How to measure AI visibility when there is no Search Console

Classic search has Google Search Console: an authoritative, first-party record of impressions, clicks and positions. AI search has nothing equivalent. No platform publishes how often it named your brand, in what context, or to how many people.

So the measurement has to be constructed. The good news is that a reasonable method is within reach of any marketing team willing to be systematic. Here is the one we use.

1. Build a prompt set, not a keyword list

Keywords and prompts are different things. “ai search agency” is a keyword. “Which agencies should I consider for getting my brand into AI answers?” is a prompt. People type the second kind into assistants.

Aim for fifteen to twenty-five prompts, grouped by intent:

• Category discovery, “Who are the leading providers of X?”
• Comparison, “X vs Y, which is better for a mid-sized business?”
• Problem-aware, “My business isn’t showing up in ChatGPT, what do I do?”
• Brand-specific, “What does [your brand] do?” and “Is [your brand] any good?”
• Local or segment-specific, whatever qualifiers your buyers actually use

The brand-specific prompts matter more than people expect. They surface what the models currently believe about you, which is frequently wrong in ways you can correct.

2. Decide what you are recording

For each prompt and each assistant, record:

Field	Why it matters
Were you named?	The headline metric
Position in the answer	First mention carries more weight than a trailing list
Which competitors were named	Your real share of voice
How you were described	Accuracy problems live here
Were you linked or cited?	Attribution varies a lot by platform
Date and assistant version	Results drift; undated results are useless

That last row is the one teams skip and later regret.

3. Handle the variance honestly

The same prompt will not give the same answer twice. Responses vary by user, session, region, model version and sometimes apparently at random.

Three practices make this manageable:

Repeat. Run each prompt at least three times per assistant per cycle. A single run is an anecdote.

Use clean sessions. Memory and chat history contaminate results badly. Use a fresh session, logged out where the platform allows it and do not run your prompts in an account that has been researching your own brand all week.

Report as a rate, not an event. “Named in 7 of 12 runs” is information. “ChatGPT mentioned us” is not.

4. Fix the access problem first

Before concluding that you have a visibility problem, confirm you do not simply have an access problem.

AI crawlers are controlled separately from Googlebot in robots.txt. The common ones to check are GPTBot and OAI-SearchBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot and Google-Extended, which governs Google’s AI products independently of ordinary search indexing.

Blocking these is a legitimate business decision some publishers make deliberately. Blocking them accidentally, through an inherited robots.txt or a security plugin’s default, is common and entirely self-inflicted.

Note the limits of robots.txt: it controls crawling, not indexing and it is honoured voluntarily. It is a signal, not a lock.

5. Segment the referral traffic you can see

Some AI traffic is directly measurable. In GA4, build a segment for referrers including chat.openai.com, chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com.

Two cautions. First, this counts only people who clicked through, which is a minority of the people who received an answer about your category. Second, referrer data from these platforms is inconsistent and some traffic will be misattributed as direct. Treat it as a floor, not a measurement.

6. Set a realistic cadence

Monthly is right for most businesses. Weekly produces noise that will tempt you into reacting to variance. Quarterly is too slow to catch a factual error about your company propagating through several models.

Read the trend over three cycles before drawing conclusions about whether a change worked.

What good looks like after a quarter

You should be able to answer, with evidence:

• Which of our buying questions do we appear in and at what rate?
• Who consistently beats us and what do their pages do differently?
• Is what the models say about us actually correct?
• Has any of it moved since we started?

That last question is the one that justifies the budget and it is only answerable if you wrote down the baseline before you started changing things.

---

This is the method behind our AI visibility audit. The prompt set goes to you with the report so you can re-run it yourself.

Platform guides: How to get your business recommended by ChatGPT, Perplexity SEO and How to show up in Google AI Overviews and AI Mode.