An AI citation audit is a dated, repeatable check of whether search systems can access a site, understand the business and surface it for a small fixed set of customer questions. It records the question, platform, answer and sources so you can compare like with like after the website changes.
What should the audit measure first?
Begin with conditions the business controls: public access, a clear identity, service and location language, direct answers and accurate structured data. These do not guarantee a citation. They tell you whether the website has a fair chance to serve as usable evidence.
Google says pages in its AI search features must meet normal Search eligibility. OpenAI, Anthropic and Perplexity publish separate search-crawler names and controls. Check the search-oriented crawler relevant to each product instead of treating every AI bot as the same thing.
The free AI IQ check turns those page-level conditions into a saved baseline. If a supporting request fails, the report shows that measurement as unavailable rather than quietly scoring it as zero.
Which questions belong in the test set?
Use five to ten questions a real customer might ask while choosing or preparing to hire. Include the town where location matters, and avoid prompts designed only to force the company name into the answer.
A practical mix looks like this:
- Who provides this service in this town?
- What should I look for when choosing a provider?
- What affects the cost of this service?
- How long does the work usually take?
- What should I do before I book?
Use the same wording for the baseline and follow-up. If you rewrite the prompt each time, you cannot tell whether the website improved or the question changed.
What should you record for every answer?
Record the full question, platform, date, location context, answer, cited sources and whether the business appeared. Save enough context to reproduce the observation. A cropped screenshot that hides the prompt cannot support a serious before-and-after comparison.
If a platform gives no sources, record that too. Do not convert “no citation shown” into “the system did not use the site.” Those are different claims.
OpenAI says visits from ChatGPT search carry utm_source=chatgpt.com, so analytics can provide a second signal when referral traffic occurs. Referral traffic still does not reveal every answer in which a brand was mentioned.
How do you compare competitors fairly?
Run the same page checks and the same question set for each business. Keep directories and publishers separate from direct local competitors. A directory appearing as a source may explain where an answer found information, but it is not a company competing to perform the work.
Look for repeated differences: the competitor clearly names a town you never mention, answers a question your site avoids, exposes stronger first-hand evidence or remains accessible to a crawler your firewall blocks. Why AI recommends competitors walks through those gaps.
When should you run the follow-up?
Run it after meaningful work has had time to become publicly accessible and searchable. Keep the grader and question set stable. Record exactly what changed between the two dates.
A responsible result sounds like this: “The same page moved from 52 to 68 after crawler access, service-area language and answer sections were corrected.” It does not leap from a technical score to a claim about revenue. Leads and sales need their own evidence.
For ChatGPT-specific details, read how small businesses get found in ChatGPT search. The same discipline applies across platforms: distinguish documented crawler controls from assumptions about how every answer is assembled.
What do you do with the findings?
Rank findings by impact and effort, assign the first action and keep the page evidence beside it. That turns the audit into Your 90-Day AI IQ Roadmap: a measured baseline, a prioritized sequence and a follow-up check.
If the site needs execution as well as diagnosis, LocalCraft’s website plans show the monthly options and build price. The audit should make the scope clearer before anyone commits to the work.