citelity.

AI Overview citation disappeared: is it real or just noise?

Lost an AI Overview citation? Sample the query before you change anything, check Search Console's AI report and Google's 2026 update calendar, then work through the causes in order of likely impact.

By Ed Grows · Updated 13 min read

Your page was a source in an AI Overview last week. Today it isn't. Before you edit anything, run the same query eight to ten times over two or three days and count how often you're cited. One observation can't distinguish a lost citation from ordinary variation, because the set of sources behind an AI Overview changes from run to run and from model to model.

Then check two calendars: Search Console's AI impressions for that page, and Google's own list of updates. Only after that is it worth looking for something wrong with your page.

How much do AI Overview sources move?

The best measurement of a large shift comes from Google's model change on 27 January 2026, when it made Gemini 3 the default model behind AI Overviews worldwide (TechCrunch). SE Ranking compared 100,000 US keywords across 20 niches before and after:

About 42.4% of previously cited domains were replaced, and the average Overview grew from 11.55 to 15.22 sources. The churn was concentrated in the long tail: among the 500 most-cited domains, only one disappeared. SE Ranking also recorded a short bug window in late January when 10.63% of AI Overviews showed no sources at all, falling to 1.27% once it was fixed.

What nobody has published is a reliable figure for how much the sources vary between two runs of the same query on the same day. A number claiming 45.5% of citations change per re-run circulates widely; we couldn't find the study behind it. That gap is the reason to measure your own query rather than borrow a stability assumption.

Step 1: sample the query

Run the same query 8–10 times across 2–3 daysSigned out, in a clean browser profile. Record which sources appeared, in what order, on what date.6+ of 10Nothing was lostThe run that worried youwas normal variation.Re-check in a few weeks.1–5 of 10Partial lossYou still appear, lessoften. Often a competitorsharing the slot.Go to step 2.0 of 10Gone, for nowSomething changed: atGoogle, on a competitor'spage, or on yours.Go to step 2.

These thresholds are a working heuristic, not a published standard. With eight runs instead of ten, read them as shares: cited in most runs, in some, or in none.

Sample signed out, in a clean browser profile. Since 27 May 2026 Google lets users pick preferred sources that then appear more often for them in AI Overviews and AI Mode, so a personal, signed-in session can show you a version of the answer most searchers don't see.

Step 1b: check Search Console

Since 31 August 2026 every site has Search Console's Generative AI performance report (it started on 3 June as a beta for a subset of UK sites). It shows AI Overview and AI Mode impressions by page, country, device and date. There are no clicks, no queries and no citation text.

Use it as corroboration. If the page's AI impressions fell to near zero on the same dates your sampling points to, the loss is real and roughly dated. It can't tell you which query you lost or who replaced you; for that you still need the samples.

Step 2: does the timing match a Google change?

Check the date of the loss against Google's own calendar before looking at your page. Every entry below is on Google's Search Status Dashboard or in its documentation.

If your loss lines up with one of these, the cause is more likely on Google's side than on your page, and the useful response is to wait for the rollout to finish and sample again. If nothing lines up, the cause is more likely a competitor or your own page. Continue to step 3.

The source pool has also been drifting away from classic rankings. Ahrefs' study of 863,000 keywords (March 2026, updated May) found only 37.9% of AI Overview citations came from pages in Google's top 10, down from 76% in its July 2025 study, and 31.0% came from pages ranking beyond the top 100. A citation loss and a ranking loss are less tied together than they were.

Step 3: what changed?

Five causes, roughly in order of how often they're the answer.

1. A competitor now answers better

Open the current Overview and read each cited source. If they're more direct, more specific, or answer the exact sub-question the Overview is built around, you lost the slot to better content rather than to a fault on your page.

The fix is specific, not "more detail". List the facts the cited pages supply that yours doesn't (a number, a named source, a definition that fits the answer's shape) and add them to your page, near the top and in plain words.

2. You lost the organic visibility underneath

If the page also lost rankings recently, the citation loss is probably downstream, and the fix has to start there. Ahrefs measured the link between organic traffic and AI mentions per engine: a moderate 0.47 correlation for Google AI Overviews, weaker than Perplexity's 0.66 and stronger than ChatGPT's 0.33. For a post-update ranking drop, the sequence is in the traffic drop diagnostic.

