In this article
Before you start
Step one: write twenty questions, not five
Step two: record four things per answer
Step three: read the sources, not just the answers
Step four: find the wrong answers
Step five: score it, then repeat monthly
What you will probably find
You do not need a tool to find out whether AI search is recommending you. You need about three hours, a spreadsheet, and the discipline to ask the same questions the same way every time.
This is the manual version of what Citera automates. Run it once and you will know more about your AI visibility than most of your competitors know about theirs.
Before you start
Open the engines in a logged-out, incognito window. Personalisation and chat history will skew your results otherwise — an engine that already knows you work at your company is more likely to mention it.
Cover at least four surfaces:
ChatGPT
Perplexity
Google AI Overviews (search normally, note the AI panel)
Gemini
If you sell outside your home market, repeat the exercise in the relevant language. Answers differ substantially between languages, and most brands never check.
Step one: write twenty questions, not five
The mistake is testing only the obvious query. Cover the whole buying journey:
Category discovery — “what tools track brand visibility in AI search”
Comparison — “X vs Y”, “alternatives to X”
Qualification — “is X worth it for a small team”, “how much does X cost”
Problem-first — “my competitor shows up in ChatGPT and I don’t, what do I do”
Industry-specific — the same questions with your vertical attached
Twenty questions is the minimum that produces a signal rather than an anecdote. Write them down before you run any of them, so you are not tempted to test only the flattering ones.
Step two: record four things per answer
For every question, on every engine, log:
Mentioned — were you named at all, yes or no
Position — first brand named, somewhere in the list, or an afterthought
Sources cited — which URLs the engine linked or referenced
Competitors named — everyone else that got a mention
The sources column is the one people skip and the one that pays. It tells you exactly which pages the engine trusts for your category, which is the map for everything you do next.
Step three: read the sources, not just the answers
Take every URL that appeared across your twenty questions and sort by how often it showed up. You will usually find a short list of five to fifteen pages doing most of the work: a Reddit thread, two or three review-site listings, a couple of industry articles, maybe a competitor’s comparison page.
That list is your target. For each one, ask:
Are we present on it at all?
Is what it says about us accurate and current?
Can we legitimately be added to it — a listing claimed, a correction submitted, a review earned?
Step four: find the wrong answers
Look specifically for answers where the engine says something about you that is false or outdated — wrong pricing, a feature you removed, a market you left. These are worth fixing before you chase new mentions, because a confident wrong answer costs you deals you never hear about.
Trace it back. The bad fact almost always comes from a specific stale page, and that page is usually yours.
Step five: score it, then repeat monthly
Turn the log into three numbers:
Mention rate — answers naming you, as a share of all answers
Citation rate — answers citing one of your own pages
Share of voice — your mentions against the most-named competitor
Then diarise it. Once a month, same questions, same conditions. The absolute numbers matter less than the direction, and the direction only becomes visible after the third or fourth run.
What you will probably find
Most first audits surface the same three things: the brand is invisible on problem-first questions, the engines are citing third-party pages rather than the company’s own site, and at least one widely-repeated fact about the company is out of date.
None of those need a tool to fix. They need someone to look.
Summarize this article with AI
Hand this page to your assistant of choice — the prompt is pre-filled:
ChatGPT · Claude · Perplexity · Gemini · Copilot · Grok · DeepSeek · Mistral · Qwen
