Assistants and answers
One request for a page by a declared automated client of an AI system — the only unit of attention in answer engines that a site can count on its own server.
Also called AI crawler request · Обращение ассистента
Three different events are routinely reported as one. An index build — OAI-SearchBot, PerplexityBot, Claude-SearchBot — makes a page eligible to be cited later. A training crawl — GPTBot, ClaudeBot, PetalBot, Amazonbot — returns no citation, no link and no visit at any point. A user-triggered fetch — ChatGPT-User, Claude-User, Perplexity-User — means a person asked a question a moment ago and is waiting. Only the third can send a reader today, and only the first can send one next month.
It is counted from the user-agent header at the edge, which means it records a declared identity rather than a verified one. Operators publish address ranges that allow verification and this site does not currently perform it, so the count is an upper bound on genuine agent traffic. Stating that is not a caveat; a number whose failure mode is unstated is not a measurement.
Nothing about a human reader is recorded by the same instrument. A request whose user agent matches no known automated client passes through unlogged, which is the property that makes counting the rest defensible.
- OAI-SearchBot · OpenAI · search index
- 33 fetches, 31 distinct pages
- PerplexityBot · search index
- 22 fetches, 20 pages
- PetalBot · Huawei · training
- 16 fetches, 16 pages
- Bingbot · search index
- 9 fetches, 6 pages
- Amazonbot · training
- 7 fetches, 7 pages
- Googlebot · search index
- 4 fetches, 1 page
- ChatGPT-User · a person waiting for an answer
- 4 fetches, 3 pages
- Applebot, ClaudeBot, Baiduspider
- 4 fetches between them
99 fetches from 10 declared agents in a day: 71 for search indexes, 24 for training, 4 triggered by a person. OpenAI’s index took 31 distinct pages; Google’s took one page four times. On this site, on this date, the answer engines paid more attention than the search engine — which is the argument for GEO stated in requests rather than in opinion. Six further rows were this instrument checking itself and are marked at the moment of observation, so they are excluded here.
A single number merges a citation that may arrive, a corpus that will never link back, and a person waiting right now. They have three different values and two of them cannot be converted into the third. Record the purpose at the moment of observation — reconstructing it later from a user-agent string is possible and never gets done.
Definitions of the unit. Nothing here is calculated, and almost every argument about a marketing number is really an argument about one of these.
Is an input to
- Citation
- Being named or linked inside an answer a machine assembled — the unit of answer-engine visibility, and the one your own server cannot see.
An error in this figure does not stay in it. It reaches two numbers above it, and each of those is quoted as though it were measured.
Knowing the definition is not the same as being able to check the figure. These are the procedures that do the second thing.
- Testing a promise about AI answers
- “We will get you cited by ChatGPT and the other assistants.” · 20 minutes, 5 questions.
The definitions are the easy part. Whether the figure on your dashboard was computed this way is a different question, and usually the more expensive one.