LLM Visibility: What You Can Know and What You Can't
- The knowable list is short: a read on your server, a visit that lands, a sale you can attribute, and a rough visibility rate across many runs.
- Fewer than 1 in 100 runs of the same prompt set returned the same list of brands in a 2026 SparkToro study.
- A percentage across many runs holds up better than a rank. Rank is the part that falls apart.
- Nobody can tell you what one specific person was told, or hand you a clean trend line that survives a model update.
- Kymo records what AI engines did on your own site. It does not track mentions.
You can know what happened on your own site, and you can get a rough rate across many runs of the same prompt. You cannot know your rank, your trajectory, or the sentence someone else was shown.
That split matters, because most LLM visibility products sit on the wrong side of it. They type prompts into assistants, count how often a brand appears, and sell the percentage as a score. The work is real. The output is a snapshot of a system that does not hold still. AEO vs GEO: Two Acronyms, One Shift untangles the naming, but the measurement question stays the same.
What you can know
Reads on your own server
When an AI system fetches a page, your server sees it. That is the piece of AI visibility data nobody can argue with you about.
Kymo runs a server-side receiver that classifies crawler user agents into three buckets. ai_answer covers a live fetch made for one person's question. indexing covers a crawler building a search index for later answers. training covers a bulk crawl gathering model training data. So you can see that Claude-SearchBot (Anthropic's search index crawler) pulled your pricing page, or that ClaudeBot (Anthropic's training crawler) took the blog in one pass. Bytespider (ByteDance) and CCBot (Common Crawl) land on their own rows, with the user agent retained so the classification stays auditable.
This is not about following a person across days. It is about a fetch. Cookieless identity means the visitor count is a salted server-side hash that rotates at UTC midnight, so a unique visitor is unique per day. Raw IPs are never stored, and no human visitor's user-agent string is ever kept.
Visits and sales
A read is not a visit. Several crawlers send no traffic at all. GPTBot, ClaudeBot, CCBot, Bytespider, meta-externalagent, Amazonbot and cohere-ai never arrive as a person.
Others can lead to a referral click, and that visit is countable. How to See AI Traffic in Your Analytics covers the mechanics. You can also know whether the visit converted, because goal tracking with revenue on a goal works here the same way it works for any other channel.
A rough visibility rate across many runs
This is the honest version of a visibility score. Run the same prompt set many times, count how often you appear, report it as a rate.
The 2026 SparkToro study did exactly that. 600 volunteers ran 12 prompts through ChatGPT, Claude and Google's AI surfaces 2,961 times. 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 visibility percentage across runs was steadier than the ordering. Gumshoe.ai, which sells AI tracking, worked on the study, and the write-up discloses it.
So a rate is defensible. A rank is not.
What you cannot know
A stable position
Any tool reporting a ranking position in AI answers is, in Rand Fishkin's phrase, full of baloney. There is no stable slot to occupy. The list changes between runs of the same question, so a "#2 in ChatGPT" badge describes one roll of the dice.
A trend line across a model update
Models change under you. Thinking Machines Lab ran 1,000 completions of the same prompt at temperature 0 and got 80 unique outputs, with server load changing the batch size as the cause. If the same prompt cannot return the same text twice, a month-over-month line chart is measuring the model as much as your content.
Larger shifts happen too. Chen, Zaharia and Zou measured GPT-4 scoring 84% on a prime-number task in March 2023 and 51% on the same task in June 2023. OpenAI rolled back a GPT-4o update on April 29, 2025 because the model had become overly flattering. A number you logged the week before either change describes a system that no longer exists.
What a specific person was told
You cannot know this. The answer was generated for one person at one moment, and no log of it reaches you. Screenshots from your own chats are a sample of one.
Why the numbers from different tools disagree
| Claim you will see sold | What the evidence supports |
|---|---|
| "You rank second in ChatGPT" | Nothing stable. Fewer than 1 in 1,000 runs returned the same brand order in the SparkToro study. |
| "Your AI share of voice is 20%" | It depends on the method. One brand on one dataset scored 20% on mention-based share of voice, 16.8% position-weighted and 31.4% citation-based, per Digital Applied. |
| "AI Overviews and AI Mode are the same channel" | Ahrefs compared 540,000 query pairs and found the two surfaces cited the same URLs 13.7% of the time. |
| "AI answers definitely send traffic" | Some do, at a lower rate. Pew Research Center found users clicked a result in 8% of visits to a Google page with an AI summary, against 15% without one, and clicked a link inside the summary in 1% of visits. |
None of those studies is selling you a dashboard. That is part of why they are useful. If you have already seen a visibility product and distrusted it, What Is an AI Visibility Checker and Do You Need One? walks through what those tools can and cannot do.
What Kymo records, and what it refuses to claim
Kymo samples nothing. It records what AI engines actually did on your own site: a read, a visit, a sale, per page. The AI Traffic page shows which pages AI engines read, how many people AI answers sent to each one, and which of those people completed a goal.
Pricing is $9 a month or $90 a year for up to 10,000 events a month, and $29 a month or $290 a year for up to 100,000 events a month. Every feature is included on both plans, and the plans differ only by event volume. Crawler tracking does not count against your event limit. Both plans start with 14 days free, no card required. Exceed your limit and the dashboard pauses, with your data kept.
What Kymo does not do: it does not track mentions, and it does not tell you what an assistant said about you. That boundary is the point. Prompts typed into a chat window are not your site's data.
On crawler control, publishing an llms.txt file is a cheap bet, not a ranking lever. Adoption is partial, and no major operator has committed to honouring it. What an llms.txt file is, and whether AI crawlers fetch it explains the difference between publishing the file and a crawler fetching it, which only your server can see.
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.