A growing share of product research now starts with a question to an AI assistant: which tools should I look at, which brand is best for this, what do people recommend. The answer usually names a handful of brands. There is no second page of results. Either you are in the answer or you are not.
How the answers are formed
No AI company publishes an exact recipe, but the broad picture is clear. A model draws on what it absorbed in training and, in many products, on pages it retrieves from the web at the moment you ask. Brands that are described often, consistently and by sources the model treats as credible are more likely to appear. Brands that are barely written about, or described in conflicting ways, are easier to leave out.
How to measure it
- Start from real buyer questions. Collect the questions your customers actually ask before they buy, and write prompts in their words, not yours.
- Run them across the main models. Different assistants give different answers, and your position can vary a lot between them.
- Repeat every prompt. The same question can produce a different answer each time. A single response tells you very little; rates over many runs tell you something.
- Record the details. Which brands were named, in what order, how each was described, and which sources were cited.
The numbers worth tracking
- Share of mentions: how often you appear across the prompt set, compared with competitors.
- Position: whether you are the first name offered or an afterthought.
- Accuracy and tone: whether the description of you is correct and how favourable it is.
- Source coverage: which pages the models lean on, and whether you are present on them.
What you can do about it
Begin with accuracy. If models describe your product wrongly, look at what your own site and the main third-party pages say, and fix the facts at the source. Then look at the publications and comparison pages that keep being cited in your category and work on being covered there. Write plainly and answer the questions buyers ask, because clear, specific text is easier for a model to use than marketing copy.
What not to expect
Nobody can guarantee a placement in an AI answer, and changes take time to show up, especially where they depend on a model being retrained. Models are also updated often, so a measurement is a snapshot. The useful habit is to measure on a schedule and watch the trend.