How do you write case studies with AI without the customer sounding like someone else?

Start from raw material, not a summary. Give the model the interview transcript, emails, support tickets or call notes, and tell it that quotes must be copied word for word from that material. If it can only rearrange the customer's sentences rather than write new ones, the voice survives.

Split the job in two. First ask for an extraction: the strongest verbatim quotes, the numbers with their dates, and the before-and-after in the customer's own phrasing. Then write the story around those pieces. Drafting straight from a transcript is where the model starts paraphrasing and smoothing.

Set a few hard rules in the prompt: no numbers that are not in the source, every metric shown with its time window, and a short list of words the customer would never say. Then run a cleanup pass that strips generic filler from the narrative parts, not from the quotes.

Finally, send the draft to the customer for approval. They will spot a quote that does not sound like them faster than anyone, and most companies need that sign-off anyway.

Listings mentioned

• Customer story builder · skill by gooseworks-ai
  Turns transcripts, survey answers, Slack quotes and tickets into a structured case study with pull-quotes and metric callouts.
• Case study writer prompt · prompt by growthack88
  Interviews you first, then writes with strict rules: every number keeps its time window and no number you did not give appears.
• Make Your Writing Sound Human Again · skill by Hardik Pandya (https://hvpandya.com)
  Strips filler, clichés and buzzwords from the narrative so it reads less like default AI copy.

Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.