--- title: "AI Memory for Interview Transcripts: Keep Evidence, Summary, and Decisions Separate" description: "Preserve consent, recordings, transcripts, timestamped evidence, analyst inferences, approved findings, corrections, and retention." canonical: "https://scalewithsearch.com/articles/ai-memory-interview-transcripts-summaries" date: "2026-08-26" modified: "2026-09-19" --- ## Site navigation - [Scale With Search](https://scalewithsearch.com/) - Real estate - Real estate - [Real estate](https://scalewithsearch.com/for/real-estate) - Work - Start here - [Send your brief](https://scalewithsearch.com/work#send-your-brief) - [Prepare your six-question brief](https://scalewithsearch.com/work#prepare-your-six-question-brief) - Build - [Site build, content library with SEO, signal desk](https://scalewithsearch.com/work) - For your business - Trades and home services - [Auto body and collision shops](https://scalewithsearch.com/for/auto-body-and-collision-shops) - [Foundation and home repair contractors](https://scalewithsearch.com/for/foundation-and-home-repair) - [Garage door and fencing contractors](https://scalewithsearch.com/for/garage-door-and-fencing-contractors) - [HVAC contractors](https://scalewithsearch.com/for/hvac-contractors) - [Janitorial and commercial cleaning companies](https://scalewithsearch.com/for/janitorial-and-commercial-cleaning) - [Locksmiths](https://scalewithsearch.com/for/locksmiths) - [Moving companies](https://scalewithsearch.com/for/moving-companies) - [Pest control companies](https://scalewithsearch.com/for/pest-control-companies) - [Plumbing and electrical contractors](https://scalewithsearch.com/for/plumbing-and-electrical-contractors) - [Restoration and water or fire damage companies](https://scalewithsearch.com/for/restoration-and-water-fire-damage) - [Roofing companies](https://scalewithsearch.com/for/roofing-companies) - [Towing companies](https://scalewithsearch.com/for/towing-companies) - [Tree services and landscaping companies](https://scalewithsearch.com/for/tree-services-and-landscaping) - [Solar installers](https://scalewithsearch.com/for/solar-installation) - [General contractors](https://scalewithsearch.com/for/general-contractors-and-construction) - [Paving, concrete, and flooring contractors](https://scalewithsearch.com/for/paving) - Practices and professional services - [Bookkeeping and tax practices](https://scalewithsearch.com/for/bookkeeping-and-tax-practices) - [Dental practices](https://scalewithsearch.com/for/dental-practices) - [Family and criminal defense law firms](https://scalewithsearch.com/for/family-and-criminal-defense-law-firms) - [Med spas and aesthetics practices](https://scalewithsearch.com/for/med-spas-and-aesthetics) - [Personal injury law firms](https://scalewithsearch.com/for/personal-injury-law-firms) - [Veterinary clinics](https://scalewithsearch.com/for/veterinary-clinics) - [Gyms and fitness studios](https://scalewithsearch.com/for/fitness) - [Therapy and outpatient health practices](https://scalewithsearch.com/for/therapy-and-outpatient-health) - [Medical billing companies](https://scalewithsearch.com/for/medical-billing) - [Insurance agencies](https://scalewithsearch.com/for/insurance-agencies) - [Financial advisors](https://scalewithsearch.com/for/financial-advisors) - [Property management companies](https://scalewithsearch.com/for/property-management) - [Recruiting and staffing agencies](https://scalewithsearch.com/for/recruiting-and-staffing) - [Architects and interior designers](https://scalewithsearch.com/for/architects-and-interior-designers) - [Logistics and supply chain companies](https://scalewithsearch.com/for/logistics-and-supply-chain) - Agencies, MSPs, and manufacturing - [IT and managed service providers](https://scalewithsearch.com/for/it-and-managed-service-providers) - [Machine shops and precision manufacturers](https://scalewithsearch.com/for/machine-shops-and-precision-manufacturing) - [Marketing agencies and freelancers](https://scalewithsearch.com/for/marketing-agencies-and-freelancers) - [SEO agencies and consultants](https://scalewithsearch.com/for/seo-agencies-and-consultants) - [Small manufacturers and fabricators](https://scalewithsearch.com/for/small-manufacturers-and-fabricators) - Restaurants, shops, studios, and nonprofits - [Restaurants and hospitality businesses](https://scalewithsearch.com/for/restaurants-and-hospitality) - [Retail stores and ecommerce sellers](https://scalewithsearch.com/for/retail-and-ecommerce) - [Photographers, event planners, and travel agents](https://scalewithsearch.com/for/photographers) - [Churches and nonprofits](https://scalewithsearch.com/for/churches-and-nonprofits) - [All industries](https://scalewithsearch.com/for/) - Learn - For your office - [Office job guides](https://scalewithsearch.com/guides/) - [Browser calculators](https://scalewithsearch.com/tools/) - Start here - [How it works](https://scalewithsearch.com/how-it-works) - [Free Starter Kit](https://scalewithsearch.com/kit/business-memory-starter-kit.zip) - [Synthetic specimen](https://scalewithsearch.com/specimen/working-session-specimen.zip) - Guides - [The Complete Guide to Business Memory for AI Agents](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [The