Creative testing: what to ask so the idea survives the research
Creative testing has a bad name in agencies, and deservedly: done badly it turns an idea into the average of six opinions and returns the version that offends nobody. Done well it does not ask whether people like it — liking predicts behaviour poorly — but what they understood, who they think it is talking to, and what they remember five minutes later.
Why "do you like it?" fails
Because the answer is polite, and because personal taste correlates weakly with conduct. Everyone says they prefer the understated ad and everyone remembers the loud one.
What does carry predictive weight is duller: whether it is understood without context, whether the category is clear, who the viewer thinks it addresses, and which word sticks. None of that requires a score.
Protocol by asset type
The principle is constant — first reaction before any explanation — and the questions change with what you show.
- Claim: show it alone, with no campaign around it. Ask what it says and who they think it comes from.
- Visual: upload it and ask where they would expect to see it. Channel fit falls out of that answer.
- Logo: ask for the first word that comes to mind and which sector they would place it in blind.
- Script or storyboard: tell it in three sentences and ask what they think happens at the end.
- Always: ask which brand they confuse it with. Confusion is more informative than approval.
Testing work aimed at a market you do not live in
This is where an English-speaking agency is most exposed: a route that is obvious in London can land as sarcastic, foreign or simply unreadable in Madrid or Mexico City, and the local partner who would have said so is not on the account.
Six profiles from that market, reacting in their own register, catch most of that in the first round — including the specific failure that costs the most: a piece that is understood perfectly and attributed to a different category.
Iterating rather than approving
The productive use is not validating the final artwork. It is comparing routes while they can still change: with 10 questions you can put three claims in front of two audiences and read all six combinations before lunch.
And the transcript is ammunition for the meeting — literal quotes about why a route reads worse, instead of an argument about taste between people paid to have it.
The risk of this method
That it gets used to kill uncomfortable ideas with a methodological excuse. A piece that divides the room is not necessarily failing; in creative work consensus is the suspicious result.
And the usual limit: this does not measure recall or effectiveness. That is pretesting with a sample and post-campaign measurement. Here you sharpen the idea before paying to produce it.
Frequently asked questions
Not yet — the engine reads images and text. Upload a key frame and tell the story in three sentences, which has the side benefit of testing whether the idea stands up without the production.
Keep reading
Last reviewed: 2026-09-14. Participants are AI-generated synthetic archetypes, not real people.