Market and Audience Research with ChatGPT

#01Which research jobs suit ChatGPT?

OpenAI's data analysis article supports spreadsheets, PDFs and text files, and asks for descriptive headers and one record per row. ChatGPT also accepts supported audio files, so you can transcribe and summarise interview recordings; OpenAI notes transcription quality may vary across languages.

#02How do you mine reviews?

REVIEW MINING
Attached: [n] reviews of [our product / Competitor A] as CSV
with columns rating, date, text.
1) Classify each review into themes (max 8). Show counts.
2) For each theme give three verbatim quotes, with row numbers.
3) List the top objections and the words customers use for them.
4) Compare themes in 1-2 star vs 4-5 star reviews.
Don't paraphrase quotes.

Then check: open three of the quoted rows yourself. If a quote doesn't match, ask ChatGPT to re-run with stricter instructions.

#03How do you synthesise interviews?

INTERVIEW SYNTHESIS
Attached are [n] interview transcripts (names removed).
For each, extract: role, trigger that started their search,
alternatives considered, the job they hired us for, and the
exact phrase they used for their main pain.
Then summarise patterns across interviews in a table, with
how many interviews mention each.

#04How do you build personas that aren't fiction?

Ask ChatGPT to build personas only from the synthesis, and to show the evidence.

EVIDENCE-BASED PERSONAS
Using only the synthesis above, draft 2-3 personas. For each:
goals, pains, triggers, objections, channels, and the quotes
that support each line. Mark anything inferred as [INFERRED].

Our view: the [INFERRED] tag is the most useful line in that prompt. It shows you which parts of a persona are research and which are ChatGPT's guess.

#05Where does public market data come in?

For category size, trends or competitor facts, use search or deep research (lesson 10) and ask for citations. Open each cited source and check the number is there, current and about the right market. Never quote a market-size figure from ChatGPT without its source.

#06Worked example: Ledgerline's review mining

Ledgerline is a fictional accounts-payable tool. Details are illustrative.

Ledgerline exported 400 of its own reviews and 300 of Competitor B's from a review site. Personal names were stripped first. ChatGPT found that "approval speed" led Ledgerline's positive reviews, while "setup took weeks" was the top complaint. Competitor B's reviews praised onboarding.

The team checked ten quotes against the CSV; one had been merged from two reviews, so they reran with "quote one row only". The finding fed a new onboarding page and a testimonial request to customers who'd praised setup.

#07What should you check before the next lesson?

• Export one review or survey set and remove personal data.
• Run the review-mining prompt and spot-check five quotes.
• Draft one persona with [INFERRED] tags and decide what to research next.
• The Claude track's version is Claude customer review mining.

Next: Content with ChatGPT: blogs, landing pages and ads. Or return to the ChatGPT for Marketers track.

#08Common questions

Can ChatGPT replace customer interviews? No. It can synthesise interviews, reviews and surveys you collect, but it has no access to your customers. Personas built without real input describe an average buyer, not yours.

How many reviews can I paste into ChatGPT? It depends on your plan context window and the file size. For large sets, upload a CSV and use data analysis, which handles more rows than pasting.

Is it safe to upload customer data? Remove names, emails and other personal data first, and use a Business workspace or switch off model training on a personal plan.