Writing a client case study is one of the highest-leverage things a marketing agency can do — and one of the most procrastinated. The result is a folder of half-finished drafts and a website that sells on claims instead of proof. At Choco Media, we solved this by building a process that takes a raw client interview and produces a publishable case study in under an hour, using AI for the heavy lifting and a human for the judgment calls. The ai case study writing process is repeatable, scales across clients, and produces work you are actually proud to publish.
This post is for agency owners, account managers, and content leads who have results worth documenting but not enough time to write them up properly. By the end you will have a 15-question interview framework, a prompting structure that generates a strong first draft, and a two-pass editing approach that gets you to publish without endless revision cycles.
We are not cutting corners on quality — we are cutting time spent on the parts that do not require human judgment. The client relationship, the insight extraction, the final voice check: those stay with a person. Everything in between moves through AI.
Why most agencies never publish their case studies
It is not a lack of results. Most agencies have genuinely strong outcomes sitting undocumented in Slack threads, client emails, and the memory of whoever managed the account. The bottleneck is almost always the writing itself — specifically, the gap between knowing a project went well and turning that knowledge into a structured, persuasive narrative.
Case studies require a particular kind of writing that feels unfamiliar to most marketers: structured narrative with a clear problem-solution-result arc, specific numbers without overstating, and a voice that is confident without sounding like a press release. That is hard to sit down and produce cold, which is why the draft gets postponed indefinitely.
- The account manager who knows the story is too busy with live accounts to write it
- The content person does not have enough context to fill in the detail
- The client approved sharing results but the permission email got buried
- Someone started a draft three months ago that is now 40% done and feels too stale to finish
AI does not solve the context problem — you still need a proper interview. But it does solve the blank-page and structuring problems, which removes most of the activation energy that keeps case studies from getting written.
The 15-question client interview
The interview is the foundation. A weak interview produces a weak case study regardless of how good your prompts are. We keep it to 15 questions across five categories, which typically takes 20 to 30 minutes by video or async voice note.
Context questions (1-3)
- What was the business situation before we started working together — what was working, what was not?
- What made you decide to bring in an agency at that particular moment?
- What had you tried before, and why did it fall short?
Problem questions (4-6)
- In your own words, what was the core problem we were hired to solve?
- What was the cost of that problem — in time, revenue, or opportunity?
- Who internally felt the pain most, and how did it show up in their day-to-day?
Process questions (7-10)
- Walk me through the first 30 days of working together — what happened and in what order?
- Was there a moment where the approach clicked, or a turning point where things started to shift?
- What surprised you about how we worked — positively or otherwise?
- Which specific deliverable or decision made the biggest difference?
Results questions (11-13)
- What metrics changed, and by how much — do you have numbers you are comfortable sharing?
- Beyond the numbers, what changed qualitatively — in team confidence, in how the market sees you, in internal clarity?
- How long before you started seeing results, and did the timeline match expectations?
Reflection questions (14-15)
- If you were describing this project to a peer considering hiring us, what would you say?
- What would you tell your previous self about what to expect?
Record the call with permission and use a transcription tool — we use Otter or Whisper for this. The transcript becomes the raw material for your AI draft prompt.
The single most useful question in the interview is number 9 — “What surprised you about how we worked?” Clients almost never volunteer this unprompted, but the answers are the most quotable, most specific, and most persuasive lines in the final piece.
Preparing the input: what to feed the AI
A transcript alone is not enough to produce a good first draft. You need to frame it with context the model cannot infer. Before you paste the transcript, assemble a short brief:
- Client name and sector — even if you will anonymise later, the model needs this to understand context
- Services delivered — reference the relevant service area so the draft can reflect what you actually did
- Key metrics — pull the three or four numbers from the transcript you definitely want in the final piece
- What to avoid — client sensitivities, competitor names you cannot reference, anything legally complex
- Target reader — who is this case study primarily for? A prospect in the same sector? A different company stage? This shapes the emphasis
- Approximate word count — we aim for 600 to 900 words for a case study; long enough to be credible, short enough to be read
With the brief and transcript ready, the prompt takes about five minutes to write. The specificity of the brief is directly proportional to the quality of the draft.
