ChatGPT is the most widely used generative AI platform in the world. The channel you can least ignore — and one whose mechanics are more complex than they appear at first glance.
How does ChatGPT work?
ChatGPT draws on two layers of knowledge that are strategically different. The first is training knowledge: what the model learned up to its cut-off date. The second is web search: ChatGPT decides for itself when a question benefits from current information and then searches automatically, citing sources with the answer — the user can also switch search on explicitly, but does not have to. For your measurement this means: the same question can be answered with or without a search step, and that difference determines what you get to see.
What has the most impact potential
- Long-term, broad web presence — our hypothesis, which nobody can verify from the outside: the more often a brand is described correctly across the web, the more likely the model knows it
- Technical accessibility for OpenAI’s crawlers — notably OAI-SearchBot, the crawler through which pages can appear in ChatGPT search results; check your robots.txt
- Off-site presence: external sources describing you, reachable for both training and retrieval
- Clear, citable formulations for the moment the search step is active
Considerations
- Without a search step, ChatGPT shows no source links — brand mention without a clickable source is the goal and the measuring point there
- With a search step, sources are visible; observability thus varies per answer, not per setting
- Factual errors are hard to correct — prevention through consistent presence is more robust than reactive correction
- How OpenAI ranks internally is not published; treat every tactic here as observation or hypothesis, not as a rule