ChatGPT Prompts for Case Studies That Convince

Case studies are the content type marketers rate highest and produce most reluctantly. In Content Marketing Institute’s B2B research, 53% of 980 B2B marketers named case studies and customer stories among their best-performing content, second only to video. The reluctance comes from the process: getting a customer on a call, extracting a story from a rambling transcript, chasing approval on every number. Most of that friction is synthesis work, and synthesis is what a model does well.

The rule that governs every prompt in this set: the model arranges evidence, it never supplies it. A case study is the one content format where a hallucinated number is not an embarrassment but a small fraud with a customer’s name attached. So these prompts run on pasted raw material - the interview transcript, the real metrics, the approval constraints - exactly like my persona prompts run on pasted customer evidence. Same method, pointed at one customer instead of a segment.

Why case studies earn this much care

The buying research says the case study’s job is bigger than most teams treat it. 6sense’s Buyer Experience Report, from roughly 4,000 B2B buyers, found buyers choose a vendor from their day-one shortlist 95% of the time, and in 85% of successful purchases they had direct experience with the winning vendor before the deal. Your case study is rarely read by a blank-slate stranger. It is read by someone validating a preference they largely formed already, often forwarding it to the committee members who have to sign off. My conclusion from those numbers: a case study’s real audience is the skeptic the champion has to convince, which is why the prompts below build for scrutiny, not for applause.

What to gather before you prompt

Fifteen minutes of collection, same as always. My kit: the interview recording or transcript (even rough auto-transcription works), every real number you are allowed to consider using, the customer’s own words about life before your product, and any constraint the account team knows about - what the customer will and will not approve, whether the company name can appear. If you have no interview, stop and get one. A case study assembled from a CRM record and enthusiasm is a testimonial wearing a lab coat.

The prompts

1. The evidence extraction

You are a case study researcher. Below is an interview
transcript with our customer, plus notes from the
account team. [PASTE BOTH]

Extract, quoting or tightly paraphrasing and citing the
transcript line for each item:
1. The situation before: what was breaking, in their words.
2. The trigger: what made them look for a solution.
3. The evaluation: what they compared us against and why
   we won.
4. The change: every specific outcome mentioned, with the
   exact number if one was given and NO NUMBER if not.
5. The friction: anything that went badly or slowly.
Use only what is in the material. Mark thin sections as
thin instead of filling them.

The friction list is the section teams skip and readers hunt for. A case study with no friction reads as fiction, because every buyer knows implementations have some. Keeping one honest bump in the story is the cheapest credibility purchase available.

2. The angle decision

From the extracted evidence, propose three different
angles for this case study, each as: a working title, the
one claim the story proves, which evidence carries it,
and which audience segment it will land hardest with.
Do not pick generic angles ("Company X grew with our
product"). Each angle must be specific enough that a
competitor could not paste their name into it.

One interview usually contains two or three stories, and choosing before outlining is the difference between a case study and a timeline. I pick the angle whose evidence is strongest, not the one whose claim is biggest.

3. The outline

Build a case study outline for the chosen angle. Sections:
headline with the customer and the concrete result, a
three-sentence summary a skimmer could act on, the before
(their words), the evaluation and decision, the
implementation including the friction point, results
(numbers only from the evidence, each with its context),
and what they would tell a peer considering us.
For each section: the goal in one line, the evidence it
uses (cited), and a target length. Flag every place where
approval or a missing number blocks drafting, as a list
at the end.

The blocked-items list is the project manager inside the prompt. It comes out as “need customer sign-off on the 40% figure, need a quote for the peer section,” which is your actual to-do list before drafting starts.

4. The results framing

Here are the approved metrics and their context. [PASTE]
For each, write the result three ways: the bare number,
the number with its baseline and timeframe, and the
number translated into what it meant operationally for
the customer. Never extrapolate, annualize, or combine
metrics into new ones. If a metric lacks a baseline, say
so instead of implying one.

The no-combining rule exists because models love to turn “saved 10 hours a week” and “team of 12” into “saved 6,240 hours a year,” a number nobody approved and the customer never said. Context beats size: in my experience “cut reporting from two days to three hours” out-converts “93% faster” every time, because the reader can feel the first one.

5. The pull quotes

From the transcript only, select the five strongest
verbatim quote candidates: specific, outcome-flavored,
and sounding like a person rather than a press release.
For each, note where in the outline it belongs and what
light editing (cuts only, no additions) would sharpen
it. Mark any quote that needs customer approval for the
edited form.

Cuts only, no additions. Tightening a rambling sentence is editing. Adding words the customer did not say, in a document with their name on it, is the line this whole workflow exists to hold - the same discipline as the fact-checking pass every claim gets before publication.

6. The derivatives

From the finished outline and approved evidence, draft:
a 150-word version for a sales one-pager, three social
posts each built on one result, and five title/subject
line options for promoting it. Reuse only approved
numbers and quotes. Keep the customer's name out of any
derivative the approval notes exclude it from.

A case study is expensive evidence, and evidence this good should not live on one URL. This is the extraction-then-reformat pattern from my repurposing workflow with the extra constraint that the approval boundary travels with every derivative.

The approval pass, unautomated

Everything above accelerates the writing. Nothing above touches the two steps I keep manual: the customer sees the full draft before anyone else does, and every number in it traces to something they said or signed off. I treat the approval email as part of the deliverable - the version they approved is the version that ships, and edits after approval go back for a second yes, however small. Tedious, and far cheaper than the alternative conversation.

FAQ

Can ChatGPT write a whole case study from a call transcript? It can draft one, and with prompts 1 through 4 the draft will be structurally sound. What it cannot do is know which claims the customer will stand behind, which is why the workflow separates extraction and outlining (automatable) from approval (never). The draft is fast now. The sign-off takes exactly as long as it always did.

What if the customer has no hard numbers? Write the case study anyway, on the operational change: what the team stopped doing, what became possible, how the workflow looks now. Prompt 4’s “translated into what it meant” framing is built for this. A specific story without percentages beats a vague one with them, and inventing a proxy number to fill the gap is how case studies end up quietly retracted.

How long should a B2B case study be? As long as the evidence supports. My typical output from one good interview is 600 to 900 words, and the outline prompt’s per-section targets keep it there. Past that length you are usually padding the implementation section, which nobody rereads.

Should case studies be gated? Mine are not. The 6sense data above says buyers arrive with their shortlist mostly set, doing their reading before they ever talk to sales - a gated case study is invisible exactly when it would work hardest. Gate the derivative assets for sales conversations if you like, but let the story itself be found.

This set is part of my prompt and workflow library. The evidence discipline comes from the persona prompts, the verification pass is the fact-checking prompts, and the wider practice starts at the complete guide to AI marketing.