Complete Small-Business Guide to AI Agent Governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [The Complete Guide to Leaving Vendor AI Memory](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - Articles by cluster - [Business memory](https://scalewithsearch.com/articles/business-memory-for-ai-agents-guide) - [Agent governance](https://scalewithsearch.com/articles/ai-agent-governance-guide-small-business) - [Migration and ownership](https://scalewithsearch.com/articles/leaving-vendor-ai-memory-guide) - For machines - [llms.txt](https://scalewithsearch.com/llms.txt) - [llms-full.txt](https://scalewithsearch.com/llms-full.txt) - [Machine view](https://scalewithsearch.com/?view=machine) - Company - Evidence - [Proof](https://scalewithsearch.com/proof) - Company - [About](https://scalewithsearch.com/about) # AI Memory for Interview Transcripts: Keep Evidence, Summary, and Decisions Separate. An AI summary turns one participant's tentative comment into a firm customer need. The product roadmap cites the summary. Three months later, nobody can find the timestamp, identify the speaker, or tell whether the transcript was corrected after review. The summary was useful. It was never sufficient evidence. Interview memory should keep the recording, transcript, evidence excerpt, analyst inference, approved finding, and business decision as separate records. The links between them let a reviewer move from a roadmap claim back to the source without pretending every spoken sentence is accurate or authoritative. For meeting decisions rather than research findings, use the [meeting-decision register](/articles/ai-memory-meeting-notes-decisions). It tracks approval and supersession without treating a proposal as a finding. ## Keep each evidence layer distinct An interview produces several artifacts with different roles. The recording is the closest retained evidence when consent and policy allow it. The transcript makes that evidence searchable. Speaker labels and timestamps locate statements. A summary compresses the discussion. An analyst note interprets patterns. An approved finding records what the research owner accepts. A decision records what the business will do. Do not overwrite one layer with another. A polished summary can hide uncertainty, disagreement, tone, and omitted questions. A corrected transcript should preserve the correction receipt and trigger downstream review. OpenAI's transcription API documents diarized transcript events with speaker labels and segment start and end times. These fields support traceability, but an application still must map a generic speaker label to a verified participant. [OpenAI: Audio API](https://platform.openai.com/docs/api-reference/audio) The [chat-history warning](/articles/chat-history-is-not-business-memory) applies here. A searchable conversation archive is evidence. It becomes business memory only after the organization defines source roles, promotion, corrections, and retrieval rules. ## Record consent and permitted use Consent is not a footnote attached after recording. Store who consented, when, how, for which purpose, and which artifacts may be retained or shared. Distinguish permission to record from permission to use quotations, train a model, share with a client, retain beyond analysis, or include a participant's name. The applicable rule depends on the interview type, agreement, location, and organizational policy. Do not infer permission from the presence of a recording file. If consent evidence is missing, quarantine the artifact and block downstream use until the authorized owner resolves it. Apply minimum access. A researcher may need audio and full transcript. A product manager may need approved findings with evidence locators. A client-facing report may need anonymized findings only. ## Inspect `interview-evidence-register.csv` Give every finding a route back to evidence and every inference a named reviewer. ```csv interview_id,consent_id,recording_path,transcript_path,speaker,timestamp,evidence,analyst_inference,finding_status,finding_id,dissent,retention_on INT-073,CON-044,recordings/INT-073.wav,transcripts/INT-073.v2.json,P03,00:18:42,"I compare the export before renewal",Export matters during renewal,approved,FND-021,,2027-02-26 INT-074,CON-045,recordings/INT-074.wav,transcripts/INT-074.v1.json,P04,00:11:08,"Maybe once a quarter",Quarterly workflow demand,review,FND-022,P05 observed monthly use,2027-02-26 ``` Avoid storing long verbatim passages in a broad register. Use the shortest evidence needed for review and point to the governed transcript. Where policy requires anonymization, keep identity mapping in a separate restricted record. Use `finding_status` values such as `candidate`, `review`, `approved`, `rejected`, and `superseded`. An analyst inference can exist without an approved finding. A finding can exist without authorizing a roadmap decision. ## Link findings to timestamps and speaker certainty Every material finding should name at least one interview, speaker label, transcript version, and timestamp. Important findings usually need multiple independent interviews or a clear statement that they reflect one participant. Record speaker confidence. Diarization may separate