After a brief passes through several people, a cautious observation can start sounding like a fact. “One interview mentioned handover difficulties” becomes “our customers need automated handovers”. An evidence register keeps those steps separate. For a team preparing ChatGPT ads, it makes the reason for choosing a customer need visible, along with the information that might change the decision.
The register is an internal editorial aid. It is not a record of individual ChatGPT users or a list of conversations guaranteed to show the ad. Use research sources you are entitled to use and provide references the team can inspect. Avoid copying identifying information when an anonymous source reference is sufficient.
Give the need a stable identity
Use a short name describing the task, such as “collect decision context before a handover”. Add an internal identifier if several needs could otherwise become confused. The name should remain useful when the ad headline changes.
A product name usually makes a poor need label. It ties the reasoning to the solution you want to sell and makes it harder to notice when the research describes different work. A broad term such as “efficiency” is also inadequate. The register should show the task behind the term.
Do not create a new need for every new expression. Two participants may describe the same work differently. If the difference does not change the task, conditions or offer, collect their observations under one need and preserve the vocabulary as examples.
Separate five kinds of note
A small register can work with the following fields. The fictional example concerns software for recording project decisions.
| Field | Example entry |
|---|---|
| Observation | A participant needed the background to earlier decisions |
| Source | Research note R-04, current project handover |
| Interpretation | Decision context may matter when responsibility changes |
| Contradiction or boundary | Another small project managed with a meeting |
| Editorial decision | Investigate handovers with scattered records further |
Keep the observation close to what the source actually says. The interpretation can reason beyond it, but must remain visibly an interpretation. The decision records what the team does with the evidence. Separating these fields allows the interpretation to change without rewriting the participant’s account.
If the material comes from reviews, use the customer-language coding method first. It helps distinguish a situation from an evaluation or condition before the note enters the register.
Include a date and a location within the source in its reference. Label a summary as a summary so it cannot later be mistaken for a verbatim quotation. If the same interview appears in both a presentation and a sales note, connect them to the same origin. Those documents contain one observation reported twice, rather than two independent observations.
From note to decision
- Observation R-04
A participant needed to find decision context.
- Interpretation
Collected history may be relevant.
- Counterexample
A small project managed with a meeting.
Reconsider - Next step
Investigate handovers with scattered records further.
Checkpoint
Explain the uncertainty that remains
A label such as “low confidence” does not tell the writer what to do. State the reason instead: few independent observations, an old product version, an uncertain sequence or a sample containing only existing customers. Different gaps need different follow-up work.
In the handover example, the need may be well described for one person but poorly distinguished from ordinary document management. The next step is then to investigate the difference between an ongoing task and a change of responsibility, rather than collect more positive comments about the product.
Avoid confidence percentages unless you have a defined method for them. A precise-looking number may conceal the fact that the judgement rests on the same few notes. A brief explanation of the gap is more useful to the next researcher.
Connect the need to the decision it influenced
When a need is used in an ad brief, link the version that relied on the evidence. Record the brief’s version number or date alongside the link. A link to a document that is continually overwritten otherwise shows only its latest wording. State which part it informed: the selected situation, leading question or scope boundary. This allows the team to locate affected drafts when the research changes.
The register should not independently approve product claims. Evidence that a customer wants collected decision context does not show that the software provides it. Use the ad-copy claim evidence matrix for that separate check. Customer need and product capability must meet, but they require different evidence.
Keep platform decisions separate too. OpenAI’s targeting documentation, checked on 29 September 2026, describes context hints as supplementary context, not exact-match keywords or delivery guarantees. Recording an audience boundary does not configure a platform restriction. Do not use the register to infer negative keywords; geographic, platform and audience restrictions belong in the campaign’s targeting settings.
Preserve the reason for removing a need
Suppose later interviews show that the difficulty mainly concerns unclear responsibility, while the product only collects notes. The original advertising task may be wrong. Mark the row as reconsidered and explain why it is no longer being used in the same form.
Do not simply delete the old interpretation. Without a recorded reason, another writer may find an earlier brief and restore the same claim. Keep a short history containing the date, decision and reference to the new evidence. This does not require an elaborate administrative system.
If the need informed hints, the hint evidence review can help identify descriptions that need attention. Follow the references from observation to brief and then to the actual drafts.
Check whether the register works without narration
Ask a colleague to choose a row and answer three questions: what was observed, what conclusion did the team draw and which advertising task changed? If answering requires a spoken explanation, improve the entry.
A useful register is not the one with the most fields. It is the one that lets a new writer follow the reasoning, notice uncertainty and make a justified revision quickly. Begin with the needs actually used in active ChatGPT ad briefs. Expand only when further decisions need to be traceable.
Keep the register available while working through the other planning steps in the targeting overview, so later decisions can still be traced to their sources.
Sources and scope
Document the path from research observation to chosen customer need so a campaign brief can be reviewed without assumptions turning into facts.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.