Compare ad copy, positioning, and offers with AI-simulated audiences. See what comes across and what needs work.
Refine your ideas before testing with real customers.
Savings app · United States · Fictional sample
31 of 50 high clarity ratings
“Can I pause automatic transfers?”
38 of 50 high clarity ratings
“Do I choose how much to save each time?”
Each person = 1 simulated response
See how each study works in a sample report.
Fictional examples of research with AI-simulated audiences.
Illustrative sample. All numbers and responses below are fictional.
AAutomatic saving
“Save automatically, without thinking about it.”
BPersonal control
“Stay in control of what you save.”
In this example, A is more appealing and distinctive. B is clearer. Intent to try is similar, so the useful next step is to explain how automatic saving and personal control work together.
Positive ratings, with the response counts behind them.
B has more clear ratings. Both messages leave some people unsure.
| Answer | A | B |
|---|---|---|
| Completely clear | 12 · 24% | 18 · 36% |
| Mostly clear | 19 · 38% | 20 · 40% |
| Partly clear | 12 · 24% | 8 · 16% |
| Mostly unclear | 5 · 10% | 3 · 6% |
| Not clear at all | 2 · 4% | 1 · 2% |
| Total | 50 · 100% | 50 · 100% |
Four fictional responses show the benefit people take away and the details they still need.
“It helps me save without remembering every week.”
“Can I pause automatic transfers?”
“I choose how much money goes into savings.”
“Do I choose how much to save each time?”
Keep A’s convenience. Make B’s control concrete. Explain who chooses the amount and when transfers happen.
“Save automatically.
Set the amount. Pause anytime.”
Untested revision · Use only if these features are true.
This sample illustrates a concept comparison. In the example setup, each simulated respondent sees one message and answers the same five rating questions and two open questions.
Real Prefield studies generate responses with AI-simulated audiences informed by market data. Rating shares are calculated from responses; AI interprets the patterns. This fictional sample comes from no completed study and demonstrates no validated difference, market demand, or conversion forecast. Use simulated findings to refine a draft and plan research with real customers.
Illustrative sample. All numbers and responses below are fictional.
A home-cleaning starter kit with two reusable spray bottles and four concentrated refills. Add water at home. Reorder when you need more.
Fictional product and prices. No cleaning-performance or environmental claims were supplied.
Interest is stronger than readiness to try. In this example, the main unanswered question is whether the concentrate cleans as well as a familiar spray. Make the first refill easy and show performance before leading with a subscription.
Three separate ratings show where enthusiasm becomes more cautious.
| Answer | Interest | Try | Priority |
|---|---|---|---|
| Very positive | 28 | 21 | 19 |
| Somewhat positive | 43 | 38 | 34 |
| Neutral | 18 | 24 | 28 |
| Somewhat negative | 8 | 11 | 13 |
| Very negative | 3 | 6 | 6 |
| Total | 100 | 100 | 100 |
The practical tradeoff sits between reducing clutter and adding another task.
Fictional theme counts from open answers, out of 100 respondents. One response can mention several themes, so the counts do not add to 100. Theme coding is an interpretation of the responses.
A possible journey, inferred from the example responses.
Show the space saved.
Question: “Will it work?”State the cost per bottle.
Question: “What do I get?”Mark the water fill line.
Question: “Did I mix it right?”Keep reordering flexible.
Question: “Can I buy just one?”This is a qualitative journey outline, not observed behavior or measured stage-to-stage conversion.
Four fictional responses put the themes in context.
“I like the idea of fewer bottles under the sink. But I need it to cut kitchen grease.”
“If there is a line on the bottle, I can mix it. I don’t want to measure anything.”
“Tell me what one full bottle costs. A tiny refill looks expensive on its own.”
“I finish bathroom cleaner slower than kitchen spray. Send more when I ask.”
Prototype the refill experience before extending the range. Include a clear fill line, explain what the starter kit contains, and make the per-bottle cost visible. Treat cleaning performance as a question to verify.
A familiar clean.
One bottle to keep.
Untested positioning direction. Support any performance or environmental claim with appropriate evidence.
This authored example illustrates an idea-exploration report. Its scenario uses a constructed audience of 100 simulated household-product decision makers in the United States. It is not a representative estimate of US households.
Each uses a five-point scale, from very positive to very negative. All 100 example respondents answer every rating question. The charts show each question separately.
Ratings describe stated responses. Drivers, barriers, and the journey summarize interpreted themes. The figures and quotations were written for this sample; no study was run. They establish no product efficacy, environmental benefit, willingness to pay, or market demand. In a real study, use simulated findings to sharpen the prototype and questions for research with real customers.
Illustrative sample. These numbers and responses are authored fiction, not a completed study.
In this example, approval chasing is a recurring burden. Readiness to try another tool is unresolved, and stated willingness to pay $79 a month falls below the chosen threshold. Prototype the handoff before committing to a paid launch.
1 of 3 hypotheses supported in this fictional sample.
Each claim uses a 50% support threshold. The scoring band determines whether the result clears it.
