How do I measure AI search visibility without a paid tool?

You measure AI search visibility with a fixed panel of buyer prompts, run by hand or by script through ChatGPT, Perplexity, Gemini and Google AI Overviews, logging two things per prompt and engine: whether a URL on your domain was shown as a source, and whether your brand was named in the answer. Pair the panel with the free AI Performance report in Bing Webmaster Tools, which is the only first-party per-page citation count any engine publishes. I ran the first 60-prompt panel for my own site on 16 October 2026. It took 98 minutes for three of the four engines and returned zero citations and zero mentions. Every later run is scored against that zero.

By Ronan Liedmeier, fractional head of growth. Last updated 17 October 2026.

What does AI search visibility actually measure?

AI search visibility is the share of relevant buyer prompts for which an AI engine cites your site or names your brand in its answer. It is a rate over a fixed set of prompts, and it has to be split by engine because the four engines pull from different indexes and show sources differently.

Two definitions, because they are different events:

• A citation is a URL on your domain displayed as a source for the answer. It means the engine retrieved your page and used it.
• A mention is your company, product or founder named in the answer text, with or without a link. It means the engine knows the entity, in most cases from a third-party page.

You can be mentioned without being cited (the engine read about you on a review site) and cited without being mentioned (your page supplied one fact and the answer never named you). Both matter commercially, for different reasons, so the sheet needs both columns.

A rank tracker does not measure either. Ahrefs' March 2026 study of 4 million AI Overview citations found only 37.9% of cited URLs also ranked in Google's top 10 for the query (Ahrefs). A position report tells you about the other 62% nothing at all.

Which free sources cover which engines?

No single free source covers all four engines, which is why the panel exists. Here is what each one sees as of October 2026.

Bing's report went into public preview on 10 February 2026 and reports per-page citation counts, which Bing says reflect how often a page is cited rather than its importance or ranking (Bing Webmaster Blog). My baseline on 15 October 2026, the day I verified the site: 0 citations and 0 cited pages over the previous three months.

Google's documentation says sites shown in AI Overviews are counted in Search Console under the Web search type, with no separate view (Google Search Central). The number to watch is the ratio: impressions holding while clicks fall on a query set.

Everything else comes from the panel.

How do I build the prompt panel?

Write the prompts the way a buyer types them, with the company size, the country and the constraint, group them by the topics you want to be cited on, and freeze the list for a quarter. The fan-out queries an engine generates depend on those details, so "best growth agency" and "fractional head of growth for a seed-stage SaaS in the Netherlands" retrieve different pages.

My panel has 60 prompts, 12 for each of five topics the site publishes on. A sample from each:

Three rules I set before the first run and will not change until January:

1. Twelve prompts per topic, so a topic score moves in steps of about 8 points per citation and a single lucky hit is visible as what it is.
2. Mix informational prompts ("what is GEO") with commercial ones ("best GEO consultants for B2B SaaS"). Commercial prompts show who gets cited instead of you. Informational prompts are where your articles can win first.
3. US only for now. Perplexity and Google answers change with the signed-in country, and I run from the US until December.

The sheet has six columns per engine: prompt id, our citation (yes or no), named (yes or no), cited domains, competing domain that took the slot, and a note. The cited-domains column is the one that pays off after four weeks, because it tells you who owns each sub-question.

How do I run the panel by hand without getting blocked?

Run each prompt in a fresh thread, read every source the answer shows, log the domains, and space Google searches out. The engines behave differently enough that each one needs its own handling.

What my first run on 16 October 2026 taught me, engine by engine:

Perplexity is the richest source of competitor data: 6 to 25 cited domains per answer, all with URLs. If you only have time for one engine, run that one.

Google is the fragile one. Thirty-five searches in three minutes was enough to trigger the captcha, and I stopped there rather than solve it, because a solved captcha from an automated session is the kind of behaviour that gets an account flagged. Space Google prompts 20 to 30 seconds apart or split them across two sessions.

Gemini shows sources as named chips without URLs, and half its answers showed none. Log the names and accept that Gemini data is thinner than the other three.

Budget two hours a week for all four engines by hand, or one hour for Perplexity and Google and skip the other two until you have something to measure. Nothing in the sheet needs a tool.

How do I read the first result when the answer is zero?

Read the competing-domains column, because that is where the first quarter's content plan comes from. Zero citations on a site that published its first articles the day before is the expected result.

Across 155 prompt-engine pairs in my first run, these were the most-cited domains:

Reddit and LinkedIn together account for 68 of 155 pairs, so every article now gets a founder post on LinkedIn on the same topic, because Google AI Overviews and Perplexity both lift LinkedIn. The companies competing for my commercial prompts (fractional growth platforms, outbound agencies) are being cited for pages that state prices and compare options, which is the format my own hub articles underused.

The informational prompts told a different story: on "what is MCP explained simply" and "what is RAG" the cited domains were vendor documentation and university pages. Those are winnable with a plain definitional page, which is what my AI training site now publishes.

What targets make sense for 90 days?

Set targets against the baseline, per engine and per topic, and publish them so you cannot move them later. Mine, set on 16 October 2026 for 19 January 2027:

Eight of 48 is a 17% citation rate across four engines. For context, when I ran a separate 40-prompt test on Perplexity across 224 vendor mentions in agency, SaaS, local-service and ecommerce categories, 42% of named vendors had their own site cited. I have no data on new sites in competitive B2B categories; my guess is under 10%, so 17% in 90 days is a stretch target.

What I'd do

This is the setup I run for a client in the first week of an AI search engagement, before proposing any content.

1. Verify the site in Bing Webmaster Tools by importing from Search Console, and note the AI Performance baseline. Fifteen minutes.
2. Write 20 to 60 prompts in the buyer's words, grouped by topic, mixing informational and commercial. Freeze them for a quarter.
3. Run Perplexity first, one fresh thread per prompt, logging citation, mention and every cited domain.
4. Run Google AI Overviews with 20 to 30 seconds between searches. Stop at any captcha.
5. Run Gemini and ChatGPT if time allows; log Gemini source names rather than URLs.
6. Count the competing domains. The top ten is the content plan: which formats they are cited for, which sub-questions nobody in the category answers.
7. Set day-90 targets per topic and write them where the client can see them.
8. Repeat weekly, same prompts, same order. Change nothing in the panel until the quarter ends.

I do this for B2B SaaS and professional-services companies whose Google traffic has been falling since AI Overviews rolled out. The first step is always the panel and the crawl test, run before any proposal, so the conversation starts from what the engines can see today. For companies that want the panel run for them every week, with the competing-domain analysis and a monthly report, that is the AI Visibility Monitoring engagement.

Re-check by 17 January 2027: whether Google has added an AI filter to Search Console, whether the Bing report has left preview, whether the Google captcha threshold has changed, and the day-90 numbers against the targets above.