Twenty prompts, three AI platforms, roughly two hours of your afternoon. That is what it costs to find out whether ChatGPT and Gemini recommend your business or your competitor’s.
Most UK business owners have never checked. They assume that ranking on Google means being visible in AI search. It does not. The two systems pull from different signals, and plenty of page-one sites never get mentioned in an AI answer.
This guide shows you how to audit AI search visibility yourself, using nothing but a browser and a spreadsheet. No tools to buy. No agency required. Just an honest picture of where you stand before you decide what to do about it.
Key takeaways
- Ranking on Google does not guarantee a mention in ChatGPT or Gemini. Different systems, different signals.
- A usable DIY audit needs 15 to 20 real buyer questions tested across at least three AI platforms.
- Log three outcomes per prompt: you are named, a competitor is named instead, or nobody is named.
- Prompts where nobody is named are the cheapest gap to fill and the fastest route to enquiries.
- Budget two to three hours for a first pass, then repeat it quarterly.
What is an AI visibility audit?
An AI visibility audit is a structured test of whether AI assistants name your business when buyers ask questions in your category. You run a fixed set of prompts across ChatGPT, Gemini and Google AI Mode, record whether you are mentioned, cited or ignored, then compare the result against your competitors.
Traditional SEO audits ask whether you rank. This asks whether you get recommended.
It is a different question with different failure points, from blocked crawlers to content that only renders once JavaScript loads. We covered that particular trap in our piece on why your website might be invisible to AI search.
How do you audit AI search visibility yourself?
Follow five steps: write down the questions your buyers ask, run each one across ChatGPT, Gemini and Google AI Mode, log whether you are named, record which competitors appear instead, and trace the sources those answers cite. A spreadsheet with five columns is all the infrastructure you need.
Step 1: Write down the questions your buyers actually ask
Start with how people buy, not how you describe yourself. Nobody types your service page headline into ChatGPT. They type things like “best commercial cleaning company in Manchester” or “how much does a facilities management contract cost”.
Write 15 to 20 of these. Mix problem questions, comparison questions and direct recommendation questions. Include at least three with a location in them if you serve a specific patch.
“In my years running marketing inside property and facilities management firms, the questions that actually won contracts were never the ones printed on the brochure.” – Marcus, Founder of Lobo Digital
Step 2: Run every prompt across all three platforms
Open a fresh, logged-out session on each platform. Personalisation and chat history will skew your results, so use an incognito window and do not sign in.
Run every prompt on ChatGPT, Gemini and Google AI Mode. Add Perplexity if your buyers are technical.
Do not rephrase a prompt because you dislike the answer. You want a clean baseline, not a flattering one. Paste each response into your sheet so you can compare the wording later.
Step 3: Log the outcome for every prompt
Score each prompt one of three ways: you are named, a competitor is named and you are not, or nobody is named at all.
Then note whether your domain was actually cited with a link, or you were only mentioned in passing. A citation sends traffic. A mention builds recall. Both count, but they are worth different amounts.
Tally the columns. If you appear in only a handful of your twenty prompts, you have a gap that is costing you enquiries every week.
Step 4: Note who gets recommended instead of you
List every business the AI names in your place, then count how often each one appears.
You will usually find two or three names dominating. Look hard at what they have that you do not: a detailed comparison page, a strong presence on review platforms, coverage in a trade publication, an About page that states plainly what they do and who for.
This is the most uncomfortable part of the exercise and easily the most useful. It shows you what good looks like in your category, in your buyers’ words.
Step 5: Trace the citations back to their source
Open every source the AI cited. Categorise each one: your own site, a competitor’s site, a directory, a review platform, a forum, or a publication.
The pattern matters far more than any single link. If answers in your sector lean on trade directories, your fix is a listings and reviews job. If they lean on long guides, it is a content job.
Check anything you plan to rewrite against Google’s guidance on content in AI experiences first.
How long does a DIY AI visibility audit take?
Budget two to three hours for a first audit of 20 prompts across three platforms, plus another hour to read the sources properly. Repeating it monthly costs the same again, because AI answers shift constantly and there is no historical record to fall back on.
| DIY manual audit | Done-for-you audit | |
|---|---|---|
| Time it costs you | 3 to 4 hours per round | About 15 minutes of briefing |
| Prompts tested | However many you have patience for | A full set built around your buyers |
| Competitor benchmarking | Manual counting | Included |
| Repeatable tracking | Starts from scratch each round | Tracked over time |
| Best for | Getting a first honest look | Acting on what you find |
Between the two sits our free AI visibility checker, which gives you a first read in a couple of minutes rather than an afternoon.
What should you do with the results?
Turn the audit into three fix lists. Prompts where nobody is named are your quickest win, because no competitor has claimed the ground yet. Prompts where a competitor is named need content better than theirs. Prompts where you are absent everywhere point to a technical or authority problem, not a content one.
Work the first list first. It is the cheapest revenue in the whole exercise.
For the technical bucket, confirm your structured data is valid and matches what is visible on the page. Google’s structured data documentation is the reference to use.
If you only want the fast version, our guide on how to check if your website is visible in ChatGPT and AI search covers the tool-led route in a fraction of the time.
Frequently asked questions
How many prompts should I test?
Fifteen to twenty is the practical minimum. Fewer than that and one odd answer skews your whole picture. More than thirty and the manual work stops paying for itself. Spread them across problem questions, comparison questions and direct recommendation questions so you can see which stage of the buying journey you are missing.
How often should I redo this?
Quarterly for most UK SMEs. AI answers move faster than Google rankings, but not so fast that a monthly manual audit is worth four hours of your time. Redo it sooner if you launch a new service, rebuild your site, or notice a competitor suddenly appearing in answers where you used to feature.
Do I need a paid tool to audit AI search visibility?
No. A browser, an incognito window and a spreadsheet will give you a genuinely useful first picture. Paid tools earn their keep when you need tracking over time, a larger prompt set, or evidence to put in front of a board. For a first look, manual testing is honest and free.
Why does ChatGPT give me a different answer every time?
Because these systems are probabilistic, not fixed. The same prompt can return different businesses on different days. That is exactly why you test a set of prompts rather than one, and why you record the outcome rather than trusting your memory of it. Patterns across twenty prompts are reliable. Single answers are not.
Ready to skip the spreadsheet?
Doing this yourself is worth an afternoon. Doing it every quarter, at scale, against a full competitor set, is a job.
Skip the DIY version: get your free audit done for you.
