Market research in Claude: personas show you before launch which version works.
Today, the sign-off meeting decides. Each persona is researched from public customer voices, with a quote and a source. You don’t need traffic, and the live test comes afterward.
Head of Marketing, SaaSBuyer
Returning to workConsumer99takes · test_page· 12 personas, 48 verdicts„…exactly what I can defend to the attorney and finance.“
Owner, ad agencyBuyer
Performance managerBuyer
Retired craftsmanVoterOne sentence in Claude is all it takes. For example:
- Test a page
“Test my homepage against my target audience and write a better version.”
You get the new version to approve first, then the verdict.
Try it free → - Compare headlines
“Which of these two headlines gets more clicks with my customers?”
You get which one leads, and for each persona the objection the other one gets stuck on.
Try it free → - Check competitors
“Which claims on my page do my competitors make the same way?”
You get each key claim marked taken, contested or open, with a verbatim quote and its source.
Try it free →
Whoever objects the loudest wins. Then it goes live, and you wait weeks for data.
You know how it goes: round three of revisions, and still nothing has been tested.
AI generated 99takes · test_page· 12 personas, 48 verdicts- 01
Not enough traffic for an A/B test?
Then test before launch, not live. A small page never reaches significance. You need the verdict before the first click.
- 02
Leading teams test before launch.
Test the message first and you start with the better version. Approve by taste and you pay in budget and weeks.
- 03
What it feels like when it works.
You know before launch which version works, and why: where the personas get stuck, get confused, or drop off.
Why not just ask Claude?
The 99takes jury runs on Claude too. The difference is who the model plays.
Ask it yourself, and the model has to guess who buys from you.
Two prompts often give you opposite verdicts.
The personas come from competitor copy and public customer voices, each with a quote and a source.
✓10 to 14 personas read the same copy, and each one also reads it with the other personas’ jobs in mind.
You see where it keeps getting stuck.
Not made up, looked up.
Each persona is researched from public customer voices, with a quote and a source. This one is from the audience we tested this page with.
Clients run my drafts through ChatGPT and come back with its feedback.
Forum post, paraphrasedIf I change one part, everything else changes too, even what I liked.
Forum post, paraphrased
AI generated Freelance copywriterSimulated persona · 2 sources “Quote and source” right at the top gives me something I could show my client, not just another AI verdict.What holds her back
My clients already show up with ChatGPT feedback. I worry this result will look like just another AI opinion to them, too.
Simulated persona, translated from German. Run on October 12, 2026: this page’s hero against the previous version.
What a result looks like: Tappla.
Tappla is our own SaaS for TV apps, sold in the US. We tested the July 2026 hero on tappla.com against a new version with 12 personas from its audience.
Screenshots of tappla.com. Same page with only the hero copy swapped, which is how it was tested. 12 personas, 48 verdicts, October 12, 2026.
Whether your readers are business buyers, consumers, or voters.
Agencies and copywriters Check your ads and landing pages before they go to the client. Write challenger versions →
Marketing teams with sign-off rounds Walk into the approval meeting with a verdict instead of taste. Check copy → 
Consumer brands and politics For readers outside procurement: consumers and voters, each as an audience of their own. Research your audience → AI-generated images of fictional people.
Four jobs, one conversation in Claude.
How often was the jury right?
The live test comes afterward. You report which version won in the market, and 99takes uses that to calculate the hit rate per jury.
A simulation isn’t a market measurement. It comes before an A/B test with real traffic and doesn’t replace one.
How the hit rate works →Connect once, then ask.
99takes runs right in Claude, where your copy gets written anyway. 350 credits for 30 days, no payment details.
- 1Create an accountAt run.99takes.com, with 350 credits for 30 days.
- 2Connect ClaudeAdd it as a connector in claude.ai or run one command in Claude Code.
- 3AskShare a landing page or paste two versions. 99takes asks for anything else it needs.
Do it yourself or have us do it.
Use it yourself
You see the price in credits before every run. Comparing two versions with 12 personas costs about 73 credits, under €9 with the 2,000-credit pack. A 1:1 run, with each persona reading with only its own job in mind, costs about 19 credits.
We do it for you
Buyer research, your audience as personas, and a report with recommendations, all done by 99takes. As a fixed-price study for €990 or a retainer from €1,500 a month, both excl. VAT.
Request a call