How to use Claude for research synthesis

Sources: customer interview transcripts, competitor blog posts, support tickets, sales calls, survey open-text responses.

For PDFs: use Adobe export or online PDF-to-text. Save as .txt files.

For audio transcripts: use Otter, Rev, or Whisper. Output as plain text.

For web pages: copy as plain text or use a "save as markdown" browser extension.

Aim for 5-20 source documents totaling 50K-150K words. Claude handles this comfortably in one conversation.

Clean up: remove headers, footers, page numbers, navigation chrome. Pure content only.

Open a new Claude chat. Paste in this structure:

"I have [N] source documents I need to analyze. Below are all of them, separated by ---DOCUMENT N---."

"---DOCUMENT 1: Customer Interview - Jane (CMO at $5M SaaS)---"

[paste transcript]

"---DOCUMENT 2: Customer Interview - Mike (Head of Growth at $20M DTC)---"

[paste transcript]

...continue for all sources.

End with: "I will now ask analytical questions. Reference specific documents by number in your answers. Acknowledged."

Claude acknowledges and is ready.

Start broad: "Identify the top 5 themes that appear in 4+ of the interviews. For each theme, cite the documents and the exact quotes."

Claude returns themes with citations. Validate the citations — open the referenced documents to verify Claude is not hallucinating.

Drill into specific themes: "Expand on Theme 2 (pricing objections). What are the 3-4 sub-patterns within this theme? Who said what?"

Synthesis with quotes: "Generate 5 customer quotes that I could use to support the claim that [insight] in a blog post."

Always ask for citations. Without them, Claude can hallucinate insights that look right but are not in the data.

Collect 5-10 of the top-ranking blog posts for your target topic.

Paste them into Claude with the delimiter pattern.

Prompt: "I have pasted 8 top-ranking competitor blog posts for the keyword '[keyword].' Identify: (1) consensus structural patterns (what every piece covers), (2) common angles (positioning approaches), (3) GAPS — topics or angles none cover well."

Gaps are where you differentiate. Build your piece around them.

Follow-up: "For each gap, suggest specific H2 sections I could write. Include word count estimates."

After thematic synthesis, prompt: "Based on all the customer interviews, build 2-3 distinct personas. For each: name (illustrative), role + company size + industry, top 3 pains in their exact words, top 3 objections in their exact words, language patterns (what words they use vs avoid), where they consume content."

Claude generates persona profiles grounded in the actual interview data.

These personas are 10x better than ones you write from intuition — they are derived from real conversations.

Save the personas in a doc. Reuse for content briefs, ad copy briefs, sales positioning.

Qualitative themes (from interviews) should align with quantitative trends (from data).

Prompt: "Based on the themes identified, what should we expect to see in our analytics data if these themes are accurate? Generate 3-5 testable hypotheses."

Validate the hypotheses against your actual data. If themes do not show up in analytics, the qualitative may be skewed.

If themes do show up, you have triangulated evidence — much stronger basis for decision-making.

Prompt: "Format the full synthesis as a research report. Sections: (1) executive summary, (2) top 5 themes with quotes, (3) 2-3 personas, (4) competitor gaps, (5) recommendations for content + positioning. Markdown format."

Claude formats the report. Copy to Notion/Google Docs.

For team sharing: also generate a 1-pager summary. "Distill the full report into a single-page brief for the marketing team."

Reuse the chat as the audit trail — anyone questioning the synthesis can see the prompts and citations.