From expert interviews to content

1. Commission an answer, not a conversation

An expert interview is a research instrument. Its value comes from the decision it helps explain, not the duration of the recording. Define what the reader should be able to compare, diagnose or implement after using the resulting content. Pick one unresolved question and identify the artifact that could illuminate it. An engineer may show an architecture decision; a customer-success lead may explain a rollout checklist. The interview brief should include intended uses and excluded uses before a recording exists. This reduces the temptation to treat all recorded material as freely reusable marketing inventory.

Team fieldwork

Before inviting anyone, write a commissioning note that an interviewer can use without a verbal briefing. Include the unanswered question, the decision it affects, the expected source material and the intended publication context. Explain what an acceptable answer would contain without prescribing its conclusion. For example, a guide about exception ownership needs evidence about responsibility and escalation, not a general statement that collaboration is important. Send the contributor enough context to prepare a relevant artifact, but avoid sending a finished argument that encourages them to supply confirming quotes. Reserve the option to publish nothing if the material cannot support a useful answer.

Copyable decision worksheet

2. Select perspectives and disclose the sample

Recruit for experience with the relevant decision, including people who saw a different outcome or rejected the favored approach. A single expert can explain a process but cannot represent an entire buyer population. Keep a sampling note with roles sought, people interviewed, recruitment route and gaps. Internal contributors offer access but may share organizational assumptions. External contributors offer another perspective but still require contextual interpretation. Do not present a convenience sample as representative research. When a claim depends on frequency or prevalence, an interview program alone may be the wrong evidence source.

Team fieldwork

Create a recruitment sheet with one row per perspective. Record the role's relationship to the decision, the episode they can discuss and the reason their experience could challenge another account. The person who sponsored a rollout and the person who operated it may describe different constraints without either being dishonest. Interviewing both helps explain that difference. Track declines and unavailable perspectives, because the final source set may differ materially from the planned one. When reporting the method, describe the actual participants and scope. Do not hide a missing operational perspective behind a larger total number of interviews with executive sponsors.

Copyable decision worksheet

3. Use the SOURCE evidence record

The SOURCE record developed here contains Situation, Observation, Underlying artifact, Rights, Confidence and Exceptions. Situation preserves context. Observation captures what the person actually reports. Underlying artifact points to inspectable material. Rights specifies permitted uses. Confidence is the editor's bounded assessment, not a numerical probability. Exceptions records where the observation may fail. This framework makes a quote usable without making it universal. Keep paraphrases separate from quotations and attach timecodes to both. If transcription is assisted by software, verify important names, figures, negations and qualifiers against the recording before using them as evidence.

Team fieldwork

Use stable identifiers such as INT03 and CL07 so claims can be traced without relying on file names or memory. A claim record should link to the relevant transcript segment and the approved version of any supporting artifact. If an artifact is replaced later, preserve enough version information to understand what was checked at publication. Separate confidence in the accuracy of a transcript from confidence in the generality of an observation. An accurately quoted opinion remains an opinion. Add a field for the evidence needed to move from that opinion to a recommendation for other teams, and name the person who can obtain it.

SOURCE record fields

Copyable decision worksheet

4. Write questions that reveal decisions

Ask about the last concrete instance before asking for general advice. Follow the sequence trigger, alternatives, action, evidence and exception. Replace leading questions such as why automation is essential with neutral questions about what was considered and why. Ask for a counterexample early enough that the contributor does not feel their task is to defend your preferred conclusion. Avoid stacking three questions into one. The interview guide is a route through uncertainty, not a script that must be completed regardless of the answer. Allow follow-ups where the contributor reveals an unanticipated constraint.

Team fieldwork

Rehearse the guide aloud with a colleague who plays an expert with limited time. Notice questions that invite a polished speech rather than a concrete account. Replace them with prompts about the last occurrence, an alternative considered or an artifact used. Prepare neutral probes: what happened next, how did you know, who disagreed and when would that advice change? Do not use every probe mechanically. When a contributor has already supplied the information, move forward. Keep a short closing question about omissions, because the most important constraint may be something the interviewer did not know enough to ask about.

Copyable decision worksheet

5. Conduct an artifact-led interview

Open by confirming purpose, recording arrangements, attribution and the contributor's ability to flag material for exclusion. Ask permission before discussing a confidential artifact. Explore one episode deeply rather than covering many topics superficially. Separate what the contributor saw from what they inferred and what another person told them. When a figure appears, ask for period, denominator, definition and source. If those are unavailable, do not press the contributor to improvise. Close by summarizing your understanding and inviting correction. An immediate verbal summary catches mistakes that become harder to notice once a polished narrative exists.

Team fieldwork

During the conversation, use a separate scratch column for interpretations. This prevents your immediate theory from blending into the source's account. If a contributor says a process became faster, ask what faster means in that episode and whether records exist. Accept a qualitative answer when no measurement exists, then label it accordingly. If they name another person's experience, mark it as secondhand and seek an appropriate original source before relying on it. At the end, read back the central decision in neutral language. A correction at this point is a research gain, not a failed interview or an interruption to the planned narrative.

