You measure AI search visibility by combining three data sources: a fixed list of prompts you run against ChatGPT, Perplexity, Claude, and Google's AI Overviews on a regular schedule to see whether your business gets mentioned or cited; referral traffic from AI platforms isolated in your analytics tool; and AI crawler activity in your server logs. None of these is a complete picture on its own, because AI platforms offer nothing like Google Search Console for answers. But together they tell you the three things that matter: are you in the answers, is anyone clicking through, and can the engines even read your site.
The good news is that a workable measurement system costs nothing except an hour to set up and 30 minutes a month to run. Paid tools like Profound, Peec AI, and Semrush's AI toolkit automate the prompt-tracking layer and add competitive share-of-voice data, and they make sense once AI referrals become a meaningful lead source. Start manual, prove the channel matters for your business, then upgrade.
Why Is AI Visibility So Hard to Measure?
Traditional SEO measurement rests on infrastructure that AI search simply does not provide. Google Search Console tells you every query you appeared for, your position, and your click-through rate. There is no equivalent console for ChatGPT answers. When an assistant recommends your firm to a user, no notification fires, no report updates, and unless the user clicks a citation link, no trace reaches your analytics at all.
Three properties of AI search make measurement genuinely different. First, answers are probabilistic: the same question asked twice can produce different recommendations, so a single spot check tells you almost nothing. You need repeated sampling over time. Second, answers are conversational and personalized: what ChatGPT tells a user with months of chat history differs from what it tells a fresh account, so your tests approximate reality rather than reproduce it. Third, roughly 60 percent of searches now end without a click, and AI answers push that further. A large share of your AI visibility produces zero traffic even when it works, because the user gets the answer, forms an impression of your brand, and arrives later as a branded search or a direct visit.
This is why measuring only referral traffic understates the channel badly. The visitors who do click through from AI tools are unusually qualified (they arrive pre-sold by a recommendation), but they are the visible tip of a mostly invisible surface. Your measurement system needs to see both the tip and, as well as possible, the iceberg.
The Free Measurement Stack: Three Layers
Here is the system we set up for clients before any paid tool enters the conversation. Total setup time is about an hour.
Layer 1: A prompt panel you run monthly (30 minutes per month). Write 10 to 15 prompts a real prospect would ask, in conversational language, not keyword-speak. For a Toronto bookkeeping firm: "I run a 10-person agency in Toronto and my books are a mess, who should I hire?" plus variations by service, city, and problem. Run each prompt in ChatGPT, Perplexity, and Google (checking whether an AI Overview appears), using a logged-out or fresh session where possible. Record four things in a spreadsheet: mentioned yes/no, cited as a link yes/no, position in the answer, and which competitors appeared. Date every run. Within three months you have a trend line, which is the thing that actually matters. Remember that assistants fan one conversational question out into multiple sub-queries behind the scenes, so include prompts targeting the sub-questions too: "how much does X cost", "what should I look for in a Y".
Layer 2: AI referral segmentation in GA4 (20 minutes, once). Create a custom channel group or exploration filter for sessions where the referrer contains chatgpt.com, chat.openai.com, perplexity.ai, claude.ai, copilot.microsoft.com, or gemini.google.com. Track sessions, engagement rate, and most importantly conversions. Expect small numbers; the pattern across published studies is that AI referral traffic runs a low single-digit percentage of organic but converts at multiples of it. Watching that conversion gap for your own site is what justifies (or kills) further investment.
Layer 3: Crawler visibility in your logs (10 minutes, once). Check your server logs or CDN analytics for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, and Google-Extended. If AI crawlers never visit, you have a plumbing problem, not a content problem: a robots.txt rule or a firewall setting is locking you out of the answers entirely. This check takes minutes and regularly turns up sites that blocked every AI crawler with a one-click CDN "AI protection" toggle.
Most businesses have no idea whether AI recommends them or their competitor. Get a free audit that includes your current AI search visibility.
When Do Paid AI Visibility Tools Make Sense?
Paid platforms automate Layer 1 at scale: they run hundreds of prompts daily across engines, track your share of voice against named competitors, and show which pages earn citations. Profound is the enterprise reference point in the category; Peec AI, Otterly.ai, and Semrush's AI visibility toolkit cover the mid-market at roughly 90 to a few hundred dollars a month.
