What it is like to use Trueness every week

Owner: Claude. Status: first draft, 25 October 2026. Why this document exists: the calibration engine and the Calibration Record are what make Trueness defensible, but they are not why a marketing team opens it on Monday. The team opens it because in one minute it can put a piece of work in front of twelve people who behave like its audience, hear them disagree, interview one of them, get a verdict, get that verdict challenged, and send a PDF to the room. The effort of building and calibrating the audiences is paid back every day in that usefulness. This document describes it, drawing on what the reference build's team actually did in its first fortnight.

1. The six things you can do, and what they feel like

Action	What you do	What comes back	What the calibration adds
Panel	Paste a concept or drop a file: an ad, an email, a web page, a deck, a key visual, a brochure spread, a social post. Say what kind of work it is. Pick the audience, or all of them.	Every person reacts in their own voice, first person, with an appeal score out of ten and the reason. Twelve reactions in under a minute. You see where they split by market, age, life stage.	Each audience carries its record: on this kind of stimulus, in this market, the panel's scores sat this far from real people, with this interval. The screen shows it beside the scores.
Verdict	One click on a finished panel.	A chair reads all the reactions and the research behind them and gives a verdict the team can act on: does this land, with whom, the one change, where the panel split and why, and how far to trust it.	The chair's "how far to trust it" is no longer the model's opinion. It is the record's measured scope of use for this audience and task.
Challenge	One click on a verdict.	A second reader argues against the chair: reactions it ignored, research it misread, a split it smoothed over, a confidence it inflated. Then a reconciled reading.	The challenger has the human holdout findings to cite as well as the research.
Talk	Open a one-to-one with any person on the panel. Ask anything, in plain language, for as long as you like.	A conversation with someone who stays in character, knows their own life, and does not know marketing words. Ask why they skipped the email. Ask what a single word in the brand's vocabulary does to them. Ask what they would tell a friend.	Every answer is labelled generated, and the person's record states which of their traits were tested against real people and which are illustrative.
Ask the research	Type a question to the evidence bench behind an audience: the segmentation, the qualitative work, the trackers, the ethnography, the reviews, everything the brand has loaded.	A cited answer: document, page or slide, every time. Or "the documents do not cover this", which is the answer that saves a bad decision.	The bench is the grounding corpus of the audiences, so the answer and the panel agree with each other.
Export	One click on any panel, verdict, challenge or conversation.	A one-page branded PDF the team passes around: what the panel saw, who said what, the verdict, the challenge, and the honesty line at the bottom.	The honesty line is now specific: this audience, this record, this scope of use, this expiry.

Also: Compare two panels on versions of the same work; Human follow-up, which turns "we should test this with real people" into a quote and a date; and History, so the decision trail is there when the board asks.

2. A week in the life, from the reference build's first fortnight

What a marketing team actually did with a panel of twelve synthetic people in the two weeks after it was switched on, anonymised:

Monday. Dropped the proposed homepage hero image, two people checking into a hotel. The core audience tore it apart: too young, why would I check into that hotel. The image was replaced.
Tuesday. Ran all seven emails of the welcome journey as one stimulus. Verdict: lands slowly with the core audience through emails two and four, fails with both conquest audiences, who read it as a bucket-list series for their parents and said they would unsubscribe. Several used the same phrase, which the team spotted as the model looping. Both findings went into the brief.
Wednesday. Interviewed people one to one about a single word in the brand's vocabulary. A recurring reaction: the word dehumanises me and removes my agency. The research lead knew which study that came from. The word stayed, with the cost of carrying it now on the record.
Thursday. Tested naming options for a front-line role and two images that used them. Both were rejected. The team learned something about literalness that fifteen research decks had not made usable.
Friday. A channel lead asked whether the panel could judge new channels, audio, short-form video, community forums, and map the journey from ad to site. The answer was: the panel can react to the creative; the journey impact is a measurement question for real people. That is the right answer, and the platform should give it unprompted.
Across the fortnight. The core persona was described as impossible to please, scoring about four out of ten unless the work matched the research exactly. The team's instinct was to tune her to be "more human". The calibrated version does not need instinct: the record says how her scores sit against real people, and the engine adjusts to that.

The lesson the owner drew: the usefulness was immediate and nobody planned the uses. People reached for the panel the way they reach for a colleague. That is the product.

3. What makes it useful rather than merely fast

Any stimulus. Not a questionnaire. The actual email, the actual page, the actual deck, read as itself.
People, not a score. Twelve reactions you can read, in voices you recognise from the research, that disagree with each other. The split is the insight more often than the mean.
A verdict you can act on, and someone arguing with it. The chair and the challenger are the two colleagues every marketing team wishes it had in the room.
A conversation. You cannot interview a tracker.
The research answers back. The bench turns fifteen decks into something you can question, with citations, and that says no.
A page to pass around. The PDF with the honesty line was what circulated, and what made the team trust it.
And now, an answer to "how do you know". The record beside every result. This is what the reference build lacked and what Trueness adds without taking any of the above away.

4. What the platform must not do to this experience

Must not slow it down. A panel in under a minute, a verdict in under two, a conversation that answers at reading speed. The calibration runs quarterly in the background; it never sits between the user and the panel.
Must not bury the people under the statistics. The reactions come first; the record is a line beside them and a page behind them.
Must not fake confidence. "No clear winner", "not supported for this use" and "ask real people" are normal outcomes, shown plainly.
Must not let a conversation become a validated score, and must label every generated answer as generated.
Must not need a researcher to drive it. Brand managers, copywriters, email producers and agency planners ran the reference panel on day one with no training.

5. Who uses what, in practice

Person	Daily use
Copywriter, email producer, content lead	Panel and verdict on every piece before it goes to review; talk to a person when a line is not landing
Brand manager	Panel on campaign routes and naming; ask the research before writing a brief; export to the steering group
Agency planner or strategist	Panel on creative territories before the client sees them; challenge their own verdicts; compare versions
Channel and CRM lead	Panel on journeys as a whole, not single assets; human follow-up when the question is behaviour, not reaction
Insights lead	Ask the research; read the records; decide when a question has outgrown the panel and needs a wave
Chief marketing or customer officer	The quarterly read, and the occasional five-minute panel to see for themselves

6. Measuring usefulness

Usefulness is a product metric with its own dashboard, separate from statistical performance: panels run per week per seat; distinct users; share of panels followed by an action (a change, an export, a human follow-up); conversations started; questions to the bench; time from stimulus to verdict; decisions recorded with their outcome. The reference build was used across brand, content, email and channel within a fortnight with no onboarding. The programme target is that the anchor partner's team runs more panels in month one than it ran agency tests in the previous year.