Copyable decision worksheet

6. Turn the transcript into research notes

A transcript is a record, not a content outline. Segment it by decision and mark observations, interpretations, recommendations and unresolved questions. Preserve timecodes and source ownership. Group related segments without silently merging speakers into a synthetic consensus. If two experts disagree, note the conditions that may explain the difference. Build an outline from the reader's question rather than the order of the interview. Remove conversational repetition while keeping qualifications. AI can help label or retrieve segments, but an editor remains responsible for whether a summary changes the source's meaning or invents an agreement that never occurred.

Team fieldwork

Build a synthesis board with separate lanes for supported observations, interpretations, contradictions and new research questions. Move segments into these lanes while retaining the original IDs. Then write one provisional answer to the reader's question and list the segments that support it. If the answer depends on a segment in the contradictions lane, investigate rather than smoothing the conflict away. Draft an outline only after this pass. An article can explicitly explain why two approaches suit different conditions. That is often more useful than forcing a consensus. Save unused source material according to the agreed handling plan, not in a permanent miscellaneous folder.

Copyable decision worksheet

7. Verify claims before quote approval

Contributor approval and factual verification solve different problems. A contributor may approve a sentence that remains unsupported beyond their experience. Check technical capabilities against current official documentation and claims about outcomes against available records. Identify statistics that originated elsewhere and trace them to their original publisher. Avoid a chain of articles citing each other. For every material claim, choose retain, narrow, qualify, investigate or remove. Send contributors the intended context around a quote, not an isolated flattering sentence. Approval should cover the actual meaning being published, including any headline implication attributed to their expertise.

Team fieldwork

Prepare a verification packet in the order a reviewer needs to inspect it: proposed claim, surrounding paragraph, source excerpt, relevant artifact and specific uncertainty. Ask a narrow question such as whether the recommendation applies only after ownership has been assigned. Avoid asking whether the article is generally fine. Maintain separate statuses for factual verification, contributor attribution approval and editorial readiness. A contributor's enthusiastic response should not move all three statuses at once. If a reviewer changes a claim materially, return it to source checking. The revised sentence may introduce a new promise that the original evidence never addressed, even if the change sounds minor.

Three separate approvals

Copyable decision worksheet

8. Design readable outputs from complex material

Choose a structure that exposes the expert's reasoning. A troubleshooting guide needs symptoms and decision branches; an implementation account needs sequence and dependencies; an argument needs alternatives and evidence. Do not preserve the question-and-answer format merely because the research was an interview. Explain diagrams and use descriptive links. W3C WAI guidance on accessible writing and image alternatives is relevant to these outputs. For recorded excerpts, prepare accurate captions and a useful text version; a transcript still needs organization to work as a reading resource. Test whether a reader can distinguish the expert's claim from the editor's recommendation.

Team fieldwork

Choose a reader task for the finished output and remove the recording from the test environment. Can a colleague use the written guide to identify the decision, apply the criteria and recognize an exception? For video, check captions against the actual speech, including names and negations. For diagrams, write the conclusion and essential relationships in prose. Do not call a raw transcript an accessible edition merely because it contains text. It may still lack navigation, speaker context or meaningful structure. Keep the expert's approved meaning while adding editorial scaffolding that lets a reader use the material without attending the original conversation.

Copyable decision worksheet

9. Repurpose through question coverage

Create a derivative map from approved insights. One interview might support a diagnostic checklist, a comparison paragraph and a short explanation of an exception. It may support none of the short video clips you initially imagined. Map each derivative to one reader question, source segments, retained caveats and review owner. Keep a lineage identifier so a later correction reaches every asset. A short format should carry a bounded point, not a sensational version of the long one. Reuse is efficient only when the cost of verifying and maintaining derivatives remains lower than their actual usefulness.

Team fieldwork

Before producing a derivative, write its standalone proposition and the source segment that supports it. A forty-second clip may need a preceding sentence to make the referent clear. A checklist may need an assumption that was explained verbally earlier in the interview. Include these elements or choose another format. Keep one derivative register that records live locations and owners. When a source correction arrives, search by claim ID and contact each owner with the exact required change. Do not rely on the author remembering that a quote was also used in a webinar deck, a sales email and a partner article.

Derivative lineage map

Copyable decision worksheet

10. Measure research and production quality

Track research throughput separately from content response. Useful operational measures include unresolved material claims, review turnaround, contributor time and derivatives requiring correction. These are diagnostic measures, not competitive benchmarks. For reader use, define an observable task such as opening a worksheet or submitting an appropriate question. Analytics events require implementation checks; a designated key event does not become a verified qualified lead by naming alone. Ask sales or support what the material helped explain, preserving the wording and context of the answer. Avoid claiming that the interview caused pipeline simply because a related article was viewed.

