AI is citing wrong information about my brand: how to find and fix the source
When AI is citing wrong information about your brand, the cited page is usually the problem. Three cases, the fix for each, and three ways to find the bad source compared.
A wrong answer with a footnote is better news than it looks. When an AI engine is citing wrong information about your brand, it is also telling you where it read it. That link is the lead. Somewhere behind it is a page that is stale, a page that is simply wrong, or a page that is fine and got misread. Each needs a different fix, and the work is mostly in telling them apart.
This post covers the three cases, how to recognize each one, and what to do. It then compares three ways of finding the bad source: clicking citations by hand, a tracker that lists citations, and Promptwatch's citation analytics with Page Tracker and crawler logs.
Quick verdict
Open the cited page before you do anything else. If it is your own page and it says the wrong thing, update it. If it is a third-party page, a directory listing, a competitor's comparison or an old forum thread, correct it at the source where you can and publish a clear page of your own where you can't. If the page is correct and the answer is still wrong, the engine blended or misread it, and you need to make the fact harder to misread.
Clicking citations by hand works for a few answers. At any real volume, you need citation data across prompts and engines. Promptwatch goes one step further than citation lists: it shows the citation rate per page alongside verified AI crawler visits, so you can confirm that a corrected page was re-crawled and is being cited again.
Three cases at a glance
| Case | How you recognize it | Fix | How you confirm it worked |
|---|---|---|---|
| Your own stale page | Cited URL is on your domain and states the old fact | Update, redirect or retire the page | Crawler visit after the change, then the claim fades in answers |
| Wrong third-party page | Cited URL is a listing, review, comparison or thread that states the wrong fact | Fix the profile, ask the editor, reply in the thread, publish your own page | The page changes, or citations move to your page |
| Correct page, wrong answer | Cited page is accurate; the answer isn't | Make the fact explicit and unambiguous on the page | Same page stays cited and the answer matches it |
Case 1: your own stale page
This is the case teams least want to find and most often do. The old pricing page that was never deleted. A 2023 launch post listing the plan you retired. A help article describing a limit you raised months ago. The engine cites it because it is still live, still crawlable and, from the outside, still official.
Fixing the page is the easy part. The harder part is finding every other page on your site that carries the same fact, because the engine will happily switch to the next one. Promptwatch's Fact Check, on Business ($579 a month), Agency Scale ($799) and Enterprise, does this directly: you mark the claim Inaccurate, and it reads pages crawled from your site for ones that state the claim or its opposite, with quotes and a suggested next step. It uses Agent Credits and shows an estimate first. Content Refresh then suggests diff-style updates for the existing pages, so you are editing rather than rewriting.
After the edit, confirm the engine has a chance to see it. On each sitemap page, the Agent Analytics tab shows crawler visits and crawler logs for that URL, and the AI Visibility tab shows the URL's citation rate and position over time, with the prompts and responses that cite it. A corrected page with no crawler visit since the change hasn't been read yet, so judging the fix at that point is premature. The Content tab shows the page as crawled, which is how you check that the crawler saw your new text and not a cached version.
Case 2: a wrong third-party page
These come in a few familiar shapes:
- A directory or review listing with your old pricing or category. Promptwatch's Citation Trends classifies citations into 20 source types, and one of them is Directory, covering G2, Capterra and Trustpilot-style sites, so you can see how much of the problem lives on listings.
- A competitor's comparison page that describes your product as it was, or as they would like it to be. Citation search marks competitor domains, which makes these quick to spot.
- An old Reddit thread where someone complained about a bug you fixed long ago. The Reddit citation report shows which subreddits and posts get cited.
The fix depends on who controls the page. Listings usually let you claim and edit the profile, so do that first. For editorial pages, send the editor a short, specific correction with a link to the page on your site that states the current fact. For a forum thread, a reply from a clearly identified staff account with the current facts is fair game, while arguing with the original poster is not. For a competitor's comparison, ask, but expect nothing, and put the effort into your own comparison page instead.
Reddit is worth a note on timing. Promptwatch's data reports found Reddit's share of ChatGPT citations fell from about 4% to about 0.5% on August 14, 2026. An old thread may matter less in ChatGPT than it did a few months ago, so check which engines still cite it before you spend a week on it.
Offsite mentions help with the rest. They list third-party pages that name your brand without linking, ranked as opportunities, which gives you a list of pages worth checking before they become the next cited source. Page Tracker lets you follow any specific URL, yours or a competitor's, so you can watch the bad page and see when engines stop citing it.
