How-To Guide

How to Use AI to Write Client Case Studies (Without Inventing Results)

A step-by-step workflow for turning project notes and a 20-minute client call into a credible case study — with a copy-paste prompt, an approval process, and the mistakes that make AI-written proof backfire.

B Biztrategy Published 3 November 2026 · 8 min read
A diverse job interview scenario in a modern office setting with a laptop and documents on the table.

Ask any consultant, agency owner or freelancer what wins them new work and the answer is nearly always the same: proof. A prospect who reads how you helped a business like theirs — the problem, what you did, what changed — is far closer to signing than one who has only read your services page. And yet most small firms have two or three case studies at best, usually written years ago, usually about a client who has since moved on.

The reason is not a lack of results. It is that writing a case study feels like a half-day job that never makes it to the top of the list. This is exactly the kind of task AI is good at shortening — provided you use it to structure and draft real material, not to invent it. This guide gives you a five-step workflow, a drafting prompt you can copy, and the mistakes that turn AI-written proof into a liability.

Why most small firms never publish case studies

When we talk to owners about this, the blockers are remarkably consistent:

  • "I don't have time to write it up." A good case study takes 3–5 hours from blank page to published, plus chasing the client for approval.
  • "I don't have the numbers." The project went well, but nobody measured the before and after.
  • "The client might not agree." Asking feels awkward, and confidentiality clauses make some owners assume the answer is no.
  • "Mine read like brochures." Self-written case studies drift into marketing copy, and prospects can tell.

AI solves the first and fourth problems almost completely. It helps with the second, by showing you which questions to ask. It does nothing for the third — that still needs a human conversation, and it is where trust is built.

What AI should and should not do in a case study

A case study is evidence. The moment a prospect suspects a figure was embellished, every other claim on your site loses value. So draw a hard line before you open ChatGPT or Claude.

Let AI do: turn a messy interview transcript into a clear narrative; propose a structure and headline; pull out the strongest client quotes; tighten the language; adapt the finished piece into a LinkedIn post, a one-page PDF and a proposal paragraph.

Never let AI do: supply numbers, percentages or timeframes that are not in your source material; write quotes and attribute them to the client; "estimate" results to make the story more impressive; fill gaps with plausible-sounding detail.

Language models are built to produce fluent, convincing text, and they will happily write "reduced admin time by 40%" if the story seems to need a number. That is why the prompt later in this guide explicitly forbids it — and why a human fact-check is a required step, not an optional one. If you want the wider picture on this risk, our guide to preventing AI hallucinations in client work covers the checks in more depth.

The five-step workflow

This process takes most teams 60–90 minutes of human time per case study, down from a typical 4–5 hours. The time saved is almost entirely in drafting and editing; the conversation with the client stays fully human.

Step 1: Gather the raw material (15 minutes)

Before you speak to the client, collect what you already have in one folder: the original brief or proposal, the scope of work, any reports or dashboards you delivered, before-and-after screenshots, and relevant emails where the client said something positive. Paste the key extracts into a single document. This becomes the factual backbone the AI is allowed to use — and nothing else.

If you never captured a baseline, ask the client for one simple before-and-after figure during the interview: hours per week spent on a task, enquiries per month, days to close the books. One honest, specific number beats three vague ones.

Step 2: Run a 20-minute client interview

Book a short video call and record it with permission. Most AI notetakers will produce a transcript automatically — just make sure the client knows and agrees, as we explain in our piece on AI notetaker consent. Five questions are enough:

  1. What was going on in the business before we started working together?
  2. What had you already tried, and why didn't it work?
  3. What made you decide to work with us?
  4. What is different now — ideally with a number or a concrete example?
  5. What would you tell another business owner in your position?

Question four gives you the result. Question five gives you the quote prospects remember. Let the client talk; the transcript is raw material, not the final piece.

Step 3: Draft with a structured prompt (10 minutes)

Give the AI the transcript, your Step 1 notes and the prompt in the next section. Ask for one draft, then one revision with your feedback. Resist the urge to regenerate ten times — the quality ceiling is set by your source material, not by the number of attempts.