Team fieldwork

Measure the research engine using a small set of operational questions. How much contributor time was required for one supported answer? Which claims still lack verification? Which approval step causes avoidable delay? Which derivatives needed correction because context was lost? These measures locate friction; they should not punish experts for careful review. Pair them with reader task feedback so fast production does not become the only goal. For website behavior, test one actual event and inspect its recorded parameters before launch. Treat a download as an observed action, then investigate whether the material helped. Naming it a high-intent event cannot establish intent.

Copyable decision worksheet

11. Diagnose interviews that produce little

Thin interviews usually fail at a specific stage. The contributor may lack direct experience, the question may be too broad, the artifact may be unavailable, or the editor may not probe decisions. Inspect those causes before increasing interview volume. If every answer sounds like company positioning, ask for an episode in which the recommended approach was not chosen. If the source cannot provide one, seek another contributor. If the interview is rich but the article is generic, inspect the synthesis step. More interviews will not repair a process that discards the distinctive evidence it already has.

Team fieldwork

Select a failed interview and replay a short section with the interviewer. Find the first moment when a concrete answer could have been pursued but was replaced by another broad question. Alternatively, identify where the interviewee lacked the relevant experience altogether. These require different fixes: coaching in the first case, recruitment changes in the second. Check whether necessary artifacts were requested in advance. If the contributor could not obtain them during the session, a short follow-up may be better than another full interview. Record the chosen repair in the next brief so the diagnosis changes behavior instead of becoming a retrospective complaint.

Copyable decision worksheet

12. Hand over a source library safely

The source library needs purpose limits as well as searchability. Separate original recordings, transcripts, approved excerpts and published derivatives. Record who may access each layer and the agreed handling of corrections, withdrawals and retention. Do not assume that permission for a written article covers a paid advertisement or an unrelated campaign. When a request changes, obtain an appropriate new agreement before reuse. The library owner maintains lineage and access; the editor owns interpretation; the contributor remains the point of contact for attribution questions. A backup must be able to identify all uses of one corrected statement without reading every transcript.

Team fieldwork

Test the library with a realistic correction request: the contributor says one approved capability description is now outdated. Ask the backup owner to locate the recording, the approved wording, the relevant permission and every published use. Time the exercise only to understand effort, not to set an arbitrary speed target. Identify inaccessible files and ambiguous versions. Fix those issues before scaling the interview program. Keep personal contact details and private material in appropriately controlled locations rather than copying them into public editorial boards. A searchable library is useful only if its access and purpose boundaries remain understandable to the people using it.

Copyable decision worksheet

13. Budget a fictional interview sprint

FICTIONAL B2B SaaS EXAMPLE: Clearstack provides access-review software. All numbers are assumptions for planning, not client outcomes. Assume three interviews of forty-five minutes, six hours of preparation, nine hours of verification and synthesis, and twelve hours of drafting and review. The total is twenty-nine and a quarter hours before distribution. The reader question concerns ownership of unresolved access exceptions. Recruit an administrator, a process owner and an implementation specialist. Plan one guide and one checklist first. Additional derivatives require unused capacity and approved source coverage. Do not describe the three conversations as a representative study of security teams.

Team fieldwork

For the fictional sprint, compare the source plan with available time before reserving publication dates. Three interviews require scheduling and preparation in addition to recording. Add verification, contributor review, editorial review and repair time. If capacity shrinks, retain the guide and checklist and remove optional derivative formats first. Ask the team which single question the first release must answer. A smaller complete answer is better than a large package with unresolved claims. Record actual hours during the pilot so the next estimate reflects your process rather than an imagined industry norm. None of these planning figures predicts audience response or commercial performance.

Fictional sprint effort

Copyable decision worksheet

14. Close the loop with contributors

Send contributors the published work and a concise note explaining how their approved input was used. Invite factual corrections without promising that every later preference change will rewrite an editorial conclusion. Review what the program learned, which questions remain and whether the next interview is necessary. Archive duplicate or unusable material according to the agreed handling plan. Google's helpful-content guidance emphasizes useful original contribution; interview volume alone does not establish it. The engine is working when a new editor can produce a supported answer from traceable material without repeatedly burdening the same expert or losing the limitations that made the answer trustworthy.

Team fieldwork

Close each interview record with a publication note, correction route and remaining research questions. Thank the contributor through the agreed channel when authorized by the project, and show the specific approved uses rather than a vague claim that their insights were everywhere. Review whether further interviewing is necessary or existing artifacts could answer the next question. Retire redundant derivatives that add maintenance without a distinct reader task. The next cycle should begin from a documented evidence gap. Avoid an engine that runs merely because there is a monthly interview slot to fill; recurring production must still justify the time it asks from specialists.

Copyable decision worksheet

Primary sources and application limits

• Google: Creating helpful, reliable, people-first content — checked 24 October 2026.
• Google: Guidance on generative AI content — checked 24 October 2026.
• Google: Publication dates — checked 24 October 2026.
• W3C WAI: Writing for web accessibility — checked 24 October 2026.
• W3C WAI: Page structure — checked 24 October 2026.
• W3C WAI: Images — checked 24 October 2026.
• Google Analytics: About key events — checked 24 October 2026.
• LinkedIn: Newsletters — checked 24 October 2026.