. It says "6.7% of the SERPs we collected were poisoned".

Each collected page is scored against the keyword's own history. Sites on at least 60% of the last 30 daily pages count as regulars.

A page showing under 40% of them is poisoned and collected again, up to three times.

The post does not say which period the 6.7% covers, beyond a chart dated 23rd Oct. It says "We are not claiming certainty on any individual page".

SE Ranking gives a third figure of a different kind. On 16th Oct it said "roughly 2% of tracked results" were still affected after its fixes, "and that share continues to shrink".

That counts what its checks still miss, not what Google served. It sits beside the other two figures, not between them.

AccuRanker gives a fourth figure, from its own stored data. Its 19th Oct post dates the onset to late on 24th Sep.

On that first day, 30.2% of desktop and 25.7% of mobile page-1 results in a fixed sample were swapped. Its filtering brought that below 2% within two days and under 1% by 17th Oct.

It says Google "tightened its detection again on October 15, mostly on mobile".

Its 26th Oct update reports two more steps, late on 20th Oct and late on 23rd Oct. On 24th Oct "Google poisoned 40% of our first desktop fetches", and 3.2% of its desktop sample still had a swapped page 1 after its checks.

By 25th Oct that was 0.6% on desktop and 1.0% on mobile. The 40% counts first fetches, not stored data, so it is a third unit on this page.

AccuRanker also sees smaller rounds of the same pattern from late January 2026. From April to July they ran at 7 to 10 times the 2025 level.

Its later shares describe what passed its checks, so they do not compare with Nozzle's 8.4%.

Semrush told Search Engine Roundtable it has seen the issue "since late summer", with "limited" impact on the rankings its users track. It gives no figure.

On 27th Oct Semrush published its own post. It calls the pattern "an industry-wide issue affecting data providers across the SEO space" and again gives no share.

It says its systems "detect clearly invalid results during collection and exclude them from our database", and that it has "added validation checks" to collect doubtful pages again. Search Engine Land, which Semrush owns, covered the post the same day.

SEOmonitor's undated help article, read 23rd Oct, says "around 8% of the daily results Google returns to us come back degraded". From 9th to 13th Oct some bad results reached its platform, touching "around 10%" of tracked keywords.

Search Engine Journal compared five vendor statements on 23rd Oct. "As of Oct. 22, none of these vendor statements include a statement from Google", it writes, and "none cites independent testing of how well their checks work".

Ahrefs gave a band rather than a share. Its 22nd Oct update says "between 5% and 20% of search results" are affected, "depending on the data provider".

The post names no sample, start date, country or device. It does not say which providers it counts, so the band is not a measurement of Ahrefs' own data.

Google has not commented. This matters for every rank-tracker figure on our statistics pages.

A measurement taken after 13th Sep may rest on a smaller or different sample than one taken before.

Update, 5th Nov: DataForSEO's 2nd Nov post describes a second mechanism. Google now throttles how fast the goto links on one results page resolve to their destinations.

Past the first five or six, an interstitial adds delays of up to 120 seconds a link. The limit is per page, not per IP address, and DataForSEO's fastest modes now resolve about half of goto URLs.

Search Console is unaffected. It is Google's own report, not a scrape.

Update, 10th Nov: a court dismissed Google's amended DMCA complaint against SerpApi for a second time on 9th Nov.

SerpApi scrapes Google results for its customers. Google's claim is that SerpApi circumvents SearchGuard, the JavaScript challenge it launched in Jan 2025.

The Northern District of California held that Google had not alleged copyright owners authorised SearchGuard. Google has 21 days to replead, or the case ends "with prejudice".

The order is a pleading ruling, not a finding on how Google blocks scrapers. It never mentions rank trackers.

A platform property is a Search Console property for an Instagram, TikTok, X or YouTube account rather than a website. It shows which search terms lead people to your posts on Google Search and Discover.

Google introduced them on 7th Jul 2026 after an earlier experiment, with Performance, Insights and Achievements reports. They became available to everyone worldwide on 29th Jul, with a guide to analysing social and video performance.

Setup is in Search Console's property selector: add a property, pick one of the four platforms and authorise the connection. Our Instagram tracker covers the launch from the creator's side.