AI Citation Audit: What the Evidence Supports

An AI citation audit is a structured review of whether AI assistants mention, describe and link to your business when someone asks about your category. The evidence supports a narrow, useful version of this practice: check the assistants directly, record what they say, compare your answers with those of competitors, and fix the gaps in your own pages. It does not support the more ambitious claim that anyone can guarantee placement inside a model's answer. What follows is what a careful audit can and cannot tell you.

Coaching and leadership work sits awkwardly with all of this, because the people buying it rarely search in neat keywords. They ask an assistant a question about a problem, not a product. That shift matters for anyone running a practice, a consultancy or a small agency, and it is why the language of search engine optimisation has started appearing in conversations that used to be about referrals and reputation. The teams that handle it well tend to be the ones that treat it as a research habit rather than a growth hack.

What does an AI citation audit actually check?

A citation, in this context, is a reference to your business inside an answer produced by a tool such as ChatGPT, Perplexity or a comparable assistant. The audit is the process of gathering those references systematically and asking why they appear, or why they do not.

A practical audit covers four things. First, the questions themselves: the prompts a real prospect would type, written out in full rather than reduced to keywords. Second, the answers: what the assistant says, which businesses it names, and whether your own pages are cited as the source. Third, the citations behind those answers, where the tool exposes them, because a link to a third-party directory tells you something different from a link to your own site. Fourth, the gap between what the assistant says about you and what your pages actually claim.

That last point is where most audits earn their keep. An assistant that describes your service accurately is a small asset. An assistant that describes it wrongly, or omits the part of your work that clients actually buy, is a problem you can fix, because the fix usually sits on your own site.

How do you run one without expensive tools?

You can start with a free approach and a spreadsheet. The method below is deliberately manual, because the manual version teaches you what the automated tools are actually measuring.

1. Write twenty to thirty questions your ideal client would ask an assistant. Include the awkward, specific ones: sector, location, budget range, the problem rather than the service.
2. Run each question through two or three assistants. Note the date, because answers move.
3. Record whether you appear, who else appears, and which sources are cited.
4. Repeat the exercise for two or three competitors, using the same questions.
5. Compare the language in the answers with the language on your own pages. Where the assistant uses a phrase you never use, that is a signal worth investigating.

Paid audit tools exist and they save time on collection and reporting. Treat their scores as a prompt for investigation, not a verdict. A tool that evaluates whether you appear in an answer is measuring the answer, not the reason for it. The reason still has to be worked out by a person reading your pages.

What the answers can and cannot tell you

Assistants are inconsistent. The same question asked twice can produce different answers, and a model update can change everything without notice. A single result is an anecdote. A pattern across many questions, repeated over weeks, is evidence.

Monitoring shows you movement over time; it does not show you causation. If your citations improve after you rewrite a service page, you have a correlation and a reasonable hypothesis, not proof. That distinction matters when someone offers to sell you a guaranteed outcome.

Why do some businesses get cited and others do not?

Several factors appear repeatedly in audits, and none of them are mysterious.

• Clear, factual pages. Assistants draw on text that states plainly what you do, for whom, and where. Vague brand language gives them nothing to quote.
• Corroboration elsewhere. Mentions on other sites, directories, trade publications and professional bodies help establish that a claim is not yours alone. Backlinks still carry weight in ordinary SEO, and the same authority signals appear to feed into how confidently an assistant repeats something.
• Consistency. The same service described the same way across your site, your profiles and your listings. Contradictions create hesitation, and hesitation tends to produce a hedged answer or no mention at all.
• Structure. Headings, short paragraphs and direct answers to direct questions make a page easier to extract from. This is ordinary good practice, and it happens to suit machine reading as well as human reading.

Where these signals are weak, the assistant usually defaults to a larger, better-documented name in your category, even if that name is a poorer fit for the question. That is the commercial impact in plain terms: not a lost ranking, but a lost conversation at the moment someone is deciding who to contact.

How does this connect to SEO and to your existing team?

It connects more closely than the marketing language suggests. Search engine optimisation and citation work share a foundation: pages that answer a question clearly, backed by evidence that other people take you seriously. The difference is the output format. A search result gives a list; an assistant gives an answer, and an answer can only name so many businesses.

That is why the work belongs with whoever already owns your content, not with a separate function. In small teams, one person can run the audit quarterly. In larger ones, it usually falls to whoever handles SEO and whoever handles client-facing content, working together, because the fix is often editorial rather than technical.

There is also a question of alignment between what your sales conversations reveal and what your pages say. If clients consistently arrive asking about one service and your site buries it, the assistant will bury it too. The AI citation audit tool is best understood as a way of surfacing that mismatch, using the assistant as an unusually literal reader of your own material.

A caution about AI-generated content

Publishing large volumes of AI-generated pages in the hope of being cited tends to backfire. Assistants are not the only readers; clients are, and thin content damages the trust that made them look for you in the first place. If you use AI to draft, use it to draft from your own expertise, then have someone who knows the work review it for accuracy. Output quality is the constraint, not volume.

What should a report contain, and what should it not promise?

A useful report is short. It lists the questions asked, the answers received, who was named, which sources were cited, and the specific pages that need attention. It separates observation from interpretation, and it names the person responsible for each fix.

What it should not do is promise placement. No one controls what an assistant says, and any supplier claiming otherwise is describing something they cannot deliver. Be equally wary of custom pricing that scales with an outcome nobody can measure reliably.

The honest commercial case is narrower and stronger: you learn what the market is being told about your category, you find the pages that are letting you down, and you fix them. That work improves your site for human readers whether or not it changes a single assistant answer.

Frequently asked questions

How do I check if ChatGPT mentions my business?

Ask it the questions your clients would ask, in plain language, and read the answer carefully. Do this across several sessions rather than once, and keep a written record. If you want to compare assistants, run the same set of questions through Perplexity and any other tool your audience uses, then note where the answers differ. You can do this yourself for nothing; the discipline is in the repetition and the record-keeping.

Can I audit my own citations accurately, or do I need an agency?

You can do a credible first pass on your own, and for many small businesses that is enough. Agencies become useful when the volume of questions, locations or brands makes manual checking impractical, or when you need someone to turn findings into an editorial plan. Whichever route you take, ask to see the raw answers rather than a summary score, and ask how the supplier distinguishes a pattern from a one-off result.

Does appearing in AI answers replace SEO?

No. The two overlap heavily, and the fundamentals are the same: clear pages, accurate claims, corroboration from other sites, and content written by people who understand the subject. Citation work adds a specific habit, which is checking what assistants currently say and correcting the record on your own site. Treat it as an extension of what you already do, not a replacement for it.