Query fan-out
You ask one thing. Before it answers, AI search quietly rewrites your question — and fans it out into several different searches, often bolting on a competitor, a year and a commercial framing you never typed. Those searches, not your words, decide which brands get found. Here's that hidden step, measured.
From one question to many searches
Pick a real question someone asked. See the searches AI fanned it out into — green is everything the model added that you never wrote.
How far it moves from your words
Every analysed search, by how much of your original prompt survived. Most land in the rewrite half of the scale.
What AI bolts onto your search
The words AI injects most often: the language of a shopping list, not your question.
It picks a format
The commercial framings AI reaches for.
A plain question becomes a "best", a "top list", a "vs". To get cited, you have to win the listicle AI went looking for.
Brands AI brings to the search uninvited
Searches that name a brand the user never did.
Almost always the current or next one — a freshness reflex that buries anything undated.
Your 8.7 words become 13.6. 63% of searches come out longer.
You phrase it as a question 27% of the time. AI almost never searches one — it strips the question into keywords.
A measured ask gets sharpened into a superlative — and the search hunts for a ranked winner.
AI guesses who you are and adds it to the search — narrowing the field to brands that court that segment.
Ask in another language and AI often searches in English — pulling answers from a market that isn't yours.
You're not being ranked on the question your customer asked. You're being ranked on the searches AI invents from it.
One prompt becomes a handful of rewritten, brand-injected, year-stamped "best-of" searches. To show up, your content has to win those — the listicles, the comparisons, the searches that name your competitors. That's the work of getting found in AI search, and it's exactly what Trakkr measures and fixes.
Built from 37K paired observations of a user prompt and a search query the AI actually issued, diffed word-by-word for additions, removals, injected brands, year qualifiers and language. "Added" words are tokens in the search that never appeared in the prompt. Rewrite distance uses Jaccard overlap. Refreshed daily.