EvalCommunity Tutorial

How to Use MAXQDA AI Assist for Thematic Coding in Monitoring, Evaluation, and International Development

A practical step-by-step workflow for using MAXQDA AI Assist to support qualitative coding, thematic analysis, code review, and evidence-based interpretation in M&E and international development research.

Summary

This tutorial explains how Monitoring, Evaluation, Accountability, and Learning professionals can use MAXQDA AI Assist to support thematic coding of interviews, focus group discussions, open-ended survey responses, field notes, case studies, and evaluation documents.

The tutorial focuses on practical use: preparing qualitative data, generating AI-assisted code suggestions, writing code memos, applying AI Coding, reviewing coded segments, developing subcodes, summarizing themes, and documenting AI use responsibly.

What You Will Learn

• How to prepare qualitative evaluation data before using AI Assist.
• How to use AI New Code Suggestions for selected text passages.
• How to write strong code memos that guide AI Coding.
• How to use AI Coding for documents, segments, and survey responses.
• How to review, accept, reject, or revise AI-generated coding suggestions.
• How to use AI Subcode Suggestions to refine a thematic code system.
• How to summarize coded segments and move from codes to themes.
• How to document AI use transparently in evaluation reports.

Authoritative Sources Used

This tutorial is based on MAXQDA’s official product pages and help documentation. Because software features change, users should verify current licensing, feature limits, and data protection details directly with MAXQDA.

• MAXQDA AI Assist product page
• What is AI Assist?
• AI New Code Suggestions
• AI Coding: Documents
• AI Subcode Suggestions
• AI Coding: Segments and Survey Responses
• AI Summary: Code Summary
• AI Chat with Coded Segments
• MAXQDA AI and Data Protection
• MAXQDA Pricing

AI-Assisted Thematic Coding Workflow

The diagram below shows the recommended workflow for using MAXQDA AI Assist in an evaluation project.

Key Point for Evaluators

AI Assist should support the evaluator, not replace the evaluator. Human judgment remains necessary for code definitions, interpretation, validation, ethical review, and final reporting.

Practical Preparation Steps

1. Collect all transcripts, open-ended survey responses, field notes, and qualitative documents in one secure project folder.
2. Remove direct identifiers where possible, including names, phone numbers, exact addresses, ID numbers, and unnecessary personal details.
3. Check whether participant consent allows AI-assisted processing or external processing of data.
4. Separate highly sensitive data, such as protection cases, safeguarding reports, health records, or gender-based violence disclosures.
5. Create a consistent file naming system, such as KII_HealthWorker_SiteA_01 or FGD_Women_DistrictB_02.
6. Prepare an evaluation question matrix before coding.
7. Decide whether your coding approach is deductive, inductive, or mixed.
8. Define who will review AI-generated suggestions and who has final responsibility for the analysis.

Important Data Protection Note

Do not use AI-assisted analysis on sensitive or restricted data unless it is permitted by informed consent, organizational policy, donor requirements, and applicable data protection regulations.

Project Setup Steps

1. Open MAXQDA and create a new project.
2. Name the project clearly, for example: Community Health Evaluation Qualitative Analysis 2026.
3. Import transcripts, focus group notes, open-ended survey responses, field notes, PDFs, or relevant documents.
4. Create document groups by stakeholder type, location, project component, or data source.
5. Add document variables where relevant, such as country, region, district, gender, age group, disability status, stakeholder type, or interview date.
6. Save a backup copy of the project before using AI-assisted features.

Example Document Groups

• Beneficiaries
• Community leaders
• Project staff
• Local government
• Implementing partners
• Donor representatives
• Field observation notes
• Open-ended survey responses

Manual Familiarization Steps

1. Read at least two or three transcripts before using AI Assist.
2. Write a short memo on first impressions.
3. Highlight repeated issues, unusual statements, contradictions, and local terms.
4. Identify possible deductive codes from the evaluation framework.
5. Identify possible inductive codes emerging from the data.
6. Flag sensitive content that may need special handling.

Practical Steps: Suggest New Codes

1. Open a transcript or document in the Document Browser.
2. Select a meaningful text passage.
3. Use the MAXQDA menu path: AI Assist > Text Selection > Suggest New Codes for Selected Text.
4. Alternatively, right-click the selected passage and choose AI Assist > Suggest New Codes for Text Selection.
5. Select the language for the suggestions.
6. Review the proposed code labels.
7. Select only codes that are analytically useful.
8. Apply the selected codes to the text passage.
9. Review the new codes in the Code System.
10. Rename vague or generic codes so they match the evaluation context.
11. Add a memo to each important code.

Example

Selected text: “Women told us that the clinic is open, but transport is expensive and some husbands do not allow them to travel alone.”

Possible useful codes:

• Transport-related financial barriers
• Household decision-making
• Gender norms affecting service access
• Access to health services

Quality warning: Do not accept every AI-suggested code. Some suggestions may be too broad, too generic, overlapping, or disconnected from the evaluation question.

