How AI Overviews decide what to say
An AI Overview is generated from documents Google's systems retrieve for the query — largely pages that already rank, plus Google's own data like Business Profiles. That has two consequences. First, the wrong claim almost always exists on some page Google can retrieve; find it by clicking the Overview's source links. Second, you correct an Overview by correcting and strengthening sources, not by arguing with the summary. Audits have found significant issues in a large share of AI answers generally, so you're not unlucky — you're dealing with a known failure mode.
The correction sequence
1.Identify the retrieved sources
Expand the AI Overview and open every source link. One of them — or your Google Business Profile — almost certainly carries the wrong fact. That's your target list.
2.Correct what you control
Fix the fact on your own pages: state the correct information plainly in the first 100 words, update the page's modified date, and make your structured data (Organization/Product JSON-LD) agree. Update your Google Business Profile — for hours, status, and location questions it's often the decisive source.
3.Request corrections on sources you don't control
Directories, old articles, review pages: contact the owners with the correct fact and a reason to update. This is the slowest step — budget days to weeks per source.
4.Get the fixes re-crawled
In Search Console, request indexing for your corrected pages. Google re-crawls on its own schedule; the request just gets you in line sooner.
5.Use the feedback button — but don't rely on it
Send feedback on the Overview (thumbs-down → report). It's feedback, not an override — it rarely resolves a wrong claim by itself and leaves no record. Treat it as a supplement to source fixes, never the plan.
6.Document while you wait
AI Overviews are ephemeral — they change and leave no history. Capture the wrong Overview with timestamps on a regular schedule. A 2026 German ruling held that AI Overview statements are legally attributable to Google, and preserved records are what make any escalation possible.
Realistic timelines
Owned-page fixes plus a re-crawl request: expect weeks, not days. Third-party source corrections: add their own days-to-weeks on top. If the Overview persists after your sources are demonstrably correct and re-indexed, escalation paths (formal complaints, legal demand letters) exist — the German precedent established attribution — but they're slow and uncertain. Throughout, weekly monitoring tells you whether the fix landed or you're still bleeding searchers.
Frequently asked questions
Can I directly edit or remove a Google AI Overview?
No. There is no edit or removal mechanism. You influence an Overview indirectly: correct the sources it retrieves, strengthen your own ranking pages and Business Profile, request re-indexing, and wait for Google's systems to re-read them.
Does the thumbs-down feedback button work?
It's a legitimate feedback channel, but practitioner experience is consistent: it rarely resolves a wrong claim on its own. Google ignored even a formal cease-and-desist in one documented case until a court intervened. Use it alongside source fixes, not instead of them.
How long does correcting an AI Overview take?
Weeks is the honest answer: your page fixes need re-crawling and the Overview needs regenerating, and third-party sources add their own lag. Anyone promising days for AI Overviews specifically is guessing.
Is Google legally responsible for a wrong AI Overview?
In Germany, a 2026 Bavarian court ruled AI Overview statements are attributable to Google (under appeal). In the US the question is unsettled and early cases have been hard for plaintiffs. Either way, escalation requires timestamped records of what the Overview said and when.
How do I know when the Overview is fixed?
Re-check on a schedule — Overviews vary by phrasing and over time, so a single clean check proves little. PositionBird samples fact-check questions weekly across engines, flags a claim only when it's persistently wrong, and confirms resolution only after consecutive clean cycles.
Find out what AI says about your store — before customers do
Run a free one-shot accuracy check right now, or let PositionBird monitor your business facts weekly across ChatGPT, Perplexity, Gemini, and Claude — with statistical debouncing so you're only alerted to real patterns, and a hash-stamped evidence record of every answer.
Wondering how common this is? Read the research: how often AI gets small businesses wrong →