How to build a quote bank for qualitative reporting

What is a quote bank in qualitative research?

A quote bank is a structured table of selected quotes, excerpts and passages that may support a claim, theme, finding or report section, with each row linked back to its source. It is not a folder of transcripts and not a highlights document of every interesting line.

A quote bank sits between raw qualitative material and the written report. It moves a team from "someone said this somewhere" to "this quote supports this claim, comes from this source at this point, sits under this theme and has been checked for use". That makes it one practical part of keeping every claim in a report traceable to its source.

A quote bank can hold several kinds of evidence:

One rule matters more than the rest: never mix exact quotes, cleaned quotes and paraphrases without labelling them. Writers and reviewers need to know which one they are looking at.

When is a quote bank worth building?

A quote bank is worth building when a project has more qualitative evidence than one writer can hold in their head. The larger or more sensitive the material, the more it matters.

It is especially useful when the project has:

• many interviews, case studies, focus groups or field notes
• more than one analyst or writer
• findings that a funder, client, technical panel or government reviewer will question
• material spread across documents, PDFs, spreadsheets, transcripts and forms
• sensitive material that needs anonymising or careful handling
• AI-assisted coding or extraction that still needs checking
• a deadline that leaves no time to reread everything during drafting

The method is the one used in the UNICEF Zambia household case studies: interview material coded to one theme standard, with a quote behind every claim.

What fields should a quote bank include?

Each row should tell a writer or reviewer what the quote says, where it came from, what it supports and whether it is ready to use. These are the fields that do that work.

For a smaller project, keep at least the minimum set below. Without these fields a bank may still help drafting, but it will not stand up to review.

Quote ID | Source ID | Participant or case reference | Locator | Exact quote | Evidence type | Theme | Claim supported | Context note | Sensitivity flag | Review status

How do you build a quote bank, step by step?

Build it in a fixed order: agree the report's questions, register the sources, then extract, locate, tag, link and review each quote before writers use it.

1. Start from the report questions, not the quotes

Agree the research questions, evaluation questions, main themes or report sections before extraction begins. Ask what the report has to explain, which findings will need direct evidence, who will review it, whether direct quotes are allowed and whether any material is sensitive. Copying every line that sounds interesting creates a large table that is hard to use. If the report must explain barriers to service access, the bank should help the writer find evidence for those barriers; if it must compare groups or places, it needs the fields that make that comparison possible.

2. Register every source and give it an ID

List every interview, focus group, case study, field note file, survey comment set and submission in a source register before any quote is taken, and give each a stable Source ID. The fields and ID conventions are set out in how to build a source register. Do not leave IDs for later: retrofitting them during drafting is where quotes lose their sources.

3. Extract quotes that serve a report need

Extract quotes that may support a claim, explain a theme, show a tension or help a writer build a section. If a quote does not serve a report need, mark it as background or leave it out.

4. Add the locator at the moment of extraction

Record the page, paragraph, line, timestamp or row as you extract. A quote without a locator is hard to verify later, especially when several people work from similar documents.

5. Tag each quote with one agreed theme

Use the agreed theme list. If a new theme appears, add it to the list openly rather than creating silent variations. "Access barriers", "barrier to access", "service access issue" and "difficulty accessing services" should be one label with sub-themes for detail.

6. Link each quote to a claim or finding

A theme says what the quote is about. A claim link says what the quote can support. For example: theme, service access; sub-theme, transport; claim supported, distance and transport costs limited access for rural families; finding link, 3.2; report section, barriers to service uptake.

7. Add review status and sensitivity flags

Before a quote reaches a writer, check that it is accurate, relevant and safe to use. A quote can be analytically strong and still unsuitable for publication because it identifies someone or contains sensitive personal information.

8. Test whether each finding has enough support

For each draft finding, check whether there are enough supporting quotes, whether they come from the right groups or places, whether the claim is broader than the evidence and whether contradicting evidence exists. This stops weak claims before they reach review.

9. Give writers filtered views

Writers should not need the whole raw bank. Create views by report section, theme, finding, respondent group, quote role and approval status.

10. Keep usage status current

As the report changes, record whether each quote was used, cut, replaced or moved. This makes review comments and final checks easier to answer.

