For three of the four questions we ask about Thailand’s dive market, AI names no business at all.
A visibility score answers one question: was this business named, with a source. It cannot say what else was in the answer. So we read the answers. For every answer the engines gave to our four Thailand phrasings between 4 September and 14 September, a second pass listed the businesses it named, in order, and discarded any name that did not occur in the text. The method is on the methodology page; on the stored answers it recovered every mention the engines had backed with a citation.
This is Perplexity, ChatGPT and Claude. Google’s own answer above its search results is not measured here yet.
The numbers
115 answers, 89 with no business in them — 77%. By engine: Perplexity 30 of 40, Claude 30 of 35, ChatGPT 29 of 40. By question:
| Question | Named nobody | Businesses named, median / max |
|---|---|---|
| “best dive centres in Thailand” | 6 of 29 (21%) | 3 / 12 |
| “where to learn to dive in Thailand” | 26 of 29 (90%) | 0 / 6 |
| “diving in Koh Tao where to go” | 29 of 29 (100%) | 0 / 0 |
| “diving in Koh Samui where to go” | 28 of 28 (100%) | 0 / 0 |
“best dive centres in Thailand” gets dive centres: a median of 3 businesses an answer, up to 12, and Perplexity and ChatGPT named at least one in every answer. The other three questions get places. “where to learn to dive in Thailand” is answered with islands and course names; “diving in Koh Tao where to go” with dive sites. In 29 of 29 answers to the Koh Tao question, and 28 of 28 to the Koh Samui one, no operator appears.
What this does and does not mean
Two explanations fit the same numbers, and we cannot separate them. One: the engines do not surface Thai dive operators unless asked for them by category. Two: three of our four questions are not questions about a business — a person asking where to learn to dive is asking about a place and about how courses work, and an answer about Koh Tao is a correct answer. Both are probably true in some share, and this measurement does not say which share.
What it does say: for three of the four phrasings, a business scoring zero is not losing to a competitor. Nobody is in that answer. That changes what a zero means, and it is why the scan now shows how many businesses an answer named beside whether it named you.
One market, one category, 10 days, four phrasings that are ours. The eleven operators measured are a hand-picked cohort, not a sample of Thailand. Claude’s answers are capped at 1,024 tokens by our own call; its count is a count of what fit. The reading is by a model and checked against the text: a count of names in prose, not of links.
How everything above is counted — methodology.