Turn Sapien research into a useful ChatGPT or Claude brief

A research readout and a campaign brief do different jobs. The readout explains what was studied and what the evidence suggests. The campaign brief turns that understanding into copy, creative choices and the next test. When a team moves between them, it is easy for a qualified finding to become an unqualified promise.

ChatGPT or Claude can help with that translation if you give it a clear evidence package and a specific assignment. The useful handoff is small enough to inspect and detailed enough to preserve why the recommendation was made. You can prepare it manually from material you are authorized to use.

Decide what the next document needs to achieve

Consider Tern, a fictional planning app for small agency teams. Its marketing lead has explored two messages in a simulated-audience study and now needs a landing-page brief. The decision is how to explain the app and what product proof belongs beside that explanation.

For this illustration, imagine the approved readout says that some simulated respondents interpreted “Your week, handled” as a promise that Tern would allocate work automatically. It also says that the product actually helps a manager organize a plan. These are invented example findings, rather than results from a Sapien study.

The next assignment is to address that misunderstanding in a landing-page brief. Asking a chatbot to “make this research into a winning campaign” would leave both the evidential standard and the deliverable unclear. Asking for two hero concepts, the reasoning behind them and the product proof each needs gives the team something it can review.

Prepare a source package with clear roles

Use an approved copy of the research material, a product-facts document and a short assignment. You can paste a concise package into the conversation or attach supported documents. OpenAI's evidence-synthesis guidance recommends inventorying the source corpus, keeping findings distinct from interpretation and checking citations against originals. Anthropic's file-upload guidance explains how to attach documents in Claude and which formats it supports.

Choose material your organization permits you to share with the selected account. A finished report can still include confidential client information or identifiable customer quotes. Review it before transfer, and use a redacted or summarized version where that gives the assignment sufficient context. The account's applicable terms and your organization's rules govern that transfer.

For Tern, the package could contain three plainly named documents:

• Study summary: the audience, objective, stimuli, questions, modeled findings and limitations, with page or section references back to the approved readout.
• Product facts: what Tern does, what it does not do, the current user flow and which screenshots or demonstrations are available as proof.
• Writing assignment: the reader, page placement, required deliverable, voice and constraints on claims.

If the report is interactive, provide the relevant authorized material in a readable form and retain a reference to the original view. A chatbot should not have to guess at an inaccessible dashboard.

Label the evidence clearly inside the summary. Preserve the difference between a simulated response, a real customer quote, a measured behavioral result and the team's interpretation. Include the original question and the version of the message that produced a reaction. A quote about one headline should not appear to support a revised headline that nobody has yet evaluated.

Confirm the material was read before asking for copy

A successful attachment is not proof that every table, chart or image was interpreted correctly. Ask for a brief source inventory first, then check it against what you provided.

For Tern, that instruction might be:

Read the attached study summary, product facts and writing assignment. List the documents and the sections you can access. Summarize the audience, the messages actually tested, the evidence type and the main limitations. Identify missing or unreadable material before drafting. Treat simulated responses as modeled evidence. Do not infer results for an untested message or invent a product capability.

Review that response. If the tool describes Tern as automatically allocating work, correct the source package before continuing. Repeating the correction in a later prompt while leaving contradictory product notes attached gives the next draft an avoidable ambiguity.

Tables deserve a direct check. Confirm the column labels, units, audience bases and which version each result belongs to. If a chart is the only place a finding appears, include a readable explanation of the relevant values. Keep the original nearby so the team can verify the summary.

Ask for a draft that explains its choices

Once the source inventory is correct, give the writing assignment. Keep the requested output focused on the decision the team is ready to make.

Draft two landing-page hero concepts for Tern's agency-manager audience. Each should include a headline, supporting sentence, button text and a description of the product proof needed beside it. Use only confirmed capabilities from the product-facts document. Address the mistaken expectation that Tern allocates work automatically. For each concept, explain which finding motivated the choice and cite the supplied document and section. Mark the concepts as new candidates to test. Do not claim that either has improved conversion or been validated by the study.

The response should make the reasoning inspectable. A proposed headline such as “Build a clearer plan for your team's week” might address the interpretation problem, but the team still needs to judge whether it is accurate, distinctive and useful in the actual hero. A screenshot showing a manager arranging work could support that explanation if it faithfully represents the product.

The research does not decide the final wording on its own. It gives the writer a reason to explore a direction and a problem the next version should resolve. Our message testing guide follows that relationship from a misunderstood offer to a specific revision.

Review the claims before polishing the prose

Read the draft against the source package. Check that each evidence-based statement points to the right finding and that the finding supports the statement at the level claimed. A source reference can be accurate while the sentence still overreaches.

For Tern, “some simulated respondents expected automatic allocation” is different from “agency managers want automatic allocation.” The first describes an interpretation in the illustrated study. The second asserts a market need the example has not established. Similarly, “this candidate addresses the misunderstanding” is different from saying the candidate will increase signups.

Check that proposed proof exists. If the draft requests a live demonstration of a function Tern lacks, revise the concept. If a useful question remains unanswered, carry it into the next research or product conversation rather than filling it with plausible copy.

Then edit for the reader. Remove a detailed methods explanation if it belongs in the supporting rationale rather than the hero brief. Keep the source references in the internal working document so a colleague can understand the recommendation without reconstructing the chat.

Return the final decision to the research record

Save the approved brief with the source versions that informed it. Record which concept the team chose, what it changed and which question the next test needs to answer. A follow-up study needs the actual revised stimulus, not a summary such as “we made the headline clearer.”

This gives the team a traceable path from research to writing to the next observation. Sapien helps explore audience reactions and their explanations; ChatGPT or Claude can help organize that material into a working brief. The quality of the handoff comes from preserving the evidence while giving the next document a clear purpose.