voices without assigning real identities. Crosstalk, poor audio, interpreters, and shared rooms can reduce certainty. A reviewer should be able to play the relevant segment when the claim matters. Store transcript version and content hash. If a human fixes a name, number, or product term, append a correction and regenerate affected summaries. Do not silently edit a sentence after it has supported a finding. The [correction-log template](/articles/ai-correction-log-template) shows how to name the old value, corrected value, source, approver, affected outputs, and retest state. ## Separate evidence from inference Participants report experiences. Analysts interpret what those experiences mean. Keep both. Write evidence as an attributed observation. Write inference as a proposed explanation. Write a finding as an accepted synthesis. Write a decision as an authorized business choice. For example, “three participants could not find export settings” is an observation if the sessions support it. “Export anxiety causes churn” is an inference requiring more evidence. “Add an export test to renewal research” is a decision. Require the system to label each sentence it produces. It should not convert hedged language into certainty or merge dissent into a false consensus. Preserve counterexamples beside the dominant pattern. ## Build retrieval around the research question Do not retrieve all transcript fragments with similar words. Start with a research question, population, date range, consent scope, and approved evidence status. If the evidence is in Claude conversations, read [Search Claude Chat History and Keep Verified Findings.](/articles/search-claude-chat-history-full-text) Return balanced evidence. Include supporting excerpts, dissent, missing segments, and participant counts. Deduplicate repeated statements from the same interview so one talkative participant does not appear as several independent observations. Filter by permission before semantic ranking. Restricted recruiting interviews should not appear in a customer-research query because the topic is similar. When the agent cannot find timestamped evidence, it should return `unsupported_finding`. It must not use a summary sentence as proof of the summary. ## Correct a transcript and rerun downstream work Use one deliberately incorrect term in a synthetic transcript. Let the first summary and candidate finding reflect the error. Then submit a reviewed correction. The correction process should: 1. preserve the original transcript version; 2. append the correction with reviewer and reason; 3. create a new transcript version and hash; 4. identify summaries and findings that cited the changed segment; 5. regenerate or block affected outputs; 6. retain an acceptance receipt for the new result. Do not recalculate every historical report without a decision. Mark affected artifacts and follow the organization's correction policy. ## Run the interview-memory test list Create four fictional interviews: 1. Include one clear supporting statement with verified speaker identity. 2. Include one tentative statement that uses “maybe.” 3. Include one dissenting participant. 4. Include one incorrect transcript term. 5. Restrict one interview from broad product-team access. 6. Ask for a finding and its evidence chain. 7. Ask the system to claim unanimous support. 8. Correct the transcript term and rerun. 9. Delete one recording under policy while retaining the permitted receipt. 10. Export the approved findings and reconstruct them with another tool. Pass requires timestamped evidence, preserved hedging, visible dissent, permission enforcement, correction propagation, and a usable export. The system must refuse the unanimous claim and identify missing authority. The [retention and deletion policy guide](/articles/ai-memory-retention-and-deletion-policy) helps assign separate schedules to recordings, transcripts, summaries, findings, and decisions. ## Require an evidence receipt Every research output should list the research question, included interviews, excluded interviews, transcript versions, consent scope, evidence IDs, participant count, dissent, corrections, model and prompt version when relevant, output hash, and reviewer state. The receipt does not prove the finding is wise. It proves which evidence and rules produced it. That is enough for a reviewer to reproduce the path and challenge it. When a model changes, rerun the same fixture. Word choice may differ. Evidence identity, permission behavior, hedging, dissent, and approved findings should remain stable. ## Approval and stopping boundary The workflow may transcribe consented recordings, label speakers provisionally, prepare summaries, link evidence, propose findings, detect dissent, apply approved corrections, and write receipts. It stops before recording without resolved consent, identifying an anonymous participant, sharing restricted material, treating a model summary as approved evidence, promoting a finding, making a roadmap decision, or deleting source artifacts. The research owner approves findings. The business owner approves decisions. The records owner approves retention and deletion. If consent, speaker identity, transcript confidence, or evidence