Answer position

Answer position is the place a brand occupies in the order an AI answer lists options — first, third, sixth. It sits between mention rate and first recommendation rate, and shows how far a brand has advanced after simply being mentioned.

Mention rate is binary for each answer: the brand appears or it doesn’t. First recommendation rate only cares whether the brand is in the top spot. If a brand moves from sixth to third, neither metric changes, yet the shift is real — buyers scanning a list pay more attention to the top few entries. Answer position fills that gap.

It is roughly the AI-answer equivalent of average rank in traditional search, with one big difference: there is no fixed results page. The list is generated fresh each time, and its length and format vary.

How to record it

Count in order of first appearance in the answer:

• Numbered list: use the item number.
• Bulleted list or running prose: use the top-to-bottom order in which brands appear.
• Multiple mentions in one answer: count only the first.

Record the position alongside the raw answer text so anyone can check the count later.

The tricky cases

• Ties. Many answers list options with wording like “any of the following would work.” Layout order does not necessarily reflect recommendation strength. Your method needs to state whether you record layout order anyway or mark these as tied.
• Non-list answers. An answer that explains how to choose but names no brands should be recorded as “no position,” not as last place.
• Same-name mistakes. If the brand in that position is actually a different company with the same name, it does not count. See name confusion rate.

Write these rules down before the first measurement and apply them unchanged in every round. A position series is only comparable if every round was counted the same way.

How to use it

• Use the median, not the mean. Report answer position as the median across a set of questions, so a few extreme values don’t dominate. If a brand is third in most answers but twelfth in one long list, the mean drifts; the median doesn’t.
• Report coverage next to it. A median position is computed only over answers where the brand appears, so always show it next to mention rate. A brand with a median position of 2 in 10% of answers is in a very different place from one with a median of 4 in 80%.
• Tie it to one question set. Position is bound to a specific question set. Change the questions and you lose comparability with earlier rounds (see baseline retest).
• Split by question type. Category keyword questions and decision queries behave differently; position matters most on decision queries, where the top one or two options get most of the attention.

What moves it

Position tends to respond to how clearly the model can tell why you belong near the top: a specific, differentiated description of what you do and for whom, consistent across your own pages and third-party evidence. Being mentioned is mostly about presence; moving up is mostly about differentiation. See generative engine optimization (GEO) for the broader method.