Recommendation
Use ChatGPT as the writing engine and Zotero (Better BibTeX or CSL JSON export) as your canonical reference library. Export the exact references you want cited from Zotero, paste or upload the export into ChatGPT, and use a short prompt template that forces ChatGPT to insert consistent citation tokens and build a reference list. Finish by validating every citation and producing a formatted bibliography with Zotero or pandoc.

When to choose this vs alternatives
- Best for: fast, repeatable market-report drafts where you control the source list and need copy with inline citations (Author, Year) and a reference list. Works well for one-author or small teams.
- Avoid if: you need legal-grade citations, verbatim archival sources without verification, or fully automated enterprise ingestion (you’ll need more tooling and QA).
- Consider Claude if your draft requires extremely long-context reasoning across hundreds of pages of source metadata — Claude scales well with long-context tasks.

Decision criteria (pick what matters most)
- Output quality required: use manual verification and Zotero formatting for high quality.
- Team size & workflow: small teams can paste exports into ChatGPT; larger teams should automate exports (Better BibTeX) and use CI to render citations.
- Budget & skill: basic (free) — manual export + ChatGPT web UI; advanced ($) — Better BibTeX, pandoc, or scripting.

Practical prompt + toolchain (copy-paste)
1) From Zotero: select items → Right-click → Export Items → choose “CSL JSON” (or Better BibTeX for persistent citekeys). Save and open the file.
2) In ChatGPT: paste the CSL JSON file content (or upload as a file), then send this prompt (replace STYLE with APA 7 / Chicago / IEEE):

Prompt template (paste after your CSL JSON):
"You are a professional market researcher. Use only the reference entries provided in the attached CSL JSON. Produce a 800–1200 word draft section titled ‘Market Drivers and Forecast’ in clear business prose. Insert inline citation tokens in-text using this format: (AuthorLastName, YYYY) and also include a citation token [@citekey] alongside it. Make sure each token matches an entry in the library. At the end, produce a numbered reference list formatted in STYLE (APA 7). Include DOI or URL and access date when available. Do not invent references or facts. If a claim cannot be supported by the provided library, mark it with [UNSUPPORTED]."

3) Post-process: Use Zotero (Create Bibliography from Items) or pandoc+citeproc to convert tokens into a fully formatted bibliography if you used citekeys/pandoc syntax.

Practical checklist before publishing
- Exported references include DOI/URL and correct metadata.
- ChatGPT output uses only provided tokens; search for [UNSUPPORTED].
- Cross-check each inline citation against the exported CSL JSON item (author, year, title).
- Generate final formatted bibliography from Zotero or pandoc (don’t rely on ChatGPT formatting alone).
- Run a factual QA pass: verify numbers, quotes, and market figures against primary sources.

Quick tips
- Use Better BibTeX if you want stable citekeys (e.g., Smith2021_product). Paste those keys in the prompt and ask ChatGPT to use them as [@Smith2021_product].
- Keep the library small and focused per draft to minimize hallucinations.

If you want, I can produce a ready-to-use prompt tuned for APA 7 with an example CSL JSON mapping and a short post-processing command line for pandoc/Zotero.

Compare ChatGPT and Gemini