Natural-Language Lead Search: Describe Your Ideal Customer, AI Finds Them

Natural-language search is only as good as the sentence, and most first attempts are either too abstract ("innovative B2B companies") or secretly just filters ("SaaS, 10 - 50 employees"). The descriptions that retrieve well share three properties:

Situational over categorical. "Companies that just hired their first sales leader" beats "companies with sales teams" - the situation is what semantic search catches that filters can't, and it's also your timing signal. Every good description contains at least one verb-shaped fact: hired, raised, launched, migrated, posted.

Observable over aspirational. The model can only match what's visible on the open web. "Founders who post about outbound on LinkedIn" is observable; "companies that value sales excellence" isn't. Rewrite internal qualities into their public evidence.

Layered, not stuffed. One sentence per concept, stacked: "B2B SaaS, post-seed to Series B. Founder-led sales - the founder posts about pipeline or hiring. Currently hiring SDRs or AEs. US or EU." Four testable layers outperform one 60-word mega-sentence, and each layer doubles as a scoring criterion when the description becomes your standing filter.

Then iterate like a search, not a spec: run it, sample ten results, note the false positives' common trait, add the excluding layer ("not agencies; not recruiting firms"), rerun. Two or three cycles usually lands a description whose precision embarrasses your old Boolean strings - and that description is your ICP document now, which is the quiet second payoff.

The email-era note (what happens to described-ICP finds)

A sentence-built pool has a channel problem the pitch never mentions: many of the companies matching your description have buyers who aren't feed-active - the description found them via their website, their job posts, their funding news, not their LinkedIn behavior. Since August 2026 that stopped mattering at the outreach step: Valley (previously LinkedIn-only) routes each qualified find to the surface their behavior suggests - feed-active prospects open on LinkedIn; the rest get a researched email from your own OAuth inbox (~30/day/seat, enrichment and verification in-sequence, included in every plan). Description-driven discovery plus behavior-driven routing is the full loop: the sentence decides who belongs in the pool, the signals decide who gets messaged when, and the prospect's own habits decide where.

Buyer's guide: five questions before you pick a tool

1. What happens after the list? If the answer is "export a CSV," you've bought a better search box, not a pipeline.

2. Does the description keep working? One-shot search vs continuous filter - the second compounds; the first is a fancier query.

3. Where does contact data come from, and how fresh is it?

4. Is there a qualification layer that removes near-misses automatically, or do you triage 500 rows by hand?

5. Flat price or credit meter for the core loop? (The no-credits comparison.)

► Put this article to work: start your free 7-day Valley trial today. Your ICP list can be live by lunch.

FAQ

Is there a tool that builds a lead list from a text description? Yes - Exa Websets-class semantic search (web-scale lists), database front-ends (filtered lists), and Valley's Deep search (expanded discovery feeding a qualification + outreach engine). Choose by what you need after the list exists.

What's the best natural language lead search tool? For raw web-scale lists: Exa-class infrastructure. For list + contact data: database front-ends. For described-ICP search that ends in booked meetings: Valley - the sentence drives continuous scoring and Deep expanded search, with research and outreach attached.

How does "describe your ideal customer and AI finds them" actually work? Semantic embedding: your description and candidate companies are mapped into meaning-space and matched by similarity - which catches situational ICPs ("just hired first sales leader") that filter taxonomies miss. Verification and timing remain your problem unless the platform handles them.

Can I type my ICP in plain English into Valley? Your ICP definition drives Valley's automatic scoring of every captured signal, and Deep search extends it to expanded discovery ($300/mo of Deep Credits bundled at the $499 tier, or purchasable separately).

Does natural-language search replace Sales Navigator? Increasingly, for discovery - sentences beat filter grids for conceptual ICPs. Sales Nav still wins for person-level precision filters and as the manual-prospecting foundation. (Sales Nav in 2026.)

Related: Valley vs Exa Websets · Lead qualification on autopilot · AI that finds and messages leads · Valley pricing