ChatGPT visibility
The engine most homeowners mean when they say they asked AI, and the one where a single check tells you the least.
Ask ChatGPT who the best heating contractor in your town is, ten times in one afternoon, from the same machine, and you can get several different sets of names. That is not a fault and it is not something an agency can fix. These systems are non-deterministic by design, which means the only honest unit of measurement here is a rate across many runs rather than a screenshot of one good answer.
Two crawlers, two different jobs
OpenAI runs more than one user agent and they do different things, which is where most robots.txt advice about ChatGPT goes wrong. GPTBot is the training crawler. OAI-SearchBot is the one associated with surfacing live results. Blocking them is not one decision, it is two, and blocking the first while expecting the second to stop is a common and expensive misunderstanding.
Read your own robots.txt before anyone sells you work here. Whichever way you decide, the file is a fact you can check in a minute rather than a claim you have to trust.
What it draws on for local questions
For a question about who to call in a specific town, the answer is assembled from material about your business that exists elsewhere: your own site, your profiles, review platforms, directories and whatever third parties have written. None of that is a ranking system you can query, and nobody outside OpenAI can tell you the weighting. What you can do is read the sources the answer cites when it cites any, and notice which of your competitors keep appearing.
How to measure it without fooling yourself
Fix your questions in advance, run each several times, log out first, and record what came back word for word with the date. Report the result as a rate. An agency that shows you one screenshot of ChatGPT naming you has shown you a coin landing heads once. Citation rate explains the arithmetic, and non-determinism explains why it has to be a rate.
Who measures this engine
Most companies on our board name ChatGPT. Far fewer say how many prompts they run or how often. The scorecards record which is which.
In more detail
- The two crawlers
- GPTBot, OAI-SearchBot and what blocking each one does, including the outcome where you are describable but not correctable.