Step 4: Fact-check, anonymise and get sign-off (20–30 minutes)

Print the draft or open it side by side with the transcript. Highlight every number, date, tool name and quote, and confirm each one against the source. Delete anything you cannot trace. Then decide on naming: if the client prefers not to be named, use a descriptor such as "a 12-person accountancy practice in Manchester" and remove details that would identify them. Under GDPR, the transcript itself is personal data — store it securely and delete it once the case study is approved.

Finally, send the client the draft with a short note: "Here's the write-up of our chat — please correct anything that's not right and let me know if you're happy for us to publish it with your name and logo." Written approval protects both of you.

Step 5: Repurpose into five assets (15 minutes)

One approved case study can become a web page, a one-page PDF for proposals, a LinkedIn post, a two-sentence proof paragraph for your sales emails, and a slide for pitch decks. Ask the AI to produce each format from the approved text only. Our guide to AI content repurposing has prompts for each channel.

The drafting prompt that works

This prompt works well in Claude, ChatGPT and Gemini. Paste it, then attach or paste the transcript and your notes underneath.

You are helping a small consultancy write a client case study. Use ONLY the facts in the transcript and notes below. Do not invent, estimate or round any number, date, percentage or timeframe. Do not write any quote that does not appear word-for-word in the transcript. If a section lacks source material, write [MISSING: what is needed] instead of filling the gap.

Structure: (1) a headline that states the result in plain language, under 12 words; (2) The client — two sentences; (3) The challenge — what was going wrong and what it cost them; (4) What we did — three to five concrete steps; (5) The result — specific outcomes from the source only; (6) one or two direct client quotes; (7) a one-line summary a prospect could skim.

Style: British English, 450–650 words, plain and specific, no hype words such as "transformative", "game-changing" or "seamless". Write it so the client is the hero, not us.

The [MISSING] instruction is the most useful line in the prompt. Instead of a smooth paragraph that quietly papers over a gap, you get a visible flag telling you exactly what to ask the client. To keep drafts in your own voice rather than generic AI prose, add two of your best existing case studies as style examples — the approach is explained in our article on making AI writing sound like your brand.

Mistakes that make AI case studies backfire

Invented precision. "Saved 37.5 hours a month" reads as fabricated when the client actually said "it saves us most of a week each month". Use the client's own phrasing; it is more credible, not less.

Polishing quotes until they sound like you. Real people say "honestly, it just stopped being a headache". AI turns that into "the solution delivered significant operational efficiencies". Keep the original — prospects trust quotes that sound human.

Making yourself the hero. The most persuasive case studies are about the client's problem and outcome. If more than a third of the text describes your methodology, cut it back.

Skipping written approval. Publishing a named client without consent can damage the relationship and, depending on your contract, breach confidentiality. Always get a written yes, even from clients who said "sure" on the call.

Pasting confidential data into consumer tools. Client financials and internal documents should go into a business or team plan that does not train on your data. If you are unsure which tier you are on, check before you paste.

Writing it and hiding it. A case study buried on a page nobody visits does little. Link it from your services page, attach it to proposals, and reference it in follow-up emails.

What it costs and what to expect

You do not need a specialist tool. A €20–€25 per month subscription to Claude, ChatGPT or Gemini, plus the transcription built into Zoom, Google Meet or Teams, covers the whole workflow. At roughly 75 minutes per case study, a freelancer or small agency can realistically publish one a month — twelve pieces of proof a year instead of the usual two.

A practical target for the first quarter:

  • Month 1: list your ten best client outcomes and ask the top three for a 20-minute call.
  • Month 2: publish the first two case studies and turn each into a proposal one-pager.
  • Month 3: add a "results" section to your proposals and track whether win rates move.

Case studies are one of the clearest examples of where AI earns its place in a small business: the human parts — the relationship, the honesty, the approval — stay human, while the hours of drafting shrink to minutes. If you are deciding which other workflows deserve the same treatment, start by checking how ready your business is overall.

Where does your business stand on AI?

Take the free 3-minute AI Readiness Quiz and get a personalised score with your next steps.

Take the Free Quiz →