Audience research for a website project that keeps buyers and end users apart

Good website research begins by separating the organisation that commissions the work from the people who must use the finished service. A procurement lead, content owner and visitor may all influence the same build, but they answer different questions. Combining them into one vague "audience" produces confident-looking notes that cannot support a decision.

This guide sets out a research design, not findings. No interviews, surveys or observations were conducted. It contains no invented persona, quotation, sample size or prevalence claim. Research closed on 6 September 2026 and relies on UK government and regulator material, with England identified where a population can genuinely be bounded.

What to take away

Write the decision before the discussion guide

A team should be able to finish this sentence: "After this research, the named owner will decide whether to ___ because we will know ___." If the blank is merely "understand users", the project is not ready for recruitment.

The GOV.UK Service Manual page on planning a round of user research recommends clear, actionable objectives tied to problems, assumptions and the next decision. Although written for government services, that discipline is adaptable to a commercial website. It does not create a universal private-sector method.

For an agency appointment, decisions might include whether to retain an assisted channel, which content must be rewritten, or whether a prototype can proceed to build. Each question needs an evidence threshold and a stop rule. "Most participants liked it" is not a threshold unless the population, recruitment and measure were defined in advance.

Keep buyers and end users in different frames

The buyer population contains people who authorise, specify, procure, operate or assure the website. A buyer sample might include the service owner, budget holder, procurement officer, content lead, security lead and accessibility specialist. They can explain constraints and approval paths. They cannot stand in for customers using the service.

The end-user population contains people who attempt the task, including people who abandon it, use another channel or need assistance. Define eligibility through recent behaviour or a likely future need, not through a convenient demographic label. The Service Manual guidance on finding participants recommends recruiting actual or likely users and explicitly considering disabled people, limited digital skills and support needs.

Create separate frame records:

Broad statistics can help locate organisations but cannot supply needs. The ONS 2025 business activity dataset reports VAT or PAYE registered enterprises and local units by industry and region from a March 2025 snapshot. It is a possible administrative frame for organisational sampling. It does not identify website buyers, user populations or demand for an agency.

Turn assumptions into recruitment criteria

List each current belief as a hypothesis with a source and a way to disprove it. For example, a team might suspect that account recovery fails for people using a particular access route. The screen should then recruit people with relevant recent experience, while avoiding a question that reveals the desired answer.

Recruiting only current customers can omit people who cannot start or finish the journey. Recruiting through an agency may omit people outside its panel. Workplace sessions may suppress criticism when managers observe. These are design limitations to document, not problems to hide after fieldwork.

The Service Manual recruitment guidance says bias is difficult to avoid. It suggests varied activities and recruitment routes, and recommends an incentive at the start of a session to avoid implying a particular response is required.

An England project should set value, currency, tax treatment, expenses and approval route before inviting participants. This article gives no amount.

Design participation around access, not convenience

Ask people what adjustment or communication route they need. Do not infer it from a diagnosis, and do not make disclosure of medical details a condition unless a qualified reviewer establishes necessity.

The GDS page on running sessions with disabled people advises teams to plan the venue and equipment, speak directly to the participant, ask before helping and avoid assumptions about what a person can do. It also notes that personal assistive-technology settings can be difficult to recreate elsewhere.

Access planning includes travel, breaks, interpretation, captioning, quiet space, accessible formats and the participant's own device where appropriate. It also includes remote alternatives for someone who cannot attend. Record which adjustments were offered, accepted and successfully provided. A mixed sample does not prove that a prototype is accessible; accessibility evaluation remains a separate gate.

Equality and accessibility review should examine the recruitment screen, incentive, venue, research artefacts and analysis, not only the final interface. A missed group should appear in the evidence gap register rather than being represented by a fictional persona.

Explain participation and data use separately

Research participation needs a plain information route. GDS guidance on getting informed consent says participants should understand who is conducting the work, its purpose, what will happen, data collected, intended use and sharing, retention, recording, observers, complaints and withdrawal.

That operational consent record does not automatically settle every UK GDPR lawful basis, special-category condition or employment issue. A named privacy reviewer must assess the actual processing. The ICO's data-minimisation guidance requires personal information to be adequate, relevant and limited to what is necessary for the stated purpose.

Use a participant code in notes and separate the contact key. Restrict raw recordings and transcripts to named roles. State the deletion date or trigger for contact details, recordings, consent evidence and analysed outputs separately. If a transcription, recruitment or survey provider receives information, document its service, role, location, subprocessor and deletion terms before collection.

Research contact must not become an undeclared sales route. The ICO page on identifying direct marketing says contact for genuine market research is not direct marketing. Mixing a promotional message into the exercise, or using research details for later sales, changes the facts and needs separate review under UK GDPR and PECR.

