How to Measure AI Visibility: Mention, Read, Click, Sale
- AI visibility has four measurable steps: a mention, a read, a click, a sale. Each one is produced in a different place.
- Prompt sampling measures the mention only, and it barely repeats. Fewer than 1 in 100 runs of the same test prompts returned the same brand list.
- Pew found users clicked a link inside a Google AI summary in 1% of visits, against a 15% click rate on Google pages with no summary.
- Only your server sees the read. Crawler fetches land in your logs, sorted into the categories explained in the bot taxonomy.
- Kymo measures reads, clicks and sales on your own site. It does not track mentions or what an assistant said.
Measuring AI visibility means choosing one of four steps and reading the number where it actually gets produced. Most tools sample prompts and report a mention score. That covers step one of four, and step one is the least stable of the bunch.
Four steps, four different measurement points
| Step | What it is | How it is measured | Does the number repeat? |
|---|---|---|---|
| Mention | An assistant names your brand in an answer to someone else's question | Prompt sampling: run prompts, count the brand appearances | No. Fewer than 1 in 100 runs returned the same brand list |
| Read | An AI engine fetches a page from your site | Server-side receiver or raw logs, classified by crawler user-agent | Yes. Counts are counts, and they attach to a specific URL |
| Click | A person arrives at your site from an AI referral | Referrer data plus confirmed click-throughs from AI referrals | Yes, as a monthly total |
| Sale | Revenue booked on a visit that started in an AI answer | Goal tracking with a revenue value attached | Yes, if the goal is set up |
Read that table top to bottom and the industry's problem becomes obvious. The step with the most tooling is the step with the least repeatable number.
The mention is the step everyone sells you
Most AI visibility tools work the same way. They type prompts into ChatGPT, Claude, Perplexity or Google, count how often your brand comes back, and sell that percentage as a visibility score. Your site never sees the request. An assistant is answering a stranger's question.
That number moves. A 2026 SparkToro and Gumshoe.ai study put 12 prompts through ChatGPT, Claude and Google AI Overviews across 600 volunteers, 2,961 runs in total. Fewer than 1 in 100 runs returned the same list of brands. Fewer than 1 in 1,000 returned the same list in the same order. The study also found that a percentage-across-many-runs figure held steadier than position, which is why some vendors report the percentage. Gumshoe sells AI tracking, and the study says so.
The wobble has mechanical causes. Thinking Machines Lab ran 1,000 completions of the same prompt at temperature 0 and got 80 unique answers, because server load changes how requests get batched. The model itself also drifts under you. Work from Chen, Zaharia and Zou measured GPT-4 scoring 84% on a prime-number task in March 2023 and 51% in June 2023. OpenAI rolled back a GPT-4o update on April 29, 2025 for being overly flattering (OpenAI).
Rand Fishkin's description of tools that report a ranking position in AI is that they are "full of baloney". Treat any dashboard selling you a rank inside an AI answer with suspicion.
The read happens on your server, and you can audit it
A read is an AI engine fetching a page from your site. Your server records it. When someone asks ChatGPT a question right now, ChatGPT-User (OpenAI's live fetcher) reaches out and pulls the page mid-answer. That request hits your logs like any other GET.
Kymo's receiver sorts crawler user agents into three categories. ai_answer is that live fetch made for one person's question. indexing is a crawler building a search index for later answers, which is where PerplexityBot sits. training is a bulk crawl gathering model training data, which is ClaudeBot (Anthropic's training crawler). CCBot (Common Crawl) belongs there too, and it sends no traffic back.
This is the step that repeats. Fetch counts for a given URL on a given day are integers, not estimates. If you block training crawlers, you do not disappear from that operator's answers, because the indexing and live-fetch agents are separate tokens. Should You Block AI Training Bots but Allow Answer Bots? walks through the trade.
The click is much smaller than the mention
Sitting inside an AI answer is not the same as getting a visit. Pew Research Center studied 900 US adults in July 2025 and found that on Google pages with an AI summary, users clicked a result in 8% of visits. On pages without one, the figure was 15%. Links inside the summary itself got clicked in 1% of visits (Pew Research Center).
That is the gap between a mention and a click, in one study. Ahrefs found a related problem inside Google itself: across 540,000 query pairs, AI Mode and AI Overviews cited the same URLs only 13.7% of the time (Ahrefs). Two surfaces, one query, different citations.
Kymo records confirmed click-throughs from AI referrals. It does not follow one person across days. Unique visitors are per day, counted from a salted server-side hash that rotates at UTC midnight. No cookies, and raw IPs are never stored.
The sale is the step that pays rent
A click that books nothing is a nice chart. Attach a revenue value to a goal and the same click becomes a number you can act on. Kymo does goal tracking with revenue, plus outbound click tracking, so you can see whether a visit that began in an AI answer turned into money or bounced.
You also get UTM tracking for source, medium and campaign, which is how you separate an AI referral from a newsletter. If you are running a shop or a one-person studio, AI search for small business covers the wider picture.
One brand, three scores, same data
The scoring mess is real. Digital Applied took one brand and one dataset and scored it three ways: 20% on mention-based share of voice, 16.8% position-weighted, and 31.4% citation-based. Same underlying data. Three answers, and the largest is over 1.5 times the smallest.
When a vendor quotes you a single AI visibility score without telling you how it was built, you have learned nothing about your site.
Where Kymo draws the line
Kymo samples nothing. It records what AI engines actually did on your own site: a read, a click, a sale, per page. It does not track mentions, and it does not tell you what an assistant said about you. No tool can honestly do that from your server, because the answer happened somewhere else.
Kymo's AI Traffic page shows which of your pages AI engines read, how many people AI answers sent to each one, and which of those people completed a goal. If you want to build the picture yourself first, How to Run an AI Visibility Audit by Hand shows the manual version. The Small Business Guide to AEO covers what to change once you have the numbers.
Pricing is Solo at $9 a month or $90 a year for up to 10,000 events a month, and Studio at $29 a month or $290 a year for up to 100,000 events. Every feature is on both plans; they differ only by event volume. AI crawler tracking does not count against your event limit. If you go over, the dashboard pauses and your data stays put. Both plans start with 14 days free, no card required.
Stop guessing. Watch.
Most AI visibility tools ask a chatbot and report a guess. Kymo watches your own site instead: which pages AI engines read, how many people their answers send you, and which of those people buy. Every number comes from your server, not from a prompt someone else chose.
Start free → 14-day free trial. No card required.