11. Hand the bank over with the report

At the end of the project, keep the quote bank with the source register, coding notes, data dictionary, review notes and anonymisation decisions. It is the record of how the report's qualitative claims were built.

How many quotes does a claim need?

Every important claim should point to at least one quote, excerpt or coded evidence entry, but one quote rarely proves a finding on its own. A quote usually illustrates a wider pattern; the pattern has to come from the evidence as a whole.

Take a draft finding such as: stakeholders felt that referral pathways were unclear, especially outside urban centres. The quote bank should let the writer check:

• which sources support it
• whether those sources span more than one group or location
• whether the quotes are about referral pathways or a different access issue
• whether any evidence challenges or narrows the claim

If the bank holds one weak quote from one respondent, the claim needs softer wording. If several sources from different groups point the same way, the finding is easier to defend. GRADE-CERQual, the approach used to rate confidence in qualitative evidence synthesis findings, treats adequacy of data as one of its four components alongside methodological limitations, coherence and relevance (Lewin et al., Implementation Science, 2018). A report team does not need a formal assessment to ask the same question: is there enough here to support the wording?

Quotes are one kind of evidence. Where a claim also rests on documents, survey results or administrative data, record it in a source-linked evidence table, which links every type of evidence item to claims and findings.

What role does each quote play?

Not every quote has the same job, and marking the role stops writers overclaiming. A vivid quote may be memorable and still be an outlier or a minority view.

If a finding says "most participants reported this issue", one strong line is not enough. The team needs evidence across participants to justify the wording, and a single vivid quote should be framed as one person's experience.

Which review statuses should a quote bank use?

Use statuses that separate checking from approval, and internal use from public use. Quote review belongs inside the evidence workflow, not only at final review, when a section may already be built around a quote nobody approved.

Sensitivity deserves its own check. In South Africa, the Protection of Personal Information Act treats some categories as special personal information, including health or sex life, religious or philosophical beliefs, race or ethnic origin and political persuasion, and restricts how they may be processed (Protection of Personal Information Act 4 of 2013, section 26). Interview quotes often touch these subjects. Check what participants consented to before a quote moves to "approved for public use".

When a quote needs anonymising, replace identifiers rather than blanking them out. UK Data Service guidance on anonymising research data advises consistent pseudonyms across the whole project and marking each replacement clearly, for example with square brackets (UK Data Service, 2024). Record every anonymisation decision in the bank, not in email threads or comments.

Where can AI help with a quote bank?

AI can help with first-pass work, but every quote and locator must be checked against the original source before use. Treat AI output as suggestions.

AI can help with:

• first-pass quote extraction and short labels
• grouping quotes by theme and finding similar quotes across sources
• identifying possible finding links and contradictions
• preparing draft writer packs by report section

AI should not:

• invent quotes, or turn paraphrases into direct quotes
• rewrite evidence without marking it
• remove source links
• decide whether a quote proves a finding, or make final decisions on sensitive material
• receive confidential material without the right data controls

A verified quote bank is also a cleaner base for later AI-assisted retrieval than a folder of transcripts, because every row already has a source, locator, theme and review status. The order matters: verify the bank first, then let AI help retrieve, compare or summarise from it. For turning approved quotes into written findings, see how to turn interviews and case studies into report-ready findings. Where the volume of material calls for it, a quote bank and evidence database, built and handed over, keeps the structure in place after the project ends.

Common mistakes

• Copying too many quotes. A bank that becomes a second transcript folder does not help writers.
• Losing Source IDs. Without them nobody can check where a quote came from or whether it was used correctly.
• Mixing exact quotes and paraphrases without labels, which misleads readers and reviewers.
• Using one quote as proof of a whole finding. Check the wider evidence before choosing the wording.
• Ignoring contradictory evidence. Capture tension and disagreement, not only neat support for the emerging argument.
• Skipping the sensitivity check until after drafting.
• Tagging themes but not claims. Quotes by theme are helpful; quotes linked to claims, findings and sections are what writers need.
• Letting AI summarise without source checks.
• Building the bank after the draft. Then it is a repair tool. Build it early enough to shape the findings.