76% 76/100+26 percentage points above threshold · band 71–80%
54% 54/100+4 percentage points above threshold · band 49–59%
32% 32/10018 percentage points below threshold · band 28–37%
The two highest answers count as support. The rest show where the uncertainty sits.
See the claim, the question, and exactly which answers support it.
76 of 100 select three hours or more. The whole scoring band is above 50%, so this sample supports the time-cost assumption.
“In a typical week, how much time do you personally spend following up on client feedback and approvals?”
| Answer | Count / share | Supports H1 |
|---|---|---|
| Less than 1 hour | 8 / 8% | No |
| 1–2 hours | 16 / 16% | No |
| 3–5 hours | 49 / 49% | Yes |
| 6 hours or more | 27 / 27% | Yes |
| Supporting answers | 76 / 76% | 49 + 27 |
54 of 100 choose definitely or probably. The band crosses 50%; a slim majority alone does not resolve the assumption.
“How likely would you be to use a dedicated approval workspace on your next client project?”
| Answer | Count / share | Supports H2 |
|---|---|---|
| Definitely | 18 / 18% | Yes |
| Probably | 36 / 36% | Yes |
| Might | 28 / 28% | No |
| Probably not | 12 / 12% | No |
| Definitely not | 6 / 6% | No |
| Supporting answers | 54 / 54% | 18 + 36 |
32 of 100 choose definitely or probably. The whole band is below 50%, so this sample refutes this specific price-and-offer assumption.
“For a workspace with five team seats and unlimited client reviewers at $79/month, how likely would you be to start a paid subscription in the next three months?”
| Answer | Count / share | Supports H3 |
|---|---|---|
| Definitely | 6 / 6% | Yes |
| Probably | 26 / 26% | Yes |
| Might | 30 / 30% | No |
| Probably not | 24 / 24% | No |
| Definitely not | 14 / 14% | No |
| Supporting answers | 32 / 32% | 6 + 26 |
Illustrative answers to the shared follow-up: “What most influenced your answers above?”
“I spend Friday finding out which version the client actually approved.”
“I would try it if the client can review from a link. Another login will slow us down.”
“We already pay for project management. Show me which part of that this replaces.”
“The price could work if it saves us one round of revisions, but I need to see that first.”
These authored reactions suggest questions to investigate; they do not establish how common each reason is.
Separate evidence of a problem from evidence for a particular solution.
Test a version history and one clear sign-off state. Watch agency leads retrieve the last approved file from a recent project.
Observe an agency and its client reviewing one deliverable. Check whether invitations, access, or existing tools interrupt the workflow.
Interview the budget owner about current tools, spend, and approval costs. Test a revised package after a real pilot; this sample does not identify a better price.
This is an authored example, not research that was run. The scenario uses a constructed audience of 100 US creative-agency owners and project leads, with the same 100 answering all three questions. It is not a representative sample of agencies.
A concept for a workspace that keeps creative files, client feedback, version history, and final sign-off together. The paid offer specifies five team seats, unlimited client reviewers, and $79/month. Those are fictional product terms, not Prefield pricing.
Each hypothesis has one choice question, explicit supporting answers, and a 50% threshold set before the responses. The three questions above are followed by one shared open question. Each result is calculated from the 100 answers shown; no answers are missing.
Bands use z = 1, approximately 68%, as in Prefield’s sampled-answer implementation. Unrounded bounds decide the verdict; displayed bounds are rounded. An interval calculated from simulated answers does not measure uncertainty about real-world demand or correct model bias.
The three claims, their questions, and the next investigations. Actual Prefield studies generate simulated responses and calculate verdicts from the study’s selected scoring method. Professionally defined audiences and custom product questions are not established as valid by population benchmarks. Check the workflow with agency staff and clients, and payment behavior with budget owners.
Audiences informed by local population data and calibrated against market-specific benchmark surveys.
Our global coverage keeps growing.
Calibration is assessed on specific benchmark questions. Custom product studies remain directional and should guide further research with real customers.
Tell us which market you needPractical guides and case studies from Prefield.
Use simulated responses to explore an offer, sharpen your questions, and plan more focused research with real customers.
Read articleExploring an e-wallet offer in the Philippines: transfer frequency, recurring payments, and the case for a prepaid alternative.
Read case studyAn exploratory lending study looks at identity verification, accountability, and the value of a route to human support.
Read case studyNo. The responses are generated by AI for a simulated audience. Use them to explore possibilities and prepare research with real customers.
Messages, positioning, and offers; product ideas; and assumptions about your audience. Start with your copy, a brief, or your own questions.
Prefield organizes a study around a defined audience and questionnaire. It brings the responses together in a report with comparisons, audience details, and method information that you can inspect and share.
Use them to refine wording, identify possible objections, compare alternatives, and prepare follow-up questions. Validate important decisions with real people and observed behavior.
No. Benchmark testing assesses specific questions and conditions. It does not automatically validate a new product question or turn a simulated response share into a sales forecast.
Tell us what you’d like to test. We’ll follow up about access.
We will follow up at the email address you provided.
