Does your brand show up when AI answers?

It depends on how the question is worded and which part of the response you are looking at. The first recommendations — the two or three names the model cites first — tend to be reasonably stable when the question is specific. The long tail — the secondary mentions, the nuances, the alternatives added at the end — varies more from query to query. A very specific question returns more consistent results than an open one. To draw real conclusions you need to repeat, compare and tell apart what holds from what changes each time — and that takes method and experience.

In a logged-in session, ChatGPT can use your conversation history, saved memories and custom instructions to tailor the response to you. If you have spent months asking ChatGPT about your sector or your own company, the answer it gives you will be biased by that context and will not represent what a prospective customer who does not know you would see. In incognito mode, that bias goes away. For an honest self-check, always use incognito. That said, no mode removes the bias of the model itself: each model has its own training corpus and its own source preferences, and that operates in any session.

Most current models cite sources in most prompts, but not always and not all of them in the same way. Perplexity displays them in a structured, visible format by default. ChatGPT and Gemini activate web search depending on the type of question: a very current or very informational query triggers it; a conversational one may not. When you do not see sources, it usually means the model answered from its trained knowledge without going to the web — which is also useful information, because it means that response comes from the weight the sources had during training, not from what is being published this week.

The first step is to understand why the AI is citing them. It is not a trivial analysis: it requires looking at which sources the model is using, what signals it is reading in the competitor’s content, and which of those signals are missing or wrong in yours. From there, the intervention is decided — and it is rarely the same from one case to the next. It is diagnostic work before it is execution work, and everything else depends on that diagnosis. That is exactly what the AI Visibility Check covers.

The self-check gives you a snapshot from five prompts. The Check is the baseline of ongoing work: it tests prompts from different angles in different models, analyses the dynamics that drive whether companies and products appear in your sector, identifies opportunities where your business could be attractive to AI even if it does not appear today, and delivers prioritised actions. We do not sell standalone checks: the Check opens a monthly improvement plan, because GEO is not a one-off audit, it is sustained work.

It depends on two things: where the source of the problem is, and how much weight the sources reinforcing it carry. When the problem is on your own website, the timelines are short. When it is external — old mentions, directories, sector listings — the timelines are measured in months, not weeks, and the strategy is different. The hard part is not identifying the problem: it is deciding what to address first and with what arguments to make the model change its recommendation. That is plan work, not checklist work.

Yes, within reason. We publish cases when there is a story with documentable evidence and it adds something to the broader conversation in the sector — we do not turn every diagnostic into an article. If you would like us to look at a specific sector, write to us with which one and we will tell you whether we can take it on, and on what timeline.