Analyzing the audience
See who's actually in your audience — the defining traits, skews, and freshness — before you spend against it.
Before you spend against an audience, look at who's actually in it. Ask, and Watt reads the audience back — the traits that define these people, how they skew, how fresh the signals are:
/watt:audience who's actually in this audience?/watt:audience what do these people look like?/watt:audience profile my customer list — what do they have in common?/watt:audience how many people are in the market for solar near Nashville — and who are they?
The default read
It works from whatever you have — a fresh build, an audience from a past session, signals you name, a list you bring, or just a description of a market, no build required. The first read back has two parts:
- Your signals — each signal in your stack, by its share of the audience, and how many of your signals each person hits. Most people hit one or two — that overlap pattern is itself a finding about how your signals relate.
- Discovered — the part you couldn't have asked for: the net-new traits that define these people against the world, by lift. Lift is plain: 5.6× means these people are 5.6 times likelier than average to carry that trait. Under-represented traits are findings too — what your audience lacks tells you who they aren't.
Every read names its basis — "aggregates over a fixed sample of 5,000 of the 2.4M" — and stays aggregates only, always: never an individual person, identifier, or record. People only exist downstream, in an export you explicitly confirm.
Ask past the default
Everything else you build by asking. Demographic skews, freshness, splits, comparisons — ask the follow-up the way you'd ask an analyst, and the read is computed on the fly:
- "how does this skew demographically?"
- "split that by metro"
- "which of my signals is actually doing the work?"
- "rank my signals by freshness"
- "how does this compare to last week's audience?"
Behind every read, each signal carries seven measured metrics you can analyze, rank, and report against:
| Metric | What it measures |
|---|---|
| Relevance | How closely the signal matches the idea you described — the one metric that depends on your ask. |
| Freshness | How current the signal is — in-market signals read live, durable identity stands. |
| Rarity | How niche the signal is, as a raw score. |
| Specificity | The same niche-ness, normalized 0–1 against the whole graph. |
| Size | How many people carry the signal, as a raw count. |
| Breadth | The same size, normalized 0–1. |
| Coverage | Calibration — how closely the signal's observed footprint tracks ground truth. |
Use it to sharpen the stack
When the discovered traits surprise you — a skew you didn't intend, a signal dragging in the wrong people — go straight back into the build: "rebuild it without the RV angle", "add something about day hikes" — and run the stack again. Analyze again, and the loop converges on the audience you actually meant.
Use it to build a report
When the read needs to travel, ask for it as a report — "turn this into a report for my client", "make a one-pager my media team can act on" — and Claude builds a self-contained file from the real audience data: it opens offline in any browser, framed your way, and every figure traces back to the measured aggregates. Nothing in it identifies a person, so it's safe to send anywhere — your team's channel, a client deck, a kickoff email. For market-profile work, the report usually is the deliverable.
And when the read comes back right, the audience is ready to activate.
From a list
Bring a list you already own and Watt expands it, learns what defines it, or slices it by intent — and every play chains with the rest of the build.
Activating the audience
Export your audience in the format your destination expects — social platforms, DSPs, outbound lead lists, direct mail.