A product manager asks whether younger customers ever said the refill pack felt too expensive. Three people remember a study. Nobody can find it, so the team commissions a new one. A UX research repository exists to stop that from happening. It is where the output of surveys, trackers, interviews and thematic analysis lands, tagged so the next question starts from evidence instead of memory. This guide covers what to store, how to tag it, who should see what, and how to check answers that AI search pulls from it.
Key takeaways
- A research repository is a searchable home for studies, findings, data and verbatims, built so teams reuse what they already know.
- Store atomic insights, one finding per entry, each linked to the evidence behind it.
- Tag every item with date, sample, market, method and theme, using a short controlled list.
- Most repositories fail because they become dumping grounds with no tags and no owner.
- When AI search answers from your repository, open the sources before you repeat the answer.
What goes into a research repository?
A research repository holds more than finished reports, which are the least reusable part because findings are buried in slides. Store these five things:
- Studies: the project record, with the brief, questionnaire or discussion guide, and fieldwork dates.
- Reports: the final deliverable, as delivered.
- Atomic insights: single findings written as one or two sentences, each with links to its evidence.
- Verbatims and clips: the quotes and open-ended answers that support the insights.
- Metadata: date, sample size and type, market, method, audience and who ran it.
What is an atomic insight?
An atomic insight is the smallest finding that still makes sense on its own. “Younger lapsed buyers judge the refill on price per use, not pack price” is atomic. “Key themes from Q1 research” is not.
Each atomic insight should link to its evidence: a chart, verbatims or interview clips. That link lets someone trust a finding they did not produce. If you build insights from coded open-ends, the codes from your qualitative coding scheme are a natural starting point for themes.
Why do research repositories fail?
In our experience, repositories fail on habits, not technology.
- They become dumping grounds. Every report gets uploaded, nothing gets broken into findings, and search returns 40 slide decks.
- Nothing is tagged consistently. One person tags “price,” another “cost,” a third “value for money,” and none of them find each other’s work.
- Nobody owns it. Without one person curating tags and retiring stale findings, quality decays quickly.
How should you tag and structure an insights repository?
Use a short, controlled list for each tag type, because free-text tags split one idea across five spellings. Add values only when a real study needs one.
| Tag type | Example values | Why it matters |
|---|---|---|
| Theme | Price, packaging, trust, ease of use | The main way people search |
| Audience | 18–34, 35–54, lapsed buyers, B2B buyers | Findings rarely hold for everyone |
| Market | US, UK, Germany | Stops one market’s answer being used for another |
| Method | Tracker, concept test, interviews, usability test | Tells the reader how much weight a finding can bear |
| Date | Fieldwork month and year | Old findings may no longer hold |
| Sample | n = 1,200 online panel; 12 interviews | Size and source shape confidence |
| Status | Current, superseded, under review | Keeps stale findings out of answers |
Write a one-line definition for each theme, as you would for a codebook. “Price” might mean any mention of cost, value or price per use.
How do you answer a new question from past studies?
Here is a worked example. The question from the product manager is: “Is price a barrier to the refill pack for 18–34 buyers?” The repository holds three relevant studies, tagged as below. The numbers are illustrative.
| Study | Method | Market | Date | Sample | Theme tags |
|---|---|---|---|---|---|
| Brand tracker, wave 3 | Tracker with open-ends | US, UK | Oct 2025 | 1,200 | Price, trust, availability |
| Refill shopper interviews | In-depth interviews | US | Feb 2026 | 12 | Price, packaging, habit |
| Refill concept test | Monadic concept test | US | Apr 2026 | 400 | Price, purchase intent |
Filter on theme “Price” and audience “18–34,” and three pieces of evidence come back:
- Tracker open-ends: of 310 US respondents aged 18–34 who considered the refill but did not buy, 84 mentioned price. That is 84 ÷ 310 = 27.1%.
- Interviews: 5 of the 12 participants raised price per use without being asked.
- Concept test: in the 18–34 cell of 150, 51 chose the top two boxes for purchase intent at the tested price. That is 51 ÷ 150 = 34%.
In under an hour, you can say price is a real but not dominant barrier for this age group in the US, and say what you do not know. Only the tracker covers the UK, and its UK base of 18–34 non-buyers is too small to read on its own. So the new study, if you need one, is a short UK survey, not a full project.
Who should see what in a research repository?
Decide access rules before you load data, not after.
- Findings and reports: usually open to everyone in the company, because reuse is the point.
- Raw data and recordings: limited to researchers, especially where they contain personal data or identifiable faces and voices.
- Client or partner data: kept to the named team on that account, so one client never sees another’s work.
- Consent terms: stored with each study, so nobody reuses a clip beyond what participants agreed to.
UX research repository vs market research repository: how do they differ?
A UX research repository is mostly qualitative and project-based. A market research repository has to handle surveys and trackers that repeat over time.
| Factor | UX research repository | Market research repository |
|---|---|---|
| Typical studies | Usability tests, interviews, diary studies | Surveys, brand trackers, concept and ad tests |
| Main evidence | Clips, notes, observations | Crosstabs, open-ended answers, wave-on-wave trends |
| Sample sizes | Often 5 to 20 participants | Often hundreds to thousands per wave |
| Time structure | One-off projects | Repeating waves that must line up |
| Main risk | Insights scattered across projects | Comparing waves or markets that do not match |
For trackers, store the questionnaire version with every wave, because a wording change can look like a market change. When open-ends are part of the evidence, use the same approach you use to analyze open-ended survey responses, so tagged themes mean the same thing across waves.
How does AI search help, and how do you check its answers?
AI search lets people ask a repository a plain-language question and get a written answer, so it helps people who would never learn the taxonomy. Some tools let you ask questions of your own research in plain language and show the source with each answer.
The risk is an answer that sounds right but blends studies that should not be blended. Check every answer:
- Open each cited source. If an answer has no source, do not use it.
- Check date, market and sample. A 2023 UK finding should not answer a 2026 US question.
- Check counts against the claim. “Many respondents” might mean 4 of 12 interviews or 27% of 310.
- Look for missing evidence. Ask the same question a second way and see whether other studies appear.
The same checks apply when AI codes the open-ends in the first place, as our guide to AI thematic analysis explains.
What are the common mistakes?
- Storing reports but not findings. A deck is hard to search and harder to reuse.
- Letting tags multiply. If your theme list passes 40 values, merge before you add more.
- Dropping the evidence link. An insight without its source is an opinion, and people will treat it that way.
- Never marking findings as superseded. An overturned 2024 finding keeps appearing in answers until someone flags it.
Start your UX research repository small. Pick the last six studies your team ran, break each into five to ten atomic insights, and tag them with the seven types above. Then test the setup with one real question from a stakeholder. When that works, connect findings across studies and waves so teams reuse what they know instead of starting over.