Plan audience research interviews for AI marketing tools

What to take away

1. Set the inclusion criteria and the geography

Write the criteria down before you approach anyone. A workable panel mixes the person who releases budget, the person who uses the tool each week and the person who could block the purchase.

Anyone who has not been near a buying decision in the past year is a weak source. State the geography too: buyers in England may be only one part of a UK-wide market, so include buyers in Scotland, Wales or Northern Ireland when they are in scope.

2. Recruit for the decision, not for enthusiasm

Screening questions that mention AI marketing tools attract people who like the idea of them. Screen on the task instead: what did they need to do, what did they try, who else was involved, and where did an AI feature enter, if at all?

Enthusiasts supply quotes that flatter a product and hide the friction that stalls a deal. Five or six screening questions separate the two groups.

Recruit through existing customer communities, sales or customer-success referrals, professional networks, or specialist recruitment platforms such as Prolific, Respondent and User Interviews. Check each platform's current participant pool, screening options and terms; do not assume it covers your target segment.

3. Ask about the last decision rather than next year

Recency beats prediction. A buyer describing a purchase from four months ago can give you the trigger, the shortlist and the objection that nearly ended the process. A buyer describing a distant rollout gives you a guess.

Keep the recall window to twelve months and push for specifics. Ask what they actually used or evaluated, using HubSpot, Salesforce Marketing Cloud, Adobe Marketo Engage, Mailchimp or Jasper as prompts rather than a presumed shortlist. Ask which AI-enabled task they tried, what alternatives they considered, who approved the decision and why they bought, rejected or stopped using the tool.

4. Handle consent and interview records properly

Recordings, transcripts and notes are personal data when they identify a person. Apply UK GDPR and the Data Protection Act 2018, and document the lawful basis for processing separately: a participant agreeing to take part is not, by itself, the UK GDPR lawful basis. Before recruitment, set a specific deletion date for identifiable audio, transcripts and notes, tell participants what will be deleted and when, and keep anonymised findings only in line with your retention policy.

If a recording reaches the wrong person, treat it as a data breach. Under UK GDPR Article 33, assess whether the breach is likely to risk people's rights and freedoms; where notification is required, notify the ICO without undue delay using its online personal data breach reporting route. Record the decision and its reasoning. Consent wording that covers research in general terms is too vague to rely on later.

5. Review what the tools do with the material

For transcription, evaluate Otter, Fireflies.ai, Microsoft Teams transcription or Google Meet transcription; Dovetail is another option for qualitative analysis. Before uploading identifiable recordings, check that your organisation has approved the service and verify access, storage, subprocessors, retention and whether uploads can be used to train models. If those terms are unclear, use an approved internal workflow or take notes without uploading identifiable audio. Treat automated summaries as drafts, check them against the recording and keep a named person accountable. GOV.UK's Understanding artificial intelligence ethics and safety guidance is a useful checkpoint for reviewing and challenging outputs.

6. Keep the claims within the rules

Interview quotes turn into ad claims quickly, and a claim has to be accurate and not misleading. The Consumer Protection from Unfair Trading Regulations 2008 apply across the UK and set how commercial claims are presented. Where a figure came from two buyers in one region, label it as such.

Before findings go near a budget deck, label the sample and its limits, distinguish participant evidence from your interpretation, and corroborate any proposed ad claim against other sources. Do not present a small interview panel as a market-wide estimate.

7. Keep a short glossary

Shared vocabulary stops the write-up drifting:

Common questions

How many interviews should I run?

Three to five per segment usually shows whether the language is stable. If two people in a row describe the same trigger and objection, extra calls add little. Book another batch only when the segment is unfamiliar or high value.

Should I pay participants?

Payment is a recruitment choice rather than a research requirement. If you pay, keep the amount modest and consistent, and say so in the consent wording so nobody feels obliged to give the answer you want.

Can a tool transcribe and summarise the calls?

Yes, but treat the transcript as the record and the summary as a draft. Check the summary against the recording before it informs any claim, and keep the tool inside the data rules your organisation already applies.