Both can do it; Claude handles long context and careful analysis better, ChatGPT fits iterative pipelines.

Practical recipe:
- Chunk 10–15 pages (or ~5–10k tokens). Include chunk ID and source pages.
- For each chunk ask: “Extract claims, exact quotes, page numbers, and supporting sentences” output as JSON.
- Synthesize: supply all JSONs, ask model to merge, tag each claim with source IDs and confidence.
- To reduce hallucination: require verbatim evidence for every claim; reject unsupported claims.

Compare: Compare Claude and ChatGPT