Citino teamAI answers are not the same for everyone. Two people asking ChatGPT, Google AI Overviews or Perplexity the same question can see different brands named, for three reasons: personalisation, context and chance. For anyone tracking a brand, the practical consequence is simple: what you see in your own account is not what your buyers see.
Understanding the three reasons is the basis of measuring it properly, and so of AI search analytics.
Reason 1: Personalisation
Assistants increasingly remember things about the person asking.
- Memory of earlier conversations. Anthropic describes Claude’s chat search and memory, which let it search previous conversations and build on earlier context. Other assistants offer similar features.
- Saved instructions and preferences. Claude’s personalisation features include profile instructions that apply to every conversation.
- Account activity. Google says its Web & App Activity setting saves searches and activity, and that people may get more personalised experiences from it.
If you have spent weeks asking an assistant about your own company, it may bring that context into a buyer-style question. Your colleague, or a real buyer, starts without it.
Reason 2: Context
Even without personalisation, the situation of the question changes the answer.
| Context | How it changes the answer |
|---|---|
| Location | Google says location can make search results more locally relevant, using signals such as IP address. Engines that search the web inherit this |
| Language and market | The same question in another language or country can draw on different sources |
| The conversation so far | Earlier messages in the same chat shape the next answer |
| Settings and model | Perplexity, for example, lets you leave the model on Best or choose a specific model, which can change the answer |
| Wording | Language models are sensitive to small changes in a prompt (Sclar et al.), so a reworded question is a different question |
Reason 3: Chance
The biggest source of variation is often neither. Ask the same assistant the same question twice, from the same clean session, and the answer can name different brands or put them in a different order. Language models generate text with some randomness, and assistants that search the web may retrieve different pages each time.
This is why one answer proves little, and why a visibility number needs a margin of error. How AI search engines choose brands explains where the variation comes from.
What it means for tracking your brand
| Mistake | Better practice |
|---|---|
| Checking from your own logged-in account | Use a clean session with no memory, instructions or history |
| Checking from your office location | Ask from the market your buyers are in |
| Asking a follow-up in the same chat | Start a new chat for every question |
| Rewording the question each time | Keep the wording fixed |
| Judging from one answer | Ask many times and report the share of answers that name you |
The goal is to measure what a new buyer would most likely see, not what you see. How to track your brand’s visibility in ChatGPT sets out the routine step by step.
Does personalisation make tracking pointless?
No. Personalisation changes some answers for some people, but the pages engines read and the brands those pages recommend are the same for everyone. If the sources for your buyers’ questions do not mention you, a clean-session answer will not name you, and neither will most personalised ones. Tracking from a clean session measures the part you can influence: what the shared sources say.
A clean-session answer is the closest thing to what a new buyer sees. Your own account is the furthest.
Where Citino helps
Citino is an AI search analytics tool. It asks ChatGPT, Google AI Overviews and Perplexity your buyers’ questions every day, with Gemini, Google AI Mode and Claude as add-ons, from a clean session with no account history or personalisation and with fixed wording, and reports each number with a margin of error, as the methodology sets out. See how Citino measures.