Know which message lands, and why
Rank every draft headline, claim or value prop with a thousand digital twins in minutes. Score each one on clarity, belief, relevance and difference. Then take the finalists to real people overnight and hear which words did the work.
Run it with
Methods
MaxDiff ranking
People pick the most and least compelling line from small sets, so twenty drafts come back as one ranked list with a preference share for each.
Message diagnostics
Each finalist is scored on the four standard measures: clear, believable, relevant to me, and different from what others say.
Monadic or sequential monadic
Show each person one message for a clean read you can benchmark, or several in rotated order when you want more lines per respondent.
Playback and phrase marking
People say the message back in their own words and point to the words that pulled them in or put them off, so you see misreadings and know what to cut.
TURF for claim sets
When a pack or page carries several claims, find the combination of two or three that reaches the most people instead of the same people twice.
Screen twenty drafts before lunch
Run every variant past a thousand digital twins and cut the weak ones in minutes. A typical panel round tests three or four lines and takes weeks.
Catch the line that sounds good but says nothing
A headline can win on preference and still fail on clarity or belief. Diagnostics flag it, and the AI interviewer asks people what they think it means and follows up on anything they misread.
See which audience a line loses
Twins give a realistic spread of answers, so you can see a headline that wins with new customers and falls flat with current ones. Then confirm that split with real interviews, by text or voice, in any language.
How you run it
- 1Minutes
Screen
A thousand digital twins answer first, so the weak options drop out.
- 2Overnight
Interview
The AI interviewer talks to real people and follows up on what they say.
- 3By morning
Decide
The number, the reasons behind it and the quotes that carry it.
Questions
What is message testing?
Message testing shows a target audience your draft headlines, claims or value props and measures which ones they understand, believe and prefer. Good message testing also tells you why, so you know which words to keep.
What metrics should a message test measure?
Most teams use preference plus four diagnostics: clarity, believability, relevance and differentiation. Preference tells you which line people pick. The diagnostics tell you whether it works, since a line can be liked and still misunderstood.
Should I use monadic, sequential monadic or MaxDiff?
Use MaxDiff to rank many drafts, roughly eight or more. Use monadic when you want a clean score for each finalist that you can compare across studies. Sequential monadic sits between them and needs fewer people. A common flow is MaxDiff with twins, then monadic reads with real people.
How is this different from an A/B test?
An A/B test needs live traffic and only tells you which version got more clicks. With Jean you test before launch and hear people explain what they took from each line, so you can rewrite the loser instead of just dropping it.
Can digital twins replace real respondents for message testing?
Use twins to screen many drafts fast and narrow to a short list. Then confirm the finalists with real people through the AI interviewer. Twins run on Alyosha, Jean's model of human behavior, and each is built from a real person's interview.
Can I test claims in more than one language or market?
Yes. The AI interviewer runs in any language, by text or voice, and themes come back coded in one report so you can compare markets side by side.