3. The query's intent shifted

Sometimes the Overview starts answering a different version of the question: more transactional, or focused on another aspect. Same words, different reading. If so, either rework the page for the new intent or pick another query. Pushing the old framing won't bring the slot back.

4. The page is out of date

Refresh a page when its facts are out of date, not because of its age. For AI Overviews specifically, there's no evidence of a freshness preference. Ahrefs' study of about 17 million citations found AI assistants overall cite content about 26% fresher than organic results (1,064 against 1,432 days old on average), but AI Overviews showed no such preference and cited slightly older pages than organic results did.

You'll also see Seer Interactive's 65% figure quoted as evidence for freshness. Its log analysis of 5,000+ URLs found 65% of AI-bot hits went to content published in the past year (79% within two years, 89% within three). Those are crawler visits by publish year, not citations, and they don't isolate Google's AI Overviews.

If a refresh is due, change the visible text: corrected figures, current examples, what has changed since publication. Updating dateModified without that misrepresents the page to readers.

5. Your markup broke (a hygiene check)

A CMS update or template change can leave JSON-LD that no longer parses. That's worth fixing for the rich results that still exist, but it's an unlikely cause of a lost AI Overview citation: Google's guide to generative AI features says "there's no special schema.org markup you need to add". Check it last, or in passing. Details are in schema markup and AI Overviews.

Step 4: fix in order of likely impact

AI
Free tool · No signup
Free AEO Readiness Score
Paste content or a URL → a 0-100 score across ten AEO signals, with the fixes worth doing first.
Score your content →

When to let a query go

Look at the three most-cited sources in the current Overview. If they are consistently a major Wikipedia article, a large YouTube channel, a mainstream publisher or a Google property, your realistic ceiling for that query is an occasional lower-ranked citation. Both studies above point that way: YouTube is the single most-cited domain in Ahrefs' data (5.6% of AI Overview URLs), and SE Ranking found the 500 most-cited domains almost untouched by the Gemini 3 churn.

Where the cited sources are independent sites of roughly your size, the query is winnable, and that's where the effort belongs. Three weeks spent recovering one citation that isn't coming back costs more than a week each on three queries you can take.

What a lost citation costs

Less than a lost ranking used to, and more than nothing. Seer Interactive found organic click-through on informational queries with an AI Overview fell from 1.76% to 0.61% between June 2024 and September 2025 (reported by Search Engine Land). Ahrefs' February 2026 update, across 300,000 keywords, measured the top-ranking page losing 58% of its clicks when an AI Overview is present, up from 34.5% in its April 2025 study.

Being cited changes which side of that gap you're on. In the same Seer data, brands cited in an AI Overview received about 35% more organic clicks than uncited brands on the same queries.

FAQ

How often should I re-sample a query I care about?
Weekly is enough for most queries, plus a fresh sample a few weeks after any change you make to the page. Sampling daily mostly records noise. Keep the method identical each time (same number of runs, signed out, same location) so changes in the rate reflect the Overview rather than your setup.
Should I ask readers to add my site as a preferred source?
It can help with those readers: Google says preferred sources appear more often for the people who choose them, and it has documented a button sites can add. It won't restore a citation for everyone else, so treat it as a loyalty feature rather than a recovery tactic.

Sources cited in this piece

Updated October 1, 2026: corrected the freshness evidence (AI Overviews showed no freshness preference; Seer's figures are crawler hits), added Search Console's AI report and the 2026 update calendar, updated the Ahrefs CTR figure to 58%, and moved the markup check to last.

Keep reading

aeo · 8 min readClaude-SearchBot vs ClaudeBot vs Claude-User: what to allowAnthropic runs three crawlers, and Claude's web search also draws on Brave. What each bot does, what blocking it costs according to Anthropic, and how to check Claude can see you.saas · 8 min readSaaS AEO strategy: decision-stage content that gets citedB2B buyers ask comparison, alternatives and pricing questions, and AI answers increasingly cover them. Where a SaaS team's effort pays off, and the 2026 risk in self-ranking comparison pages.citation tracking · 10 min readTracking AI citations: how many runs make a rate you can trustAI answers cite different sources from one run to the next, so tracking means estimating a rate. How many runs and prompts make that rate trustworthy, what the engines report directly, and what the tools cost.