Contextual inquiry: watch real local and family historians research, then fix where they never find us

Date: 2026-10-31. Type: field research skill (reach-focused). Distinct from think-aloud buy-path tests (small-sample-usability-tests-buy-path-5-users.md), cognitive walkthrough, heuristic evaluation and PURE: those check our pages; this watches the reader's own research routine, on their own devices, before they ever reach us.

What it is (primary sources, read 31 Oct 2026)

• Developed by Hugh Beyer and Karen Holtzblatt (Contextual Design). A small sample of people is watched doing the real activity in their own setting, and asked about it while they do it. NN/g: people "attempt to summarize their processes" in interviews and leave out reasoning and habits; "users can easily talk about what they are doing and why when they are doing it."
• NN/g example: three specialists all described a copy-paste routine in interviews; watching them showed extra unreported steps (cross-checking another tool, habitual saving). Interviews alone would have missed both.
• Master/apprentice model: the participant is the expert, the researcher is the apprentice.
• 4 principles (NN/g, Beyer and Holtzblatt): Context (their place and tools, not a lab), Partnership (they steer, you follow), Interpretation (check your reading with them on the spot: "so you went to the catalogue first because...?"), Focus (a stated research focus decides what you watch).
• 4-part session (NN/g): primer/intro, transition to the work, observe and ask (contextual interview), wrap-up where you summarise back and they correct you.
• Remote version (NN/g "Remote Contextual Inquiry: Lessons Learned"): video call with screen share works; keep each session to about 2-3 hours including intro and wrap-up; plan the tech check; have a second note-taker.
• When NOT to use it (NN/g, plainly): "not especially useful for targeted design tasks such as redesigning an ecommerce product page or testing a newsletter signup form". So do NOT use it to tweak /ebooks/. Use it for the complex, habit-heavy activity: how a local or family historian actually researches a farm, a parish or a surname.
• Sample size: NN/g answers in three parts (users per visit, visits per day, total visits, which depends on the project). For a narrow focus like ours, 4-6 sessions is a practical start; stop when new sessions stop showing new routes (saturation). This number is our judgement, not an NN/g figure.

Why it matters for reach

Real outside /ebooks/ views are about 45 a day. We guess where readers look (search, Welsh Newspapers, archive catalogues, Facebook groups, family history forums, library catalogues). Watching 4-6 real researchers shows the actual first places they go, the words they type, and the moment they give up. Every place they look and we are absent is a reach gap with a named fix.

How to run it (step by step)

1. Write the focus in one line: "How does someone researching a Vale of Glamorgan farm, parish or family history find sources, and where do they look first?"
2. Recruit 4-6 people who already do this as a hobby or for study: local history society members, family history society members, a heritage studies student, a volunteer at a local archive. Recruitment goes through the project's owner or existing public channels (talk audiences, society newsletters), with consent. No cold DMs to strangers.
3. Ask them to bring a real current question of their own (not ours). Their question, their device, their bookmarks.
4. Session (60-120 min, video call with screen share or in person): - Intro 5 min: purpose, consent to record, "you are the expert, I am learning how you work". - Transition: "Show me what you would do next on your question." - Observe and ask: silence most of the time; short in-the-moment questions only ("What made you pick that site?", "What did you type?", "What would you do if nothing came up?"). Never show or mention our site during this part. - Interpretation check: say back what you think you saw; let them correct it. - Wrap-up 10 min, then ONLY now: "Have you come across this site before?" Watch them find it from scratch (search, not a link). Note the query, the result position, and whether they would trust it.
5. Notes per session, same sheet every time: sites visited in order, exact queries typed, tools (Welsh Newspapers Online, archive catalogues, maps, Facebook groups, forums, library catalogues, AI chat), breakdowns (dead ends, paywalls, give-ups), workarounds, trust cues they mention, what they save or share and where.
6. Interpret within 24 h: one affinity pass across sessions. Group into "first places they look", "words they use", "where they get stuck", "what makes a source trustworthy".
7. Turn every pattern into one reach action with an owner: a place we are absent (list there through the matching skill), a word they use that our pages do not (add it to titles/headings honestly), a dead end we can answer (a free page that answers it with a link to the e-book).

Risks

• Researcher effect: people behave better when watched. Ask for their real current question and keep quiet.
• Small sample: patterns, not percentages. Never report "60% of researchers..." from 5 sessions.
• Do not pitch during the session; it turns observation into a sales call and spoils the data.
• Privacy: store notes without names; no recordings kept beyond analysis; nothing personal goes into public files.

How we apply it to Cedwyn

• Focus: how local/family historians research Llandough, Leckwith, Cogan and the Vale, and Welsh families in Patagonia.
• Recruit 4-6 through existing talk audiences and society channels already in our skills (u3a-speaker-circuit-local-history.md, family-history-society-journals-forums.md, civic-local-history-society-talks-penarth.md). Owner approves any outreach; agents never contact people directly.
• Agents prepare the session guide and note sheet, then code the notes into the four groups above.
• Output: a ranked "absent where they look" list feeding existing reach skills (worldcat-library-hub-discover-catalogue-visibility.md, people-also-ask-paa-mining-full-answers-place-pages.md, topic-clusters-place-pages-local-history.md), plus their exact query words for page titles.
• Measure: for each fix, outside /ebooks/ views from that source over the following 4 weeks against the ~45/day baseline.
• Do not use it on the buy page itself; that is what the 5-user test and PURE skills are for.

Sources

• nngroup.com (read 31 Oct 2026: principles, session structure, when not to use, interview vs observation example)
• nngroup.com (read: 2-3 h session rule of thumb, remote set-up)
• nngroup.com (summary read: three-part answer to how many users)
• Beyer and Holtzblatt, Contextual Design (Morgan Kaufmann; 2nd ed. 2017), the four principles (via NN/g, book not read this run)

Raw Markdown: contextual-inquiry-local-historians-research-routes.md · version and changes: changelog · machine index: agents.json