Table of Contents
Why this needs measuring at all
AI visibility is whether AI assistants mention your business when someone asks a question you should be the answer to. Unlike rankings, there is no console that reports it — so if you do not check deliberately, you simply will not know.
You can measure it well enough with a fixed set of questions, a fixed schedule, and a record of what came back. The method matters more than the tooling.
Expect variation
The same prompt can return different answers on different runs, and answers differ by account, location, and model version. Treat any single result as a data point, not a verdict. Trends across repeated checks are what mean something.
Step 1: Build a fixed prompt set
Write 15 to 30 questions a real prospect would ask, and then do not change them — a stable list is what makes results comparable month to month. Cover four types:
- Category questions — "Who are the best SEO agencies in Los Angeles?" Tests whether you are in the consideration set at all.
- Problem questions — "My website traffic dropped after a redesign, who can help?" Tests whether you surface at the moment of need.
- Brand questions — "What is Evoba?" Tests whether AI systems describe you accurately.
- Comparison questions — "Evoba vs other SEO agencies" Tests how you are positioned against alternatives.
Brand questions are the ones people skip, and they are often the most alarming. An assistant confidently describing your business wrongly is a live problem.
Step 2: Run the same check across platforms
Run every prompt in each place your customers might ask, and record four things per result: were you mentioned, were you linked, was the description accurate, and who else appeared.
| Platform | Why it matters |
|---|---|
| ChatGPT | Largest assistant audience; cites sources when it browses |
| Google AI Overviews | Reaches people who never left Google |
| Perplexity | Search-first and citation-heavy, so attribution is explicit |
| Gemini / Copilot | Distribution through Google and Microsoft products |
Use a clean session where you can, so personalisation and chat history do not flatter the result.
Track competitors in the same sheet
"We were not mentioned" is much more useful when you also know who was. The names that keep appearing tell you which sources these systems currently trust for your topic — and those pages are a readable brief for what you need to publish.
Step 3: Check whether AI crawlers actually reach you
Prompt testing shows the output. Server logs show the input. Filter your access logs for AI user agents to confirm your pages are being fetched at all:
OAI-SearchBotandChatGPT-User— OpenAI search indexing and in-conversation fetchesPerplexityBot— Perplexity's crawlerGooglebot— still the gateway to AI OverviewsClaudeBotand other assistant crawlers
No hits at all usually means a robots.txt block, a firewall or bot-protection rule, or pages that require JavaScript to show anything. Fix that before analysing your prompt results, because nothing downstream can work until pages can be read.
Step 4: Look for AI referral traffic
When someone clicks a citation, it arrives in your analytics as a referral. Segment for sources such as chatgpt.com, perplexity.ai, and copilot.microsoft.com and watch the trend.
Volume here is usually small compared with organic search, and comparing the two head-to-head is the wrong reading. The number worth watching is direction over months, and what those visitors do — AI referrals often arrive further along in their decision, because they have already had their question answered.
Step 5: Make it a routine
Monthly is enough for most businesses. Same prompts, same order, same record. What you are looking for is movement: questions where you appeared and then stopped, competitors who have started showing up, and descriptions of your business that have drifted from the truth.
When you publish something aimed at a question you were missing from, note the date. It gives you a before-and-after rather than a vague impression that things improved.
Key takeaways
- No platform reports AI visibility for you — a fixed prompt set and a schedule are the method.
- Record mention, citation, accuracy, and competitors for every prompt.
- Check server logs for AI crawlers; being unreachable explains most invisibility.
- Segment AI referral traffic in analytics and judge it on trend, not volume.
- Single results vary. Trends across repeated checks are the signal.
Common Questions
How often should I check my AI visibility?
Monthly suits most businesses. The value comes from comparing like with like, so keep the prompts and the order identical between runs rather than checking more often with a changing list.
Why do I get different answers to the same question?
These systems are probabilistic, and results also vary by account, location, and model version. That is why a single result is a data point rather than a verdict, and why trends across repeated checks are what you act on.
What if AI describes my business incorrectly?
Treat it as a content problem. Make sure your own site states the facts plainly on pages that can be crawled, keep your details consistent across the web, and use structured data to remove ambiguity. Assistants generally reflect what they can find.
Is AI referral traffic worth tracking if the numbers are small?
Yes, on trend rather than volume. It is usually a small fraction of organic search today, but the direction over months tells you whether your visibility is improving, and those visitors often arrive further along in their decision.
Want this tracked for you?
We monitor how AI engines describe and cite your business, and work on the gaps.
See our AEO service All AI Search & AEO guides