Describe a sentence, get an audience

Here's a sentence: "SaaS founders complaining that their sales tool tripled prices."

Ten years ago, turning that sentence into an actual list of live conversations was a job description. Somebody had to translate it into boolean queries per platform, guess the subreddits, tune the exclusions, check the results daily, and re-tune forever. The sentence was easy. The compilation was the moat, and most teams never crossed it.

That translation layer is what's collapsing right now, and I don't think the implications have fully landed yet.

The sentence is the spec

What makes that sentence powerful is how much it encodes. A persona (SaaS founders). A relationship (customers of a competitor). An emotion (complaining). A trigger event (price increase). A timeliness assumption (this is happening now). Every word narrows a firehose by an order of magnitude.

Keyword systems could never hold that spec. "Price increase" as a string matches airline news and gas stations. The persona is invisible to grep entirely. The emotion, likewise. So we all learned to write degraded versions of our real intent, and then to tolerate the junk the degradation let in.

What changed is that language models can now hold the whole spec. The sentence compiles: into platform-native searches broad enough to catch odd phrasings, and into a rubric precise enough to reject a student, a vendor, and a person who explicitly wants the opposite of what you sell, even when all three matched the same keywords. When someone writes "our sales tooling costs more than our office now," no keyword connects it to your sentence. The meaning does.

Audiences stop being platform-shaped

The second thing the sentence dissolves is platform boundaries. The people matching that spec are on X today, in three subreddits tomorrow, on Bluesky since the last migration wave, and occasionally on Hacker News at 2am. Under the old model, each platform was a separate query language, a separate tool, a separate patrol. So teams picked one platform and called it strategy.

When the interface is a sentence, "where" becomes an implementation detail. The audience is defined by what they're going through, not where they typed it. That matches how buying actually works, and it's why we made ThreadSonar listen everywhere by default rather than selling platform add-ons.

What this is not

It's not magic, and the failure modes are worth naming. A vague sentence compiles into a vague audience: "people interested in productivity" deserves the noise it gets. The precision comes from the specificity you'd give a competent colleague. Triggers, constraints, exclusions: "actively comparing, not just venting," "exclude agencies," "our own customers only."

And a compiled audience needs a feedback loop, because your first sentence is never quite your real intent. The tools in this class that work let you correct them (this one's good, that one's not, here's why) and get sharper. The ones that don't are writing your boolean strings for you once, which is a nice save of an afternoon, but not the actual unlock.

The actual unlock is that "who do we need to talk to" and "go find them, live, everywhere" are becoming the same act. Ten years of martech was built around that gap. Sentences are closing it.