Find out which wording people pick
People compare two to four wording options, pick one, and say why, so you stop guessing at your own headline.Paste your URL and Fiuto builds a free copy test for you in minutes. No experience required.
A preview player for the copy test. It steps through the four blocks a respondent sees, and is paused on the third: the wording options they compare, with the Fiuto panel floating over the page holding the question.
What is a copy test?
A copy test shows people two to four text variants, such as headlines, calls to action, or onboarding lines, and asks which one they prefer. It is the text-only cousin of a preference test.
Every respondent sees the same options for the same job and picks one. Turn on the optional reasoning question and they also say why, in their own words.
The result is a count per variant, with each respondent's reasoning listed under the option they chose. The count settles the argument, and the reasons tell you what the winning line communicated that the others did not.
Run one once you have a small set of real options to compare, not a blank page.
Is a copy test the right method?
It is worth two minutes to check a copy test is the method you want.
Reach for it when
Comparing headline, button, or value-proposition wording
Choosing between two onboarding or call-to-action lines
Testing a wording change one variable at a time
Settling a wording debate with evidence instead of opinion
Reach for something else when
You're comparing visual designs, not wording a preference test is the visual equivalent of a copy test.
You only have one piece of copy to react to not several to compare. An open text question or a design feedback survey fits a single piece better.
Copy testing tips
Test one variable at a time
Compare headline against headline before testing combined variants. Compound changes make preference impossible to attribute.
Keep every option doing the same job
Each variant should compete on the same job, so the comparison actually measures wording rather than concept.
Read the open text before the split
A close split, such as 55/45, means wording isn't the deciding factor on its own. The reasoning behind each choice is where the real insight is.
Frequently asked questions
What is a copy test?
A copy test shows respondents two to four text variants, such as headlines or CTAs, and asks which one they prefer, with an optional question about why.
How many options can I compare in a copy test?
Between two and four text variants. Keep every option focused on the same job so the comparison stays useful.
Can I ask people why they chose an option?
Yes. Turn on the optional reasoning question, and each respondent's explanation appears under their chosen variant in the results.
What is the difference between a copy test and a preference test?
A copy test compares text-only options, such as headlines or CTAs. A preference test compares visual designs or images.
How many respondents do I need for a copy test?
Run with at least 30 people to get a stable split on a two-option test. Add more as you add options.
What does the results view show?
Response counts per variant, plus any written reasoning underneath the respondents who chose each one.
Do AI respondents go through the study before real people?
Yes, on every plan including Free. Up to five AI respondents work through the study first so you get a read in minutes and catch an unclear task early, and up to two of them can be AI experts in conversion, usability, copy or SEO. Then real people answer the same study, recruited from Prolific's global panel or reached through your own link.
Can I create and launch this study from my coding agent?
Yes. Fiuto's research assistant works in the app, over MCP from Claude Code, Cursor, Codex, Lovable, v0, Replit and Bolt, and over the API. It creates, launches and reads back the study on every plan including Free.
Does Fiuto run AI interviews?
AI interviews are an upcoming feature, in development now, and this page will list them the day they are out of alpha. Until then Fiuto runs unmoderated studies across 18 methods, with AI respondents first and real people after.