In our last guide, we covered how to audit what AI Overviews say about your brand. You searched your business name, scored each result, and probably ended up with a few red rows.
Maybe Google says you close at 5pm. Maybe it's quoting a price you dropped two years ago.
This guide picks up from there.
Errors like these aren't rare. Oumi tested AI Overviews for The New York Times and found that about 91% contained the correct answer, but only 39% were both correct and fully supported by the sources they cited.
When something's wrong, the instinct is to "fix the AI."
You can't, at least not directly.
What you can fix is whatever the AI read. So the first job is working out what it read.
Find out where the error came from before you fix anything
Google has been fairly open about why AI Overviews go wrong. In a May 2024 post, Google's VP of Search, Liz Reid, wrote that errors usually come from "misinterpreting queries, misinterpreting a nuance of language on the web, or not having a lot of great information available."
For a business, that comes down to three causes. To find yours, open every source the AI Overview cites and ask one question: does this page say the wrong thing?
What you find in the cited sources | Cause | Where you fix it |
|---|---|---|
Your own page says it | 1: your site | On your site |
Someone else's page says it | 2: a third-party source | On that platform |
No source says it, or there's barely any source at all | 3: Google misread or guessed | In your wording, plus feedback to Google |
Diagnose every red row this way before you change anything. It takes about five minutes a row, and it stops you from rewriting a page the AI never read.
Cause 1: your own pages are wrong or contradict each other
This is the cause you can fix fastest, and it's more likely than you'd think. Google doesn't only pull from your newest pages. An Ahrefs study of 17 million citations found that pages cited in AI Overviews were, on average, 3.9 years old and last updated 2.9 years ago, the oldest of any AI platform Ahrefs measured. That pricing page from 2023 you forgot about can still be the source.
What it means: if your site holds two versions of the same fact, Google has to pick one. It won't always pick the newer one.
What to do
- Search your own site for stale facts. Try site:yourdomain.com pricing, site:yourdomain.com hours, and the old figure itself in quotes.
- Update, redirect, or remove pages carrying old information. Old announcement posts and archived promo pages are easy to miss.
- Keep your core facts (hours, prices, locations, services) on one clear page, and make every other page agree with it.
- Add Organization structured data to your homepage or About page, with sameAs links to your official profiles. Google says this markup can help it "disambiguate your organization" in search results. Check it with our Structured Data Test.
- Ask Google to recrawl the corrected pages with the URL Inspection tool in Search Console.
Our guides to content decay and keeping content fresh for AI cover how to stop this happening again.
Cause 2: a third-party source has it wrong
This is where most small businesses get stuck, because the page isn't theirs. Your Google Business Profile, Yelp page, Facebook page, and a dozen directory listings all hold facts about you, and they drift apart over time.
That costs you more than an AI summary. BrightLocal's 2023 Local Business Discovery & Trust Report found 62% of consumers would avoid a business if they found incorrect information about it online. And 56% said they run into incorrect business information at least every few months.
Find the inconsistencies
Build a simple grid. Columns: your website, Google Business Profile, Yelp, Facebook, and your three biggest directories. Rows: business name, phone, address, hours, prices, and main services. Visit each listing and fill it in.
Every cell that doesn't match your website is a possible source of the error, even if the AI Overview didn't cite it this time. If you haven't thought much about listing consistency before, our quick guide to NAP explains why it matters.
What to do, source by source
- Google Business Profile. Edit it from Search or the Maps app. Google reviews your changes before they go live, so check the edit status a few days later.
- Yelp. Claim your page and update the details through Yelp for Business.
- Directories. Claim the listing and correct it. If you can't claim it, use the site's suggest-an-edit or contact form.
- Wikipedia. If your business has a page, don't edit it yourself. Wikipedia's conflict of interest guideline strongly discourages it. Post an edit request on the article's talk page, disclose your connection, and cite a reliable source for the correction.
- Reviews, Reddit, and forums. You can't delete these. Reply publicly with the correct facts, stay polite, and don't argue about opinions. Your reply becomes part of the page. Our post on brand subreddits covers how to show up in those conversations.
- Articles and blog posts. Email the author or editor. Name the exact error, give the correct fact, and link to a page that proves it. Short, specific requests get answered far more often than vague complaints.
- YouTube videos. If it's someone else's video, contact the creator and ask for a correction in the description or a pinned comment. If it's yours, make that change yourself.
- Content that's false and damaging. Google has a legal removal process for things like defamation. It's for content that may be illegal, not content you simply dislike.
Cause 3: the sources are right, but Google got it wrong
Sometimes every cited source is correct and the summary still isn't. Oumi's study found that at the claim level, only 67% of claims in AI Overviews were supported by the sources cited. Roughly a third of the time, the AI says something its own sources don't.
Gaps cause the same problem. If nobody has clearly written down your pricing, or whether you're legitimate, Google fills the hole with whatever it can find.
What to do
- Report it. Click the thumbs-down icon under the AI Overview, choose "Report a problem," and add details. Google's help page says the report includes the search you ran and its results, so use the text box for the correct fact and the page that confirms it. You won't get a reply. Send it anyway.
- Rewrite the passage Google misread. One fact per sentence. Put dates on prices. Swap vague phrasing like "plans from $29" for the actual plan names and prices. Our piece on how LLMs parse content explains why plain structure helps.
- Fill the gap. If Google is guessing at "[brand] pricing" or "is [brand] legit," publish a page that answers the question directly. A clear FAQ on your own site gives the AI something better to quote than a three-year-old forum thread.
How long fixes take, and how to know they worked
This part needs patience. Google says crawling "can take anywhere from a few days to a few weeks", and requesting a recrawl doesn't guarantee anything. Third-party platforms add their own review time on top.
What to do
- Add a "fixed on" date to each row of the scorecard from the audit guide.
- Re-check two to four weeks later, with the same setup as the original audit: private window, signed out, same location.
- If nothing has changed after a month, go back to the diagnosis. You may have fixed a page the AI wasn't reading while the real source is still out there.
Know when the answer changes
Re-checking every red row by hand works, but it's slow and it only covers Google.
SEO Site Checkup's LLM Visibility Checker tracks how AI Overviews, ChatGPT, Gemini, Perplexity, Claude, and Copilot mention your brand, which sources they cite, and whether the tone is positive, neutral, or cautious. Set it up once your fixes are in. You'll see when the corrected answer lands, and you'll catch the next error before your customers do.
