Key·Agentic Talent Search·Query analysis

Most searches just name a category.

Across 432 distinct queries, the work is plain category and demographic lookups typed like a search box. The audience-affinity engine that sets Key apart — find creators by who follows them — is real and valuable, but only about one in eight searches reaches for it. There's room to pull more usage toward the thing that differentiates the product.

40%
Name a vertical / category
12%
Search by audience affinity
7 words
Median query length
69%
Bare-keyword phrasing

Method. Run on 432 distinct organic queries (Mar–Jun 2026). Before analysis, 212 internal demo, eval, and test traces — 28% of all queries, mostly a rehearsed Target pitch — were filtered out, then exact duplicates collapsed. Full funnel in the footer.

01

What they search for

Ten input types, ranked by share of queries. Most queries carry two or three signals at once, so shares sum past 100%. The top two — naming a category and naming a demographic — are things a basic filter could do; the value-add capabilities sit lower down.

Long tail / uncategorized 33% · 144 queries

The remainder is idiosyncratic but real: named-talent seeds ("put on Dearica Hamby, Maya Brady"), brand / show / URL seeds (creiadrinks.com, HGTV, Sun Bum, Love Island), niche descriptors ("black haircare," "salty snacks," "old Hollywood glamour," "cowboy / midwestern / heartland"), and broad browse-all requests — the surface area a fixed keyword vocabulary misses.

02

How a query looks

Short and typed like a search box, not a chatbot. Two-thirds are eight words or fewer, and bare keywords beat full sentences three to one — people expect to drop in a noun phrase and get a list back.

Query length

words per query · n=432
21%
1–3
45%
4–8
25%
9–20
5%
21–50
4%
50+

Phrasing style

how the query is worded
Bare keyword / noun-phrase298 69%
Imperative command84 19%
Natural-language question50 12%
03

Verticals & brands

Two vocabularies show up: verticals that describe the creator, and brands that describe the creator's audience. Sports, beauty, and lifestyle lead the creator side; the brand side is a long, varied tail of real campaign anchors.

Verticals & niches

mentions

Brands as audience anchor

mentions
04

Signals worth acting on

Three patterns that read as product surfaces rather than search modes — each a concrete thing to build or fix.

9

Under-delivery complaints

Customers counting results and pushing back when a set feels short or wrong. The clearest evidence pointing at recall and coverage — your top reported bug cluster — coming straight from the query log.

why did you only find 27 Why only 95 instead of all 149? This list is only 4 — can you expand to 20
Treat as → recall / coverage bug
40

Seed-based search

Users hand the agent a person, brand, show, or URL and expect "more like this" — named talent, Sun Bum, HGTV, Love Island, a drinks brand's homepage. A distinct intent from attribute filtering, and currently unsupported.

put on Dearica Hamby, Maya Brady… dude perfect and other similar creators can you pull creators for Sun Bum
Build → find-similar-to-X
16

Browse-all intent

A recurring "show me everything, I'll filter" request — users who want to explore the catalog rather than craft a query. Suggests a browse / faceted entry point alongside the search box.

find all talent all industries and all demos show me what talents I have, I'll select
Add → browse / faceted view