How to write client case studies with AI (without sounding fake)

I have a testimonial on my site that says I’m “a pleasure to work with”. It has never, as far as I can tell, influenced a single buying decision. I also have a case study describing how a client went from spending two days a month on a reporting process to about twenty minutes, including the part where my first attempt made things worse. That one gets referenced on sales calls.

The difference isn’t quality of writing. It’s that one is a compliment and the other is evidence.

Why case studies outperform testimonials

A testimonial says someone liked you. A case study shows a person with a problem you might recognise, and what happened when they addressed it.

Prospects are running one calculation: will this work for my situation. A testimonial gives them nothing to reason with. A case study lets them see themselves in the before state, which is the entire persuasive mechanism.

This is also why generic case studies fail. “We increased their revenue” gives a reader nothing to recognise. “They were manually copying orders from email into a spreadsheet every morning” gives them a mirror.

The interview is the actual work

Everything good in a case study comes from the conversation. If you skip it and write from memory, you get your version of events, which is always tidier and less interesting than what really happened.

Twenty minutes on a call, recorded, with these questions:

• What was happening before? Walk me through a typical week. You want the mundane detail. The tedium is what readers recognise.
• What made you decide to actually do something about it? The trigger moment is the most useful sentence in the whole interview.
• What did you try before this? Establishes that the easy answers didn’t work.
• What were you worried about? This becomes your objection-handling section, in your client’s words rather than your guesses.
• What’s different now, specifically? Push for concrete before-and-after.
• What surprised you? Almost always produces the best quote.
• What would you tell someone considering this? The closing line, written for you.

Record it. AI meeting notes tools that actually save you the follow-up covers the tooling, and having a transcript is what makes the AI-assisted drafting work at all.

The structure that works

Same skeleton every time:

Headline result. One line, specific, up top. “Cut monthly reporting from two days to twenty minutes.”

The client, briefly. Who they are and why a reader might be like them. Two sentences.

The before. The problem, in concrete daily terms. This section should make the right reader wince in recognition.

What we tried and what happened. The work. Include at least one thing that didn’t go smoothly.

The result. Numbers where you have them, specific qualitative change where you don’t.

A quote. Their words, unedited apart from cutting filler.

What this means for you. One short paragraph and a next step.

Where AI helps, and where it ruins it

Good jobs for AI: turning a rambling transcript into a structured first draft, tightening your own overlong paragraphs, generating three headline options from the result, and rewriting the whole thing for a different format like a one-page PDF or a LinkedIn post.

Bad jobs for AI: anything requiring specificity it doesn’t have. Ask a model to write the “before” section from a one-line brief and it will produce beautifully written mush about inefficiency and lost productivity. That mush is exactly what makes case studies unreadable.

The prompt shape that works:

Here’s a transcript of a client interview. Draft a case study using this structure [paste structure]. Use the client’s own phrasing wherever possible and quote them directly at least twice. Do not add any detail, number or claim that isn’t in the transcript. Where the transcript is vague, flag it rather than filling the gap.

That last sentence is the important one. It turns the model into a drafter rather than an inventor, and the flags tell you what to go back and ask about.

For anything where you’re producing these regularly and want consistent voice, Jasper with a saved brand voice removes the re-prompting overhead. My Jasper review covers whether that’s worth a subscription, and best AI writing tools for creators has the wider comparison.

The uncomfortable truth that makes it credible

Include one thing that went wrong, or one limitation, or one thing the client still isn’t happy about.

This feels counterproductive and it’s the single biggest credibility lever available. A case study where everything went perfectly reads as marketing, because readers know projects don’t go like that. An admission that the first approach failed and you changed course tells them two things: that this is a real account, and that you handle problems.

Mine says my initial solution was too complicated and the client didn’t use it for three weeks. Every prospect who has mentioned that case study has mentioned that part.

Getting approval without friction

The step that kills case studies. Make it easy:

Send the draft with a clear note: here’s what I’ve written, here’s the one number I’d like to include, tell me anything you want removed. Give them a deadline, because open-ended requests sit forever.

Offer an anonymised version as a fallback. “A fintech client with a six-person team” is less powerful than a name and still far better than nothing, and it converts a no into a yes surprisingly often.

Ask at the moment of success, not six months later. Right after a good result, people are delighted and generous. In February they’ve forgotten and moved on.

My take

Write three case studies rather than thirty testimonials. Interview properly, use the client’s own words, include the part that went wrong, and put a specific number in the headline.

AI makes the drafting fast, and that’s genuinely useful, because the reason most freelancers have no case studies is that writing one felt like a day’s work. From a recorded twenty-minute interview, a good draft takes about half an hour.

But keep your hands on the specifics. The generic, well-written version of a case study is worth almost nothing, and it’s exactly what you get if you hand a model a summary and ask it to make something impressive. The awkward, specific, slightly-too-detailed version is the one that closes deals. For the wider client-facing stack, AI proposal and contract tools for freelancers covers what happens once the case study has done its job.