Use it to find objections, confusion, trust risk, and rewrite moves before real spend.
How the panel is built, how reactions are generated, what the scores mean, and where it fits.
Use it to find objections, confusion, trust risk, and rewrite moves before real spend.
Built from the shipping roster, with coverage stats computed at build time.
Individual reactions, clusters, segment splits, and ranked edits.
Computed at build time from src/personas/_compiled/americans.json — the same file the run engine loads.
Coldread takes any written asset (an ad, landing page, email, speech, debate answer, public statement, agent output) and runs it past a focus group of specifically-defined people. Each "person" is a written persona — name, age, occupation, region, voice, biographical detail — that the model speaks as. You get back individual reactions, scores, recurring objections, and ranked rewrite suggestions in minutes.
The architecture is genuinely multi-agent: parallel calls with private context, a blind synthesis layer, and graceful degradation when a batch fails.
Your ad, email, speech, statement, or URL becomes the stimulus. Optionally grounded with voter data and source notes.
Each panel member is conditioned only on their own persona file — name, age, occupation, region, voice, biography.
One independent agent call per person. No cross-talk, no shared verdict, no consensus pressure. Disagreement is emergent.
A separate judge pass aggregates the reactions without ever seeing persona demographics — so it can't stereotype the read.
Topline score, sentiment split, objection clusters, segment splits, trust and confusion risk, ranked rewrites.
Audience-mode runs work the same way but generate archetype agents from your audience description, source context, and the simulated population size you enter, then sample variant personas inside each archetype. The hosted website generates a 240-respondent sample (40 archetypes × 6 variants) to represent that population; CLI and MCP callers can request larger direct runs within server caps. Hybrid runs read the named panel and the generated audience together.
Across trades, healthcare, public service, professional work, retail, agriculture, military, education, and creative work. Written one at a time, not generated in bulk — 509 distinct occupations across the roster.
Specific, not archetypal — not "suburban mom" but Beth Howell, 36, suburban Indianapolis, ex-marketing manager who sees the manipulation in your copy and dismisses it in two seconds.
Pulled straight from the persona files the engine runs. Click anyone to read the opening of their file; shuffle for a different cut of the roster.
coldread panel show americansA public focus group baseline on AI and deepfakes — 3 human groups, 39 participants — scored against a fixed 10-category insight rubric, 0–2 points per category.
Coldread recovered trust erosion, scams and impersonation, political misinformation, verification behavior, accessibility benefits, consent risk, bias, and audience relevance gaps. The two partial misses matter: the human groups gave more texture to the "liar's dividend" problem (real evidence later dismissed as fake), and surfaced institutional disruption themes more strongly.
That is why we do not claim AI replaces human research. The claim is narrower: a fast first read on what people will understand, reject, trust, ignore, or question — before you spend weeks recruiting a panel.
The research-grade detail, collapsed so the page stays readable.
Each panel member is described by a multi-tier profile. The ordering reflects empirical priority: a narrative biography predicts how a person responds far better than any structured demographic field (Park et al., Generative Agent Simulations of 1,000 People, 2024 — interview-conditioned agents replicated their source individuals at 85% of the source's own two-week test–retest reliability; demographic-only agents reached 71%).
The Americans baseline was hand-authored before this schema landed; it uses Tier 2 narrative + a subset of Tier 1/5 fields and is being progressively backfilled. Auto-generated panels ship with the full schema.
These are mitigations, not eliminations. Validity work (held-out human benchmarks, GSS / ANES / WVS calibration) is ongoing.
Each panel member gives the asset a 1–10 score and a short reaction in their voice. The headline Score (0–10) reflects how good the message is: it is derived from the share of the audience won over — the people moved enough to act — on a calibrated curve where winning a majority of a cold audience reads as genuinely strong, not merely average. The raw "won over" percentage is reported alongside it. The aggregate also reports the average individual reaction, sentiment buckets, consensus, polarization, confusion risk, trust risk, persuasion lift, backlash risk, message discipline, opposition vulnerability, factual exposure, top and bottom responders, recurring objections, and ranked rewrites. Use them as a directional read — the value is in the named reactions and recurring objections, not the single number at the top.
This is not literal market research. The panel members are written personas, not recruited respondents. Scores are a structured pre-flight signal — directional, fast, and cheap — not a forecast of conversion lift or population truth.
It is not a substitute for real voter contact, user testing, or paid pilots. Use Coldread as the read you do before spending on those — to kill the obviously-bad version and walk into the expensive test with a stronger draft.
For a candidate under compliance scrutiny, defensibility is the product. A Coldread number is built to survive a campaign lawyer, a hostile reporter, or a skeptical data team: it is impossible to misread as a forecast, it is reproducible from a stamped fingerprint, and — when the parity measurement lands — demonstrably even-handed across parties.
Coldread is a directional, bipartisan preflight read on how a message is likely to land and where it breaks. It is NOT a poll, NOT a prediction, and does NOT forecast vote share, turnout, or election outcomes. Treat scores as directional signal; use the language, objections, and segment splits operationally. Every reaction is generated by a language model, not a real person.
The same line is single-sourced and rendered on every surface a number can leave on — the on-screen report, the PDF (a standing disclaimer block plus a footer on every page), the CSV / JSON / Markdown exports, the CLI, and the MCP response envelope. No surface can emit a naked point estimate.
Each run is stamped with a config fingerprint — provider, model id, date, panel composition, and archetype × variant counts hashed to a short, stable tag. A staffer can say “this is the read from this run, on this model, at this config; here's the fingerprint; re-run it and you'll get the same directional answer within the noise floor.” If the model id changes, the fingerprint changes — silent model drift becomes visible on the report instead of invisibly stale.
Bipartisan-parity badge: measurement pending (bipartisan-fairness-moonshot). We publish this only once it is measured on matched, human-validated R-vs-D pairs — a parity claim that doesn't survive scrutiny is worse than none. Built bipartisan from the substrate up, Coldread can make a symmetric parity claim that a tool validated on mostly one side's messaging structurally cannot.
The real voter data baseline is hand-written, not scraped or synthesized on the fly, and Census-matched on the marginals shown above. Political personas carry explicit, partisan-by-design voting history — the panel is built to be even-handed, not accidentally one-sided. They are not real people, and we never claim they are; they are a structured, documented instrument whose composition you can inspect in the roster above.
A growing patchwork of states regulates AI and synthetic media in political communications. We maintain a versioned, dated map of those requirements — 5 states currently mapped, registry 2026.06.0, last reviewed 2026-06-20 — and generate a starting-draft disclosure line you can adapt. Example draft:
This ad script was created or substantially generated using artificial intelligence. (Drafted toward California CA AB 2839 / AB 2655 (2024). Confirm exact wording and placement with counsel.)
Maintained map of state AI-disclosure requirements for political communications. Not legal advice. Verify with counsel before relying on any entry; the law changes frequently.
Draft attack lines and not-yet-public messaging are among the most sensitive material a campaign holds. Runs are dispatched server-side with a Coldread-managed key; we don't write your asset or reactions to disk. Recent runs live in your browser's localStorage, on your device — not ours. Full disclosure at /privacy.
Early Access $19/mo, Pro $29/mo, or Max $120/mo — every tier includes the real voter data baseline, hosted generated-audience reads for any simulated population size, follow-ups, MCP/CLI, and exports. Teams (up to 5 people) are included with Pro and higher — every member runs on their own plan. Enterprise Teams (more than 5 people, central billing): hello@peachstateai.com. Billing via Stripe.