Audience research
Part of Researching creator-platform buyers and creators without mixing the two populations
Five interview prompts that surface how creator-platform decisions were really made
Use five evidence-led creator-platform interview prompts with explicit recruitment, consent, accessibility, safeguarding and inference limits.
What to take away
- Anchor each interview to one dated decision and the participant's authorised role, not a general view of platforms.
- Ask which record they used, what its fields meant, and how a wrong entry gets corrected.
- Trace where work stopped or changed hands, so governance gaps appear without pitching a product.
- Code answers against named decision stages and keep an unclear category rather than forcing tidy themes.
- Findings cover these participants only, never national prevalence, adoption or demand.
Fix the sample and the scope first
Research date: 6 September 2026. Scope: an England-linked buyer, creator or governance study of a defined discovery database, marketplace, workflow service or manual register. Decide which of the six creator discovery routes the decision sat in.
Set a floor before recruiting. A minimum of six completed interviews per role group works, with reviewers and suppliers counted separately. That is a design rule, not a market estimate.
Inclusion needs direct, documentable experience of the named decision inside the recent window. Exclude follower-based recruitment, guessed platform use and general social-media opinions.
The five prompts, in order
| Prompt | Target | Record |
|---|---|---|
| "What decision were you authorised to make?" | The dated event and the role held | Role label plus evidence of the event |
| "Which record did you rely on, and what did its fields mean?" | Provenance and permitted use | Field definition, source, timestamp, correction route |
| "Where did the work stop, change hands or need approval?" | Hand-offs and sign-off | Evidence that crossed each hand-off |
| "What would have disproved the proposed action?" | Contrary evidence | Disconfirming factors offered without prompting |
| "What evidence had to survive correction, withdrawal or exit?" | Retention and deletion | Records kept, exported, corrected or deleted |
Prompt one anchors the account to a dated discovery, procurement, hand-off, correction or exit event. Record whether the participant acted as advertiser user, budget authority, creator, intermediary or supplier. Do not merge answers from people with different authority.
The GOV.UK Service Manual recommends clear participant criteria. A role label alone is not enough, because eligibility needs evidence of the event.
Prompt two covers the record itself. Ask for the field definition, source, timestamp, permitted use and correction route. A redacted artefact often helps. Do not invite guesses about age, residence, protected traits, interests or audience composition.
The ICO's transparency guidance explains what people must be told about personal data gathered directly and from elsewhere. A public profile is not permission for research or reuse.
Prompt three traces the work. Follow each hand-off, from buyer user through procurement and agency to the creator and the reviewers. Ask what evidence crossed it and what stayed unresolved.
Prompt four invites contrary evidence: stale provenance, absent creator authority, inaccessible participation, missing rights or an adequate manual baseline. The Government Analysis Function's questionnaire design guidance warns against leading wording, so apply that discipline here.
Prompt five asks what evidence had to survive correction or exit. Which records were kept, exported, corrected or deleted, and who confirmed completion? Do not request confidential campaign data to make an interview vivid.
Give participants clear information on recording, use, retention and withdrawal. Later testimonials need separate permission and a fresh review.
Example: coding one answer to a stage
Interviewer: "Which record did you rely on, and what did its fields mean?"
Buyer: "I looked the account up, it seemed fine, so I briefed the creator."
Coded against the stage "record verification", that answer sits in the unclear category. It names no field, source or timestamp, so it cannot count as verified provenance. The stage list keeps it visible for follow-up instead of tidying it into a favourable theme.
Publish this pack before fieldwork starts
Publish these items before any call, so the method can be checked while fieldwork runs.
Publish before fieldwork
- Frame and recruitment route
- Invitation and screener versions
- Dates and eligible counts
- Responses per item
- Incentive terms and withdrawals
- Accessibility and safeguarding route
- Missing populations
- Scope, research date, inclusion and exclusion rules.
- Recruitment route, invitation wording and screener versions.
- Incentive terms, accessibility provision and safeguarding route.
- Consent, recording, retention and withdrawal rules.
- The decision-stage list you will code against.
Administer, code and stop fieldwork
Use the same core prompts for participants in the same role, allowing relevant follow-up. Log the interviewer, mode, recording status and any interruption or withdrawal.
Code observations against the pre-declared stages, keep an unclear category, and count contrary cases. Quotations need explicit permission and a note of any editing.
For screening, buyers and creators have to be kept apart so the two populations are not blended into one theme.
Stop fieldwork if recruitment relies on scraped data, under-18 risk lacks a safeguarding protocol, access needs cannot be met, or withdrawal cannot be honoured. Publication stays on hold until those points are resolved.
Common questions
What must be published before fieldwork begins?
The scope, recruitment route, invitation and screener versions, plus dates and invited counts. Add incentive terms, withdrawals, accessibility provision and the safeguarding route.
How should interviewers handle answers that do not fit a theme?
Code them against the decision stages and preserve an unclear category. Count contrary cases and report them rather than forcing a favourable theme.
How many interviews are enough?
A floor of six completed interviews per role group is workable. Stop adding participants in a role once new accounts repeat what earlier ones covered.
When should fieldwork stop?
Fieldwork stops as soon as any of the four conditions above is met. Publication stays on hold until the method is fully documented.
