Trakkr Data

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.

Searched as you typed
queries left unchanged
Average word overlap
27.5%
your words kept
Completely rewritten
30.5%
no words in common
Pairs analysed
37K
real prompt → query

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.

Year injection
get a year added

Almost always the current or next one — a freshness reflex that buries anything undated.

Length growth
words longer

Your 8.7 words become 13.6. 63% of searches come out longer.

−10 wordskept+10 words
Question, dropped
stay a question

You phrase it as a question 27% of the time. AI almost never searches one — it strips the question into keywords.

Intent escalation

A measured ask gets sharpened into a superlative — and the search hunts for a ranked winner.

Audience fabrication

AI guesses who you are and adds it to the search — narrowing the field to brands that court that segment.

Language switch

Ask in another language and AI often searches in English — pulling answers from a market that isn't yours.

What this means for you

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.

Methodology

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.

Trakkr Query Translation Study·CitationsRankingsCC BY 4.0

Common questions

Does AI search use your exact words?

Almost never. AI engines rewrite your prompt before they search — across 37,352 real prompt → query pairs, exact-match searches are rare. The engine expands one question into several different searches.

What is query fan-out?

Query fan-out is what an AI engine does between your question and its answer: it rewrites and splits your prompt into multiple search queries — often injecting brand names, a year and a commercial framing — then merges the results.

Does AI add brands to your search?

Sometimes, yes. The data shows AI engines injecting brand names that were not in the original prompt, which shapes which companies end up in the answer.