location is unresolved, label the gap and block the affected finding. ## Sources - [OpenAI: Audio API](https://platform.openai.com/docs/api-reference/audio) - [NIST: AI Risk Management Framework 1.0](https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10) - [Granola: Document customer research interviews](https://www.granola.ai/blog/document-customer-research-interviews-product-insights) ## Questions about AI Memory for Interview Transcripts: Keep Evidence, Summary, and Decisions Separate ### What should you verify when AI summarizes interview transcripts? Check every material finding and quotation against the transcript, timestamp, and speaker certainty. Keep the recording, transcript, observation, analyst inference, approved finding, correction, and retention state as separate evidence layers. ### Can AI perform thematic analysis of user interviews without human review? Use AI as a bounded analysis aid, not as the final evidence owner. Require the research question, source-only rules, timestamped support, unknown states, and a reviewer who can trace each accepted finding back to the transcript. ### How do you keep AI interview findings from turning one participant into users? Maintain the chain from raw quote or clip to observation, interpretation, and recommendation, with participant and timestamp attached. Do not promote an inference or cross-interview theme until a reviewer confirms the supporting evidence and permitted use. ## Related: Owned Memory - [Decide whether your business needs retrieval-augmented generation](/articles/what-is-retrieval-augmented-generation) - [AI Memory for Content Research: Keep Sources, Claims, and Rejected Angles Between Sessions](/articles/ai-memory-content-research) - [AI Memory for URLs: Store the Page, the Claim, and Why It Matters](/articles/ai-memory-for-urls-bookmarks) ---- ```text .|########||. .|########||. .|########||. |##||. .||##|. |##||. .||##|. |##||. .||##|. |#|. .|#|. |#|. .|#|. |#|. .|#|. |#| |#| |#| |#| |#| |#| .#| |#. .#| |#. .#| |#. .#. .#| .#. .#| .#. .#| |#. .#| |#. .#| |#. .#| |# ...... #| |# ...... #| |# ...... #| .# ||#########| #| .# ||#########| #| .# ||#########| #| .#. |######||######|. .#. .#. |######||######|. .#. .#. |######||######|. .#. #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# #. .##|###|##|#|######| .# || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || || |##|#||||||||||||#||#| || # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. # |#||||||||||||||||||||#| #. || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # # .#||||||||||||||||||||||||#| # || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || || ....|||#||||||##|#|||#|#||##||||....|. || #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# #. . ....|# ....#||| ||| .#||| ....#. .# # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. # . ||||#| .#####|| . .#| .#||| ||||# #. .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || .| ||||| #. |#|||||. || #. |#||. .||||| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || || |.... #. ....|#. |. ...|. ....|| || #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# #. .||||||##||||||##||####|||||||#||||||#| .# # .||####|###########################|. # # .||####|###########################|. # # .||####|###########################|. # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#||||||.#.|| # |. ..# #| #|||||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||||| .. |# ## |#| ...#|#|||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#|||#|# .# .#| #| ##|.#..#|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # .#||||||############|######|||||#| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # # |...||#||||||#||||#|#||||||#|||||#||| |.| # #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# #. .. ||||# .|||##|. |#| .|| || .|||#. #| # .# || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || | ...|| ...##| | #| .|. |. ##### .. .| || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || || ||||| || ||||#| . |. |. |#. ||||| |#| || || .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. .# |.....#|....|#.||||.|||##.|#|....||.#||.#. #. #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# #.|||||||#############################| ||| .# || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || || ##||#||#||||||#||#||#||#||##. || # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # # .#|||||||||#||#|||||||||||#| # || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || || |#||||||||||||||||||||||#| || .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. .# |#||||||||||||||||||||#| #. || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || || |#||||||||||||||||||#| || #. .######|#|#########| .# #. .######|#|#########| .# #. .######|#|#########| .# .#. |#####||||#####| .#. .#. |#####||||#####| .#. .#. |#####||||#####| .#. |# ||########|| #. |# ||########|| #. |# ||########|| #. |# ...... #| |# ...... #| |# ...... #| |#. .#| |#. .#| |#. .#| |#. .#. |#. .#. |#. .#. .#| |#. .#| |#. .#| |#. |#| |#| |#| |#| |#| |#| .|#|. .|#|. .|#|. .|#|. .|#|. .|#|. |##||. .||##| |##||. .||##| |##||. .||##| .||########||. .||########||. .||########||. Scale With Search 2026 [scalewithsearch.com](https://scalewithsearch.com) ```