Employees require another boundary. Interviews about internal publishing may involve people at work, while screen recording or usage logs can become worker monitoring. Current ICO worker-monitoring guidance is marked under review after legislative change and stresses purpose, necessity, proportionality and transparency. Managers should not receive identifiable comments by default.

Choose the method for the uncertainty

An interview can reveal language, circumstances and a person's account of events. It cannot prove how frequently something happens. An observed task can reveal behaviour within the test setting, but a prototype may not reproduce the live service. A survey can estimate a defined measure only if the frame, question, response process and analysis support the inference.

Use a small method map:

GDS guidance on using in-depth interviews recommends current or likely users, open neutral questions, real examples and a maintained research record. It allows interviews to be combined with moderated usability work. The distinction matters: what a participant says, what the researcher observes and what the team infers should occupy separate columns.

Use personas as changeable evidence summaries

A persona card is useful only when it points back to evidence and exposes uncertainty. Do not invent a name, portrait, quote or biography to make the page feel real. Start with a blank card containing population, task, context, access route, constraint, source records, contradictory evidence, confidence, excluded groups, owner and expiry date.

The team should be able to update or delete the card when research disagrees with it. A buying role and an end-user pattern belong on separate cards. Several people may share a task without sharing age, job title or motivation. Conversely, one buyer may authorise a project but never use the public journey.

No card represents population prevalence unless a suitable quantitative study supports that claim. A vivid interview episode is evidence about that episode. It is not proof that a segment is large or commercially valuable.

Procure survey software as a data route

Tool selection follows the research design. Define respondent access, question types, languages, collaboration, export, retention, hosting, user administration, accessibility tests, contract terms and exit. Then compare the exact product edition available to the organisation.

Public records for Microsoft Forms, SurveyMonkey self-serve and Typeform Free differ in scope and detail. Product pages and vendor security statements are supplier claims, not independent tests. A buyer must obtain the applicable contract, data terms and assurance evidence for its edition, test an accessible questionnaire with representative users, export authorised data and verify deletion. Unknown remains a procurement question.

Price is excluded here because the three services use different plan structures and no like-for-like England quote was collected. An existing approved tool or a manually managed questionnaire may be the better baseline. Introducing software is not a research achievement by itself.

Analyse without manufacturing certainty

Create an evidence table immediately after each session. Give each item a stable reference and classify it as participant statement, observed behaviour, artefact or researcher interpretation. Include contradictory cases instead of voting them away.

GDS guidance on analysing a research session recommends organising and interpreting material soon after the round, with observers involved. For this context, the team should also log who coded the material, how themes were defined, whether another reviewer challenged them, and which original records remain accessible.

Report recruitment and non-response before findings. State invitations, eligible responses, attended sessions, withdrawals and unusable records as separate counts when real fieldwork occurs. Do not invent those numbers in a plan. Explain how timing, channel, incentive, observer presence and agency-client relationships might have shaped participation.

The decision log should connect one finding to one action, owner, evidence threshold and review date. A single session may justify investigating a defect, not rebuilding an entire service. If the evidence does not change a decision, stop collecting personal data and reconsider the objective.

Hold publication until the boundaries are reviewed

The research lead signs the method and limitations. The service owner accepts decision use. Qualified reviewers separately consider accessibility and equality, privacy and PECR, participant or worker monitoring, security, IP and database rights, advertising and consumer claims, procurement, contracts and finance. Passing one gate cannot offset failure in another.

Before any public claim, remove or lawfully authorise identifying material, trace every statement to an evidence record and confirm participants were told about the intended use. Report absence of evidence plainly.

A strong audience-research programme does not sound certain; it lets a reader see exactly whom the team heard, whom it missed and why the resulting decision was proportionate.

Before you act

Common questions

Why should buyers and end users be researched separately?

Buyers authorise, specify, procure, operate or assure the website, so they explain constraints and approval paths. End users attempt the task, including people who abandon it or need assistance. Combining them into one vague audience produces notes that cannot support a decision, and buyers cannot stand in for customers using the service.

What should a team settle before recruiting participants?

A team should finish the sentence naming the owner, the decision and what will be known. Objectives must tie to problems, assumptions and the next decision, with an evidence threshold and stop rule. The article says if the blank is merely understand users, the project is not ready for recruitment.

Does a consent record cover all data protection duties?

No. The article says operational consent does not automatically settle every UK GDPR lawful basis, special-category condition or employment issue. A named privacy reviewer must assess the actual processing, and personal information must be adequate, relevant and limited to what is necessary for the stated purpose.