An expert interview can contain direct experience, remembered estimates, second-hand reports, and the interviewer's own inference—all in the same paragraph. A polished summary can hide those differences.
Use a claims-and-sources table before quoting or publishing. Put each claim beside the exact source, the speaker's basis for knowing it, the level of support, and the next verification step.
Start with the claims that could change a decision. Those deserve stronger evidence than background colour or a useful lead.
Begin with permission and purpose
Before recording or reusing an interview, confirm what the participant agreed to: the purpose, who may access the material, whether quotations can be attributed, and how long the notes or recording may be retained. The GOV.UK Service Manual guidance on research notes and recordings stresses informed consent and using research material only for the agreed purpose.
Do not put sensitive personal information into a claims table merely because it appeared in the conversation. Use the minimum source detail needed for the permitted workflow, and keep restricted evidence in its approved location rather than copying it into a widely shared document.
Separate the statement from the claim
An interview statement is what the person said. A claim is the specific proposition a reader might rely on. One sentence can contain several claims, so copy the source statement once and split its testable parts into separate rows.
“Our rooftop beds usually need less watering after we add a thick mulch layer, especially in July.”
This sentence contains at least three elements: a practice (adding mulch), an observed outcome (less watering), and a condition (hot July weather). The table should not silently turn that experience into a universal percentage reduction.
Use fields that expose uncertainty
| Field | Question it answers |
|---|---|
| Claim ID | How will this statement be referenced without rewriting it? |
| Claim | What exact proposition might a reader rely on? |
| Source statement | What was actually said or observed? |
| Evidence state | Is this direct experience, a cited source, a recollection, or an inference? |
| Source link | Where can an authorized reviewer inspect the supporting material? |
| Scope | Which place, period, population, or condition does the claim cover? |
| Uncertainty | What is unknown, estimated, disputed, or missing? |
| Next check | What evidence would confirm, narrow, or reject the claim? |
The GOV.UK guide to analysing a research session recommends first capturing what was seen or heard, then determining findings and actions. The same separation helps here: preserve the observation before adding interpretation.
Choose evidence states that are descriptive
- Direct experience: the expert describes work they personally performed or observed.
- Source supplied: the expert points to a document, dataset, image, or other inspectable source.
- Source recalled: the expert remembers a source but has not yet supplied enough detail to check it.
- Second-hand: the expert reports what another person or organization said.
- Interviewer inference: the claim was derived during analysis and was not stated by the expert.
- Conflicting evidence: another permitted source supports a materially different account.
These states describe provenance; they do not automatically label a person “right” or “wrong.” Verification still depends on the claim, its scope, and the evidence available.
A fictional urban-gardening example
The following interview and names are fictional. They demonstrate the method without presenting a real customer or research finding.
| ID | Claim | Evidence state | Uncertainty | Next check |
|---|---|---|---|---|
| CLM-01 | The team added mulch to three rooftop beds. | Direct experience | Depth and material not recorded. | Confirm the maintenance log and a dated photo. |
| CLM-02 | Those beds needed watering less often afterward. | Direct experience | No baseline frequency or weather adjustment. | Compare watering logs before and after the change. |
| CLM-03 | The largest difference occurred in July. | Source recalled | Month remembered; records not yet linked. | Check July weather and irrigation records. |
| CLM-04 | A nearby garden reported a similar result. | Second-hand | Site, method, and result are unspecified. | Ask for the original contact or report. |
| CLM-05 | Mulch may explain the lower watering frequency. | Interviewer inference | Other changes may have contributed. | Check plant mix, shade, rainfall, and maintenance changes. |
Notice that CLM-05 remains an inference. The table does not rewrite “after we add mulch” as proof that mulch alone caused the change.
Keep quotations, paraphrases, and conclusions distinct
Use quotation marks only for wording checked against the permitted source. Label a paraphrase as a paraphrase. Put analytical conclusions in their own field or row. If the interview was summarized with automated assistance, run a separate summary-verification workflow before treating any wording as exact.
When the expert discusses a product demonstration or technical capability, use the stricter claims-versus-evidence workflow for product-demo notes. “The expert said it” is provenance, not independent confirmation of performance.
Review the table in two passes
- Source pass: check that each claim points to the correct statement, note location, or permitted recording segment.
- Evidence pass: decide whether the current evidence supports the wording and scope.
- Boundary pass: remove material outside consent, access, or sharing limits.
- Follow-up pass: assign the smallest useful next check, an owner, and a review date.
A claim can remain in the table even when it is unresolved. The honest status is often more useful than a premature yes-or-no judgment.
Copy-ready claims-and-sources template
| ID | Claim | Source statement | Evidence state | Source | Scope | Uncertainty | Next check | Owner/date |
|---|---|---|---|---|---|---|---|---|
| CLM-[number] | [one proposition] | [quote or paraphrase] | [state] | [permitted link] | [where/when/who] | [what is missing] | [specific check] | [name + date] |
Try it now: choose one permitted expert interview, extract five claims, and give each row an evidence state, uncertainty note, and next check. Do not upgrade a recollection or inference into a fact merely to complete the table.
