How do I measure my website's performance in AI-powered search engines? Our method
How we measure a website's performance in AI-powered search engines for clients: visibility, citations, crawler access and AI referral conversions, step by step.
"How do we know if any of this is working?" is the question we hear in every first meeting about AI search. The answer is not a single metric. It is a short chain of four measurements, each answering a different question, and we would not report on AI search without all four.
This is the method we use for clients, what we measure in free tools, and where we move to a platform.
The four measurements
Visibility. For the prompts that matter commercially, how often is the brand named, and where in the answer? This is the headline number, and also the noisiest, since the same prompt can return different answers from run to run.
Citations. Which pages do the engines cite, ours or third parties'? Visibility without source detail is a score with no handle on it. Citations tell us where to act.
Access. Can AI crawlers fetch our pages, and what happens when they do? A page a bot cannot read cannot be cited, whatever its quality.
Outcomes. Do visitors arrive from AI platforms, and do they convert?
Classic SEO tooling covers the second half of access and outcomes only partially. Search Console does not tell us what ChatGPT cited, and GA4 does not tell us what GPTBot fetched.
What we measure with free tools
GA4. We create an exploration or channel group that collects referral sources such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. It gives a floor for AI-referred sessions. Some AI traffic arrives without a referrer and lands in direct, and most AI answers are read without a click, so we tell clients the number understates exposure.
Search Console and Bing Webmaster Tools. Google's documentation puts AI features traffic inside the Web search type, so we cannot isolate Overviews from classic results here. We still use it for query and page trends and for indexing checks. Bing Webmaster Tools matters because Bing data feeds Copilot.
Server or CDN logs. Good for a one-off look at AI crawler activity. Tedious to maintain, and user agents can be faked.
Manual prompt checks. Fifteen prompts, run by hand monthly across the main engines. Slow, but an honest sanity check on any dashboard.
These are enough for a first read. They stop being enough when a client wants trend lines, competitors, source detail or proof that the site is readable.
What we run client programs on
We run client work on Promptwatch, mainly because it measures all four layers in one place.
Visibility. Visibility Score, Share of Voice, Sentiment and Position in Answer, with prompt panels run daily per model, country and language. We use A/B period comparison for monthly reviews.
Citations. Page and domain level reports, Page Tracker for specific URLs (ours and competitors'), offsite mentions, and Reddit, YouTube and LinkedIn citation reports. Citation Trends classify citations by content type and source type and show first cited, peak rate and decay half-life per page.
Access. Agent Analytics logs AI crawlers in real time and verifies them by IP, so spoofed user agents are filtered out. We see which pages bots fetch, which get cited, and error explanations. Crawler logs start on Professional ($245 a month).
Outcomes. Visitor analytics attribute AI-referred traffic and conversions through a lightweight script, GTM template or the WordPress plugin. This replaces the hand-built GA4 filters for most clients, and shows conversions by AI source.
Joining the layers. 360 Insights joins crawler logs, citation clicks and AI-referred visits per page. Search Insights overlays Google Search Console and Bing Webmaster data on AI crawls and citations. This is where problems become visible: a page that is crawled but not cited, cited but not clicked, or clicked but not converting each calls for a different fix.
TreeHouse, which runs Pararius and Huurwoningen.nl, published a case study saying AI search became about 25% of its organic conversions after fixing crawler access and automating weekly reporting through the Promptwatch MCP. That is their result, and we cite it as an example of the method, not a benchmark.
Reporting to a client
Monthly, we report five things: the share of tracked prompts where the brand is named, the number of pages cited, crawler hits and errors on key pages, AI-referred sessions, and AI-referred conversions. We show trends, not single months. PDF and scheduled reports and a Looker Studio connector mean clients get this in the format their team already reads.
Reading the numbers
- Visibility up, conversions flat: check which pages are cited. They may not be the pages that sell.
- Citations present, no visits: many AI answers are read without a click. Do not treat this as a failure.
- Crawler errors on key pages: fix access before anything else.
- Visibility down after a site change: compare dates against the crawler logs.
- Competitor gains: look at their cited pages with Page Tracker and ask what they answer that we do not.
What not to promise
No tool can make an engine recommend a client. Visibility figures vary run to run, so we never report a single day. We tell clients clearly that AI referral numbers are a floor, that rankings in AI answers are not stable the way classic positions are, and that the real goal is a repeatable loop: measure, find the gap, fix it, measure again.
Cost
Explore is free (10 prompts, ChatGPT only), Essential is $95 a month and includes visitor analytics, Professional is $245 with crawler logs, Business is $579, and agency plans are $199, $399 and $799. Annual billing is 12 months for the price of 10. For a first look we suggest the free AI Visibility Report, then Explore or Essential once there is a prompt list worth tracking.