Case 3: a correct page, misread or blended
Sometimes you click through and the cited page is right. The price is current, the feature list is accurate, and the answer is still wrong. Either the engine blended two sources, taking your product name from one page and a competitor's price from another, or it misread yours. Common ways that happens on a hypothetical brand, Acme: the monthly price sits in a toggle that the crawler only sees in its annual state, the plan limits live in a table without a heading, or the feature list and the "coming soon" list share a section.
The fix is to make the fact hard to misread. Put the price and the billing period in the same sentence. Label what is available now separately from what is planned. Keep the facts in text that a crawler gets without running scripts. The sitemap Content tab, with its preview and markdown views of the page as crawled, is the fastest way to see what the crawler actually got. If the two-source blend is the problem, look at the answer's other citations. The second source is often the one introducing the wrong detail, which turns this into Case 2.
Three ways to find the bad source, compared
| Manual citation clicking | A tracker with citation lists | Promptwatch citation analytics, Page Tracker and crawler logs | |
|---|---|---|---|
| What you see | Sources for the answers you read | Cited URLs and domains across tracked prompts | Cited URLs with positions, trends, source types, and AI crawler visits per page |
| Spots directory and review sources | By eye | Depends on the tool | Directory source type in Citation Trends |
| Spots competitor pages | By eye | Depends on the tool | Competitor domains marked in citation search |
| Confirms a re-crawl | No | Not in our catalog entries | IP-verified crawler logs and a per-URL Agent Analytics tab |
| Effort | High; answers change run to run | Medium | Low once logs are connected |
| Plan | None | Varies | Citation analytics on paid plans; crawler logs from Professional ($245/mo); Fact Check from Business ($579/mo) |
Source for the Promptwatch column: Promptwatch llms.txt, ai-agent-instructions and changelog, read November 3 and 4, 2026.
Manual citation clicking
Ask the question, open each footnote, read the page. It's free and it's how everyone starts. It only covers answers you saw, though, and the same prompt can cite different sources tomorrow. Our fact-checking method is the manual workflow done carefully.
A tracker with citation lists
Several AI visibility tools list cited sources. From our catalog: AthenaHQ tracks citations as part of its GEO score, with the Athena Citation Engine on Enterprise (Starter is $295 a month). Scrunch AI shows the sources behind answers and reports AI agent and bot traffic, from $250 a month. Profound reports citation share and answer-level analytics on a custom Enterprise plan. A list tells you which URLs are cited. Our catalog entries don't document per-page crawler logs with a crawl-to-citation path for these three, which is the part you need to confirm a fix.
Promptwatch
Citation analytics has All Citations, URL Positions and Citation Trends tabs. Citation Trends adds content and source types, citation rate over time, and per page the first cited date, ramp-up days, peak rate and decay half-life. Page Tracker follows specific URLs. Agent Analytics, from Professional, gives IP-verified AI crawler logs with a crawl-to-citation path and a citation rate per page. That rate helps you tell the two failure modes apart: a page that is crawled often but rarely cited has a content problem, and a page that is barely crawled has an access problem. Our post on AI crawls versus citation crawls explains how the bot buckets are counted.
Who should pick which
- You have one wrong answer and want it gone: click the citations by hand, fix the page, and re-check in a couple of weeks.
- You need to know which sources carry your brand across many prompts: any tracker with citation lists. See our ranking of tools that track which sources AI engines cite.
- Your wrong citations are mostly your own old pages: Promptwatch Business, for Fact Check to find every page with the old fact and crawler data to confirm the re-crawl.
- Your wrong citations are mostly listings and forums: Promptwatch, for source types, offsite mentions and Page Tracker on the pages you're trying to get changed. Crawler logs, which you'll want once your own replacement pages go live, start on Professional.
Before you start fixing sources, sort the mistake itself. Our guide to what to do when AI makes mistakes about your brand covers the types that have no cited source at all.
Our recommendation
Treat every wrong citation as three questions: whose page is it, what does it say, and has the engine read the corrected version yet. Manual checks answer the first two for a few answers. Promptwatch answers all three at volume. Of the tools in this post, it's the one that documents a link from the cited page to verified crawler visits for that URL, so you can tell a fix that hasn't been read from a fix that didn't work. To see it on your own brand's citations, book a demo and bring the three wrong answers that bother you most.