Code Memo Template

• Code name: Short and clear label.
• Definition: What the code means.
• Include: What should be coded.
• Exclude: What should not be coded.
• Example include: A short example that fits the code.
• M&E relevance: Why the code matters for the evaluation question.

Example Code Memo

Code name: Transport-related financial barriers

Definition: Use this code when a respondent mentions transport cost, distance, travel affordability, fuel cost, road access, or inability to pay for transport as a barrier to using services.

Include: Mentions of bus fare, motorbike taxi cost, fuel cost, long distance, poor roads, seasonal flooding, or transport cost preventing access.

Exclude: General poverty unless the respondent directly connects it to transport or travel.

Example include: “The clinic is far, and I cannot pay for the motorbike every time.”

M&E relevance: This code helps assess whether service access barriers remain after project implementation.

Practical Steps: AI Coding for a Document

1. Create a code in your Code System.
2. Add a detailed code memo with definition, inclusion criteria, and exclusion criteria.
3. Open the document you want to analyze.
4. Go to AI Assist > AI Coding.
5. Select the document for analysis.
6. Select or drag the relevant code into the AI Coding dialog.
7. Review the code memo shown in the dialog.
8. Edit the memo if the criteria are unclear.
9. Start the AI Coding process.
10. Review each suggested coded segment before accepting it as evidence.

Practice rule: Do not run AI Coding across the full dataset before testing one or two documents and checking whether the suggestions match your analytical intent.

Review Checklist

• Does the coded segment match the code definition?
• Is the selected text too short, too long, or missing important context?
• Did the AI code a symptom instead of the underlying issue?
• Did the suggestion overlook gender, age, disability, location, or power dynamics?
• Does the segment help answer an evaluation question?
• Should the segment receive an additional code?
• Should the segment be deleted from this code?
• Should the code memo be revised to improve future suggestions?

Typical Human Review Actions

• Keep: The segment clearly matches the code.
• Edit boundary: The idea is right, but the selected text needs to be shorter or longer.
• Delete: The segment does not match the code.
• Recode: The segment belongs under a different code.
• Double-code: The segment fits more than one code.
• Memo: The segment raises an interpretation issue or possible finding.

Suggested Batch Strategy

• Batch 1: One transcript to test the code memo.
• Batch 2: Three to five transcripts across different stakeholder groups.
• Batch 3: Remaining documents only after quality review.
• Final review: Check coded segments before using them in findings.

Practical Steps: Suggest Subcodes

1. Choose a broad code that already has several coded segments.
2. Open the AI Assist option for subcode suggestions.
3. Select or drag the relevant code into the dialog.
4. Restrict the analysis to activated documents if you only want a specific subset.
5. Review the suggested subcodes in the selection dialog.
6. Select only subcodes that are meaningful for your evaluation.
7. Add selected subcodes to the Code System.
8. Review and edit the explanatory notes saved in the code memos.
9. Rename subcodes so they use clear evaluation language.

Example: Broad Code to Subcodes

Broad code: Barriers to service access

• Transport cost
• Distance to services
• Gender norms
• Waiting time
• Language barriers
• Informal fees
• Lack of disability access
• Fear of stigma

Practical Review Process

1. Open the Smart Coding Tool or Survey Analysis workspace.
2. Select the coded segments or survey responses you want to review.
3. Select the codes that AI Assist should consider.
4. Check that each selected code has a clear code memo.
5. Run the AI-assisted code assignment process.
6. Review each AI suggestion and explanation.
7. Accept suggestions that fit the data and code definition.
8. Reject suggestions that are too general, contextually wrong, or unsupported.
9. Bulk-edit only after reviewing a sample manually.

Practical Steps: Summarize Coded Segments

1. Choose a code with enough coded segments to summarize.
2. Use the AI Assist summary option for coded segments or code summaries.
3. Select summary language and length where available.
4. Generate the summary.
5. Read the summary alongside the original coded segments.
6. Delete or revise unsupported statements.
7. Use the summary as a draft analytical memo, not as a final finding.

Example Questions for M&E Analysis

• What are the main barriers mentioned in these coded segments?
• Do women and men describe this issue differently?
• What evidence suggests that the intervention was relevant to local needs?
• What unintended outcomes appear in the data?
• Which stakeholder group is most critical of the program?
• What implementation conditions appear to influence success?
• What are the strongest quotes supporting this theme?
• What evidence gaps remain in these documents?

Verification rule: Every AI-generated answer should be checked against the original document, coded segment, or source quote before being used in an evaluation report.

Difference Between Codes and Themes

Code: A label attached to a specific idea in the data.

Theme: A broader pattern of meaning that explains something important about the evaluation question.

Example

Codes: Transport cost, long distance, poor roads, lack of referral transport, seasonal flooding.

Theme: Physical and financial access barriers continue to limit service utilization for remote communities.

