AI message
testing.

What is AI message testing?

AI message testing collects synthetic responses to supplied copy, such as a product description, headline or offer. In Manyside, you choose the audience and questions, retain the exact text, and inspect structured responses rather than relying on one overall critique.

Use a review to prepare questions about comprehension and appeal. Synthetic responses do not establish advertising performance, persuasion or conversion.

Compare wording while holding context steady

1. Paste the full message and explain where a reader would encounter it. Keep the offer, price and other facts explicit.
2. Add named input versions in the workspace when comparing wording. Keep the audience and question set the same.
3. Start with understanding and appeal. Add a custom structured question if you need to examine a particular interpretation.
4. Review distributions and response coverage for each version. Carry a specific hypothesis into an interview or a real audience experiment.

What the existing Refill example tells us

The published Refill study included the tagline “A cleaner home, with less waste” within a full product description. On October 19, 2026, 120 simulated reviewers returned an average understanding score of 2.8 out of 3 and appeal of 3.0 out of 4, with 120 usable responses for each.

Those results apply to the full description. They do not isolate the tagline’s effect or compare alternate messages. Read the actual output and model and run details.

A concrete follow-up comparison

The following is a proposed study, not a completed result. Keep the Refill product description and price unchanged, and vary only the tagline:

• Version A: “A cleaner home, with less waste.”
• Version B: “Keep the bottle. Refill your cleaning spray.”

Ask the same understanding and appeal questions for both versions. A possible hypothesis is that the second wording makes the refill mechanism clearer. No winning version or improvement is claimed before running and inspecting the comparison.

Then show the alternatives to real prospective buyers. Ask them to explain how the product works and what “less waste” means to them. If the decision concerns sales, test actual behavior with a suitable real audience experiment.

Is this an A/B test?

It is a comparison of model-generated responses. A live A/B experiment assigns real people to alternatives and measures observed outcomes. Treat synthetic differences as hypotheses to investigate, not conversion lift or proof that a claim persuades people.

Try it with your own material

Start with a small panel and a clear question. The hosted workspace supports up to 100 reviewers per run. Sign in and verify your email to use your account’s free AI credit.

Start free ↗ Open workspace

How free credit works · Study privacy · Validation evidence

Related: Product concept testing · Message testing · Methodology and validation