How to audit the sources AI engines cite when they mention your brand
A practical checklist for evaluating whether the third-party pages ChatGPT, Perplexity, and Gemini cite about your brand help or harm your position.
A Series B fintech founder asked me last month whether the sources ChatGPT was citing about his product were good or bad. He could see his brand showing up in AI answers. He could see the citation list. He had no idea whether those citations were positioning him correctly or slowly poisoning his category story.
That question is now the first thing I check in every AI search audit I run. Not whether you are cited. Whether the pages citing you are saying what you need them to say.
What you are checking for
In the audits I run, I consistently find that what third-party sources say about you shapes the AI narrative more than your own pages. Your product page might be technically perfect, schema-complete, and clearly written. If the Reddit thread, the Capterra review, and the TechCrunch mention all position you incorrectly, that is the story AI engines will tell.
The audit is not complicated. You are reading the sources the engines already cite and deciding whether each one helps or harms. Then you are deciding what to do about the ones that harm.
Here is the checklist I use.
| Question | What it tells you | Action if the answer is no |
|---|---|---|
| Does this source position you in the correct category? | Whether buyers will compare you to the right alternatives | Brief the source (if owned), pitch a correction (if editorial), or dilute with better mentions |
| Does it attribute the correct capability to your product? | Whether the AI will cite you for queries you can actually serve | Same as above, plus update your own pages to over-clarify the capability |
| Is the claim current? | Whether you are being cited for a feature you sunset or a pricing tier you changed | Reach out for an update if the source is accessible, or publish a newer contradicting source |
| Is the tone favorable, neutral, or negative? | Whether the citation helps or harms consideration | If negative and accurate, you have a product problem. If negative and inaccurate, treat it like a correction pitch. |
| Is the source accessible to the engine that cited it? | Whether future retrieval will continue to pull this page | Check the URL. If it is paywalled, gated, or broken, the citation may drop out naturally. If it is open and wrong, you have work to do. |
The fintech founder's problem was positioning. ChatGPT was citing a 2024 product comparison article that called his platform "suitable for small teams with simple invoicing needs." He had moved upmarket six months earlier. The citation was accurate when written, outdated now, and showing up in every AI answer about invoicing software for mid-market finance teams.
That is the most common problem I see in the audits I run. Not that the sources are wrong. That they are right about a version of you that no longer exists.
What to do when a cited source is wrong or outdated
You have four options, in order of leverage.
Publish a newer, better source that contradicts the old one. In my experience running these audits, newer mentions from publications the engines already cite tend to displace older ones when the query topic suggests the user wants current information. The threshold is not high. Getting mentioned in an outlet ChatGPT already pulls from will often matter more than getting mentioned in three places it does not index.
Reach out and ask for an update. If the source is editorial and you have a relationship, pitch the correction. Explain what changed. Offer updated data, a new positioning sentence, or a link to a recent case study. If the publication runs regular roundups or "best of" lists, this works. If it is a one-time feature from 18 months ago, the editor will not touch it.
Dilute the wrong source with volume. If you cannot update the wrong citation and you cannot displace it with a single strong mention, your option is to bury it. Get cited everywhere the engines pull from. Community sites, comparison platforms, review aggregators, Reddit threads where your ICP asks for recommendations. You do not need to be in Wikipedia. You need to be mentioned accurately in five Reddit threads, two Capterra reviews, and one LinkedIn post from a respected voice in your category.
Ignore it. Sometimes the wrong citation does not matter. If it shows up for queries your ICP does not ask, if the page ranks low in the AI's source list, if the claim is minor and unlikely to shift consideration, let it go. I have seen founders spend two weeks trying to fix a citation that appeared in one Perplexity answer for a query with 11 monthly searches. That is not ROI. That is distraction.
The fintech founder chose option one. He pitched a case study to a publication ChatGPT cited frequently. The case study went live in July. By September the old comparison article had dropped out of the top five cited sources, and the case study was showing up in answers about mid-market invoicing tools. One new mention, correctly positioned, was enough.
How to prioritize which citations to fix first
Start with the citations that appear most frequently across the queries your ICP actually asks. Run ten to fifteen category and competitor-comparison prompts. Extract every citation list. Count which sources appear more than once. Those are your highest-leverage fixes.
Then filter by harm. A citation that positions you in the wrong category is worse than one that gets a feature detail slightly wrong. A citation that calls you expensive when you are mid-market is worse than one that misspells your CEO's name. Fix the positioning errors first. Fix the capability errors second. Ignore the cosmetic errors unless you have spare cycles.
In the citation audits I run for consulting clients, this prioritization usually surfaces two or three sources worth fixing and eight to twelve worth ignoring. The Pareto principle applies here harder than anywhere else in AI search. Most citations do not matter. A few citations decide everything.
If you are running training for a team that wants to take this work in-house, the citation audit is week two work. You query the engines with your core category and competitor-comparison prompts, you extract the citation lists, and you read every source that mentions you. Most of them will be fine. Two or three will be killing you. Those are the ones you fix.
The question is not whether AI engines cite you. The question is whether the pages they cite tell the story you need told.