If the team will run this research again next month, the method matters as much as the answer. Claude Code is useful here because the inclusion rules, evidence and decisions can live beside one another. We’ll use outdoor dialogue recording as our example, with our own test audio supplying the eventual lesson.
Kabo, an AI agent for creator research, collects the public references. Claude Code records the method and separates source claims from the experiment our team will film.
Prepare the collection rubric and source archive
[ ] brief.md — Teaching question and the decision the research should support. [ ] method.md — Dates, language, format and inclusion/exclusion criteria. [ ] test-assets.md — Our recordings, conditions and known gaps. [ ] CLAUDE.md — Stable attribution and evidence rules for future sessions. [ ] research/YYYY-MM-DD/ — A new snapshot for this run, preserving earlier notes.
2. Connect Kabo, then check the project context
Open the content project, install Kabo for Claude Code and complete sign-in. For a direct server connection, follow the setup guide.
3. Ask for the collection plan without changing the files
An outdoor-filmmaking channel is preparing a guide to recording understandable dialogue in windy locations. The team wants a reusable research method rather than a one-off list of microphone reviews.
Have Claude Code read brief.md and method.md, then propose its research pass in the conversation. Check whether the search terms privilege product reviews over practical recording methods. Agree on the inclusion rules first. This prevents a large source file that answers a different question well.
Read brief.md and method.md. Propose a research pass before editing files. Show search terms and inclusion rules. We want sources documenting [method], not simply product recommendations. Explain how you’ll record unavailable transcripts or incomplete samples.
4. Write a dated source index through Kabo
Collect relevant public candidates and available supporting evidence. Keep the actual retrieval date and the publication window in the project. For each source, note what was available: metadata, transcript or completed analysis. If a source lacks usable spoken or visual evidence, do not let its title become a description of its test method.
Keep one reusable spoken passage for our own field tests. If each recording uses a different sentence and location, the resulting edit can sound like a comparison without actually isolating the question.
The MCP transcript workflow covers text retrieval and attributed notes when the spoken explanation is the useful evidence.
Use Kabo to collect the agreed YouTube references. Save a dated source index with URLs, publication dates, retrieval date and evidence type. Write source-notes.md separately from our-experiment.md. Identify which recording conditions are documented and which comparisons remain unmatched. End with one lesson structure and the recordings we still need to make.
5. Separate research notes from the planned experiment
Ask Claude Code to write one file summarizing source claims and another describing your own test. A reference can suggest a question; it cannot establish your microphone settings or the weather during your shoot. Require the final report to show which conclusions still need your own recordings.
6. Refresh the question that remains unresolved
When the team notices that two references use different recording conditions, ask for a narrowly scoped follow-up. Save it as a new note and explain whether it changes the lesson structure. Do not overwrite the earlier evidence or silently upgrade a weak comparison into a settled technical conclusion.
A useful dossier entry
Our outdoor-filmmaking example turns source research into a test we can record ourselves:
- Lesson question: What changes when we move from a sheltered park to a beach?
- Reference note: Attach the returned public video URL and supported time range.
- Own test: Record the same spoken passage in both locations.
- Unresolved: Which reference comparisons used matched recording settings?
- Teaching decision: Show conditions before discussing equipment.
7. Give the editor a lesson structure, not an archive dump
Write a concise handoff with the opening comparison, necessary conditions and source notes. Keep the full index available, but lead with what the editor needs to demonstrate. Record a clear reference passage in your own footage so the lesson can be judged by what the audience hears.
The shortlist is mostly buying advice. Keep only sources that document recording conditions and find a few method-focused examples. Explain what the narrower set can support.
Give the editor the dossier and the next experiment
Use Descript for a spoken rough cut if a text-oriented edit suits your process. Keep listening comparisons in context and verify the final audio against your original recordings; the research dossier should not become a substitute for hearing the test.
Attach the accepted listening comparison to the script project. Keep a broader channel-positioning study separate unless that is the decision the editor still needs.
Worked examples are illustrative. Product screenshots show existing Kabo web results, not a run of the example in this guide. Use your own materials and the evidence returned in your task.