Sightline

Problem

A growing share of purchase research now happens inside assistants rather than search engines. There is no rank to check, no impressions to read, and no way for a merchant to know whether ChatGPT recommends them or a competitor.

Approach

1. 01

   Build a prompt set per merchant that reflects how their buyers actually ask — category, comparison, and problem-first phrasings, not brand-name lookups.

2. 02

   Query each engine on a schedule from clean sessions, so results reflect the model rather than an accumulated conversation.

3. 03

   Score two things separately: whether the store surfaced at all, and what was said about it when it did. Presence and sentiment fail independently.

What it taught me

Answers vary enough run to run that a single query says almost nothing. The product only became useful once it reported trends across repeated runs instead of point-in-time results.