PlatformHow it worksPricingFull Intelligence ReportFor small businessEnterprise Contact
Platform · OpenAI

ChatGPT visibility,
bounded by what is documented

OpenAI documents four separate agents, not one, and they do different jobs: a search index crawler, a training crawler, an ads validator and a user-triggered fetcher. Confusing them is the most common mistake made about ChatGPT visibility, because blocking the training crawler and blocking the search crawler have completely different consequences. What OpenAI does not publish is how ChatGPT decides which sources to name — so OG01 measures what it actually returns about your business rather than theorising about the model.

Measure your domain free → The four agents
Documented

Four agents, four different decisions

These are OpenAI's own descriptions, summarised. The distinction that matters commercially is between the agent that feeds ChatGPT's search answers and the agent that gathers training data, because a great many sites blocked both when they meant to block one.

  • OAI-SearchBot — indexes sites for ChatGPT's search results. OpenAI states that a site opted out of this agent will not appear in ChatGPT search answers. If you want to be findable in ChatGPT, this is the one that matters.
  • GPTBot — gathers content to train generative models. Disallowing it signals that your content should not be used for model training. That is a separate commercial decision from search inclusion, and it is the one most businesses actually mean when they say they blocked OpenAI.
  • ChatGPT-User — fetches a page when a person asks ChatGPT to visit it. OpenAI states robots.txt rules may not apply here, because the request originates with a user rather than with automated crawling.
  • OAI-AdsBot — validates landing pages submitted as advertisements, and visits only pages explicitly submitted rather than crawling the open web.
The consequence people miss
Opting out of training and opting out of search are different choices with different costs. A business that blocks everything OpenAI operates has made a defensible decision about training data and, in the same line of a text file, removed itself from ChatGPT's search answers. Whether that trade is worth it is a commercial judgement for you — OG01 reports what your current configuration permits so the decision is made with the trade-off visible rather than discovered a year later.
Not documented

Why there is no factor list on this page

OpenAI publishes how its agents reach the web. It does not publish how ChatGPT selects which sources to cite, in what order, or with what weighting — and nobody outside the company can verify a list claiming otherwise. Any page offering you "the ChatGPT ranking factors" is offering a theory dressed as documentation.

What can be established from outside is narrower and considerably more useful: whether the conditions any retrieval-based system depends on are present on your domain. Can your pages be fetched and parsed? Does your business resolve to one clear entity? Are your claims corroborated anywhere other than your own site? And, most directly, what does ChatGPT actually say when asked about your business? That last one is an observation, repeatable and recordable, which is why OG01 runs prompts and reports the responses rather than estimating a score for a black box.

What OG01 measures

Conditions you control, and responses you can read

Access reality
What your current configuration permits, agent by agent, so a decision about training does not silently become a decision about search. Reported as a finding about your own settings.
Entity clarity
Whether your organisation resolves to one thing. A system that cannot tell which business you are hedges, and hedging reads as omission.
Retrievable substance
Whether the answers a buyer asks about exist in the served document, in a passage that can be lifted, rather than assembled after load.
Observed responses
Prompts run against the assistant and what came back — which businesses were named, and how yours was described when it was. A record of a moment, reported as such.

No part of this promises a citation. Responses vary between runs, and OG01 presents them as observations rather than as stable positions — because they are not stable, and a vendor charting them as though they were is selling you a graph of noise.

Sources

Where the platform facts on this page come from

Every statement above about OpenAI's agents is drawn from OpenAI's own documentation, summarised rather than reproduced. Reviewed 2026-09-08.

Vendors change documentation without notice. If something here no longer matches the source, the source wins and this page is wrong — tell us and it gets corrected.

A full AI visibility audit → LLM visibility audit → Perplexity visibility → Google AI Overviews → Full Intelligence Report — $399 →
Questions

ChatGPT visibility questions

If I block GPTBot, do I disappear from ChatGPT?
Not from ChatGPT's search answers, according to OpenAI's documentation — GPTBot is described as the training crawler, while OAI-SearchBot is the one whose opt-out means a site is not shown in ChatGPT search answers. They are separate decisions, and conflating them is the most common configuration mistake in this area.
Can OG01 get my business cited by ChatGPT?
No, and nobody can. OpenAI does not expose how sources are selected, so any promise of a citation is unverifiable in advance and unattributable afterwards. What OG01 does is measure the conditions on your side, record what ChatGPT actually returns about you, and run the same measurement again after the work.
Does robots.txt control everything OpenAI does?
No. OpenAI states that ChatGPT-User, which fetches a page because a person asked ChatGPT to visit it, may not be governed by robots.txt rules, since the request originates with a user rather than with automated crawling. Access control and user-initiated fetching are different mechanisms.
Why does the same prompt give different answers on different days?
Because these systems are generative and their retrieval changes. That variability is exactly why OG01 reports observed responses as a record of a moment rather than as a position, and why the durable part of the measurement is the conditions on your domain — those are stable, repeatable and yours to change.
Is there structured data that makes ChatGPT prefer my site?
Not that OpenAI documents. Structured data helps because it states plainly what your organisation is and what a page is about, which supports entity clarity generally — not because any vendor has published a markup that buys preference. Anyone selling you a special schema for ChatGPT is inventing it.
Measure what it says about you.

A free reading returns in roughly a minute and a half, without a card. The deeper reading carries the prompts and the responses themselves.

Get your free score → See the full audit →