Evaluation interpretation: The program may have improved service availability, but access remains unequal because practical barriers still prevent some groups from using services.

Theme Validation Checklist

• Is the theme supported by multiple coded segments?
• Does the theme appear across more than one respondent?
• Does the theme appear across more than one stakeholder group?
• Are marginalized voices visible in the analysis?
• Are there negative cases or contradictory examples?
• Is the interpretation grounded in direct evidence?
• Can the team trace the finding back to original data?
• Has at least one human evaluator reviewed the AI-assisted coding?
• Has the team avoided claiming more than the data supports?

Finding Structure

• Evaluation question: What question does this finding answer?
• Theme: What pattern emerged from the coded data?
• Evidence: Which coded segments and quotes support the theme?
• Interpretation: What does the evidence mean?
• Implication: Why does this matter for the program?
• Recommendation: What should the program or evaluation team do next?

Example Evaluation Finding

Finding: Although health services were available in the project area, women in remote communities continued to face access barriers linked to transport cost, distance, and household decision-making.

Interpretation: The intervention improved service availability, but did not fully address the social and economic barriers that affect actual service use.

Recommendation: Future programming should combine service delivery support with transport solutions, community outreach, and gender-sensitive household engagement.

Responsible Use Checklist

• Transparency: Document where AI Assist was used in the analysis.
• Human oversight: Ensure final codes, themes, and findings are reviewed by human evaluators.
• Data minimization: Use only the data needed for the AI-assisted task.
• Consent: Check whether participant consent allows AI-assisted analysis.
• Confidentiality: Remove or mask direct identifiers where possible.
• Fairness: Check whether minority voices are overlooked or misinterpreted.
• Traceability: Keep a clear link between findings, themes, codes, coded segments, and original data.
• Validation: Review AI outputs against original evidence before reporting.

Sample AI Use Statement

MAXQDA AI Assist was used to support selected stages of qualitative analysis, including preliminary code suggestions, AI-assisted coding, subcode development, and summaries of coded segments. All AI-generated suggestions were reviewed, edited, accepted, or rejected by the evaluation team. Final code definitions, thematic interpretation, findings, and recommendations were developed by human evaluators and checked against original source data. No finding was generated solely by AI.

Exercise: Code Community Feedback from a Health Program

Use a small dataset of interview excerpts, focus group notes, or open-ended survey responses. The goal is to move from raw qualitative data to codes, themes, findings, and recommendations.

1. Import 10 to 20 qualitative responses into MAXQDA.
2. Read the full dataset manually.
3. Create three deductive codes from the evaluation questions.
4. Select one text passage and use AI Assist to suggest new codes.
5. Review and accept only relevant code suggestions.
6. Create code memos with definitions, inclusion criteria, and exclusion criteria.
7. Use AI Coding on one document for one well-defined code.
8. Review each AI-suggested coded segment.
9. Use AI Subcode Suggestions for one broad code.
10. Summarize coded segments for one code.
11. Develop two themes from the coded data.
12. Write one evaluation finding and one recommendation.
13. Write a short AI use statement.

Can MAXQDA AI Assist replace manual thematic coding?

No. AI Assist can support coding, summaries, and code suggestions, but final analytical decisions should remain with the evaluator or research team.

What data can evaluators analyze with MAXQDA AI Assist?

Evaluators can use it to support analysis of interviews, focus groups, open-ended survey responses, field notes, evaluation reports, case studies, and other text-based qualitative data, depending on project needs and data protection requirements.

What is the most important step before using AI Coding?

The most important step is writing clear code memos. AI Coding uses code names and code definitions to guide its suggestions, so vague code definitions can lead to weak results.

How should AI-generated coded segments be reviewed?

Each AI-generated coded segment should be reviewed against the code definition, the original text, the evaluation question, and the broader context of the data.

Should AI use be disclosed in evaluation reports?

Yes. If AI Assist was used to support coding, summaries, or analysis, the report should include a short AI use statement explaining how AI was used and how human reviewers validated the outputs.

• The dataset was reviewed for privacy and consent risks.
• The evaluation questions guided the code system.
• At least one transcript was read manually before AI was used.
• AI-generated code suggestions were reviewed before acceptance.
• Code memos include clear definitions and inclusion/exclusion criteria.
• AI Coding was tested on a small sample before scaling.
• AI-generated coded segments were reviewed by a human evaluator.
• Theme development was based on evidence, not only AI summaries.
• Findings were checked against original data.
• Contradictions and minority views were considered.
• AI use was documented transparently in the report.

Conclusion

MAXQDA AI Assist can help M&E and international development professionals work more efficiently with qualitative data. It can support code suggestions, AI Coding, subcode development, summaries, and analytical exploration.

The strongest use of AI Assist is not to replace the evaluator. The strongest use is to help evaluators organize evidence, question patterns, review coded segments, and develop more transparent findings while keeping human judgment, context, and ethics at the center.