Has Google removed AI Overviews? What the latest reports show
Google still documents AI Overviews as available. Trace the removal reports and use a five-field check to verify what changed.
Google’s current help documentation continues to describe AI Overviews as an available Search feature. A check on November 19, 2026, found no confirmed Google announcement of a general withdrawal. Recent reports concern specific searches, interface elements, and experiments. The exact query and the element that disappeared matter when assessing a removal claim. Google’s AI Overviews documentation
What Google’s documentation confirms
Google says its systems display AI Overviews when they determine that a generated response would be especially helpful. Its help page lists supported countries and languages and explains how an overview can lead into a conversation in AI Mode. The documentation also says the Web filter displays text-based links without features such as AI Overviews.
That creates a straightforward check. A result page using the Web filter is expected to omit the feature. An ordinary results page can also lack an overview for a particular query. Neither observation, by itself, establishes a product-wide change. Keep the search filter visible in any screenshot used to support a removal report.
Follow the dated report to the missing element
An October report documented AI-generated summaries appearing without the AI Overview label. Search Engine Roundtable added an October 22 update saying the label had returned. The summary and its heading are two different observations a reader can verify in the screenshots. The report did not establish that the feature was permanently removed. The October report and its update
Citation presentation has also changed in observed result pages. The November 18 Search Engine Roundtable recap linked coverage of an unfaded citation block. Source cards can appear differently while a generated answer remains. A useful record therefore captures both the answer and where readers can open its supporting links. November 18 search recap
An earlier removal concerned selected health searches following an investigation, rather than a general shutdown. That event was reported in January 2026. A headline recirculating today should retain the original event date and query scope. The Guardian’s January report
Use a five-field check before interpreting a screenshot
Start with the exact query, copied as text. Record the country and search language, desktop or mobile device, account state, and selected filter. Add the date and local time. These conditions make a second observation easier to compare.
Then record five visible fields independently. Is an AI-generated answer present? Is the AI Overview label present? Are supporting links visible before expansion? Is an expansion control available? Is AI Mode offered? Save the whole result page as well as a close-up of the relevant module. A cropped screenshot of an absent label cannot show whether the answer or source links remain elsewhere.
Repeat the query in the same conditions first. If you then change account state or device, treat that run as another condition in your log. Keep every run, including those with no overview, to avoid describing only the screenshots that fit the initial claim.
This is a proposed verification procedure. This article’s evidence comes from Google documentation and the dated reports linked above. It does not claim a fresh, representative sample of Google result pages.
A worked example for a monitoring report
Suppose a team monitors 20 fixed queries once a week. Twelve produce an overview in the first run, and eight do so in the next. Its observed coverage fell from 60% to 40%, a decline of 20 percentage points. The result describes those queries under the recorded conditions. It cannot establish global coverage or explain the difference.
Now suppose the same eight answers still appear, but their label is missing in three cases. A label-only detector would count five overviews and report 25% coverage. A review of the answer text and controls would count eight, or 40%. This illustrative calculation shows why monitoring should keep the label field separate from the answer-presence field.
Manually check a small sample after a layout change, and preserve the raw capture the detector uses. If the parser reads a different part of the page after an interface experiment, fix the extraction rule and annotate the break before comparing trend lines.
What publishers should check next?
For a publisher, the practical question is whether a specific page is visible and accessible through the answer’s source links. Record the exact destination URL, link position, and whether an expansion is required. Compare those observations with the page’s organic placement and its measured referrals over an appropriate period.
Use the AI visibility measurement guide to record appearances, referrals, and outcomes in separate fields. The SERP testing guide explains how to hold test conditions steady. If an AI impressions chart changes, check the Search Console reporting guide and Google’s documented data anomalies before attributing the shift to an interface change.
The evidence checked on November 19 supports continued availability, alongside query-specific and interface changes. A future removal claim needs an exact scope, a dated source, and a reproducible observation.
How to get rid of an AI Overview for a search
Google’s current help page describes AI Overviews as a core Search feature and directs users who want text-based links to the Web filter. Run the search, select Web where the interface provides it, and check that the filtered result is the view you intended. Interface placement can vary, so follow the controls actually shown on your device. Google’s current instructions
Turning off an AI experiment in Search Labs is a separate action. It changes participation in that experiment. Record whether your screenshot shows ordinary Search, the Web filter, or a Labs configuration before describing the outcome. The distinction explains why two people can report different experiences without either having observed a general removal.
Check a viral removal claim against its actual scope
A November 20 research check surfaced a recent TikTok video claiming Google was testing the removal of AI search features. Treat the video’s wording as a claim to verify. Ask which feature, which query, which market, and which account condition the creator observed. A missing label, a missing answer, and a missing entry point require different evidence.
Use a small comparison sheet with fields for query, time, language, market, device, account state, selected filter, answer presence, and source-link presence. Keep an unchanged query set for a repeat check. If a feature is absent in three of ten searches, the narrow observation is three absences in that sample. It does not establish removal for all queries or all users.
A useful countercheck repeats the same query in the ordinary results view and the Web-filtered view. Record both outcomes without assuming that either represents every search session. A difference can be consistent with the selected filter. It cannot, by itself, explain Google’s business motives or justify an advertising-profitability decision.
For publishers, retain a traffic and citation baseline while investigating the interface. A changed screenshot is one observation, and a change in qualified visits is another. The AI visibility measurement guide explains how to keep those outcomes separate. Teams using new models to review the evidence can use the GPT-6.2 evaluation method.
The November 20 update adds a practical Web-filter explanation and a verification method for circulating removal claims. The method is a proposed audit, not a completed population study of Google result pages.
Keep learning
Continue this topic
Audit citations
Locate the failing answer layer
Compare evidence
Separate mentions, citations and visits
Measure referrals
Check AI search visits in GA4
AEO & AI Search
Ask a question or join the discussion