The honest guidance: do not start here. A paid dashboard on top of a site that AI engines have no reason to cite measures failure with great precision. The sequence that works is fix the fundamentals, run the free stack for a quarter, and buy tooling when two conditions hold: AI referrals are producing real leads, and you are competing for visibility against specific rivals you need to track daily rather than monthly. Understanding why AI engines cite some content and ignore the rest will do more for your numbers than any subscription.
One more measurement source deserves mention: branded search volume. When AI assistants recommend you without links, some users search your name afterwards. A steady climb in branded impressions in Search Console, without a matching PR or ad push, is circumstantial evidence that invisible recommendations are happening. It is indirect, but it is free and it moves.
What to Do With the Data Once You Have It
Measurement only earns its time cost if it changes what you build. Fold the numbers into a simple monthly review, the same discipline behind the weekly AI growth brief, and act on three signals.
Signal 1: Mentioned but not cited. The engines know you exist but link a directory or a review site instead of you. That is an authority gap on your own pages: add citable specifics, proof, and clearer answer-first structure so you become the source worth linking.
Signal 2: Competitors own a prompt category. If rivals appear for "best X in city" prompts and you never do, look at where the engines are sourcing those answers (Perplexity shows its citations openly) and earn presence on those sources. Often it is two or three directories and one comparison article.
Signal 3: AI referrals convert but are tiny. That is the normal, healthy starting state. It says the channel is qualified and worth feeding with more answer-shaped content, not that it failed. Small and highly qualified is what early AI referral traffic looks like everywhere.
Rerun the panel monthly, not daily. Answer variance is high enough that daily manual checks produce noise and anxiety, while monthly runs produce trend lines and decisions.
Frequently Asked Questions
Can I see ChatGPT mentions of my business in Google Analytics?
Only partially. GA4 shows you sessions referred from chatgpt.com when a user clicks a citation link, but it cannot show mentions that never produce a click, which is most of them. That is why referral tracking has to be paired with a prompt panel you run yourself and with branded search trends. Treat GA4 as the conversion-quality layer of measurement, not the visibility layer.
How often should I check my AI search visibility?
Monthly is the right cadence for a manual prompt panel, and quarterly is acceptable for small firms. AI answers vary between runs, so frequent spot checks mostly measure randomness, while monthly runs of a fixed panel reveal genuine trend. If you use a paid tool that samples daily, review it weekly or monthly anyway; the decisions you can make (publish, restructure, earn sources) operate on that timescale.
What is a good benchmark for AI referral traffic?
Published studies and our client data agree on the shape: AI referrals typically run between one and five percent of organic search sessions, but convert at two to four times the rate of organic visitors. So benchmark on quality, not volume. If your AI-referred sessions show strong engagement and produce enquiries at a higher rate than organic, the channel is working, even while the absolute session count still looks like a rounding error.
Do AI visibility tools like Profound actually work?
They do what they claim: run large prompt panels across engines on a schedule, compute your share of voice against competitors, and surface which prompts and pages drive citations. That is real, useful data. What they cannot do is see actual user conversations, which are personalized and private, so every tool is sampling an approximation. They are worth paying for once AI search is a proven lead source for you, and premature before that.
How do I know if AI crawlers can access my site?
Check two places. First, your robots.txt file for disallow rules against GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, or Google-Extended. Second, your server logs or CDN dashboard (Cloudflare in particular) for whether those user agents appear and what status codes they receive. Blocked crawlers cannot read your content, and unreadable content cannot be cited. This five-minute check is the first thing to rule out when visibility is zero.
Read Next
- Why AI Engines Cite Some Content and Ignore the Rest
- The Weekly AI Growth Brief: What to Measure and Why
Get Your AI Visibility Baseline
The free audit includes a prompt-panel snapshot for your business: what ChatGPT, Perplexity, and AI Overviews currently say when prospects ask for what you sell, which competitors they name instead, and whether AI crawlers can even reach your site. You cannot manage a channel you have never measured. Start with the baseline.