The AI draft prompt structure
We use a structured prompt rather than a freeform request. Here is the skeleton we use:
- Role: “You are a B2B content writer for a marketing agency. Write in first-person plural. Tone: calm, direct, specific. No hype words.”
- Structure instruction: “Write a case study with these sections: Challenge, Approach, Results, What the client said. Each section 100 to 150 words.”
- Source material: paste the brief, then the transcript
- Constraints: “Do not invent numbers. If the transcript is ambiguous on a metric, write [VERIFY: metric] in the draft.”
- Quote instruction: “Pull two or three direct quotes from the transcript and integrate them into the relevant sections.”
The [VERIFY] pattern is particularly useful — it forces the model to flag uncertainty rather than produce a plausible-sounding number from nowhere. In client work we have found this single instruction eliminates about 80% of the fact-checking work on the back end.
Our AI content creation service uses the same principle: AI handles structure and volume, a human handles accuracy and voice. The division of labour is explicit, not assumed.
The two-pass edit
The AI draft is almost always structurally sound but tonally imperfect. Two focused passes gets it to publish-ready without a full rewrite.
Pass one: accuracy and specificity (10 minutes)
Read with the transcript open. Check every [VERIFY] flag. Confirm numbers match the source. Replace any generic phrases the model defaulted to with the specific language the client actually used in the interview — their phrasing is almost always better than a paraphrase. Cut any section that is padding rather than evidence.
- Every number should be traceable to either the transcript or your internal reporting
- The Approach section should name specific things you did, not describe a category of activity
- The client quote should read like something a human said, not a testimonial template
Pass two: voice and flow (5 minutes)
Read aloud, or use text-to-speech. The goal is to catch anything that sounds like AI filler, generic agency positioning, or passive constructions that weaken the narrative. Tighten the opening sentence — that is the one most readers make a decision on. Check the closing: it should end on the result, not on a sales message.
What to do with results you cannot fully share
Not every client will approve specific revenue numbers or percentage improvements. That is a real constraint and it is more common than agencies like to admit. A few approaches that maintain credibility without overpromising:
- Directional language with timeframe: “Organic traffic increased consistently over six months, with the strongest growth in months four and five.”
- Relative improvement: “Cost per lead fell to its lowest point in 18 months.”
- Qualitative outcomes the client did approve: team capacity freed up, process clarity gained, confidence in a channel that previously felt uncertain
- Anonymous framing: “A B2B SaaS client in the HR sector…” — this removes the attribution concern while preserving the specificity
A case study with specific directional evidence and strong qualitative quotes is more credible than one with vague numbers that feel inflated. Readers — especially B2B buyers — have a well-calibrated sense for when metrics have been optimised for impressiveness rather than accuracy.
Publishing and placement
A case study that lives only on a /case-studies/ page gets found by roughly no one. The pieces that drive actual pipeline are the ones placed where the right reader already is. For agencies managing content at scale, the bespoke retainer model is how we embed systematic content production into an ongoing client relationship — case study production included.
- Service pages: embed the relevant case study or a short excerpt on the service page it demonstrates
- Proposals: a PDF version of the case study sent with a proposal increases close rates more than almost any other addition
- Sales emails: a single paragraph summary with a link, sent at the right moment in a conversation
- LinkedIn: the client quote, reframed as a short post, often outperforms original content because it is evidence rather than opinion
- Blog cross-links: case studies make excellent supporting evidence for how-to posts — link to them from relevant articles rather than treating them as a separate content type
Building the habit: making case studies automatic
The teams that consistently publish case studies are not the ones with the best writers — they are the ones with a trigger built into the account management process. We add a case study flag to our internal project checklist at the 60-day mark: if results are tracking well, the interview gets scheduled then, not six months later when the account manager has moved on and the client has forgotten the specific before-state.
- Set the trigger at 60 days, not at project close — you need results to have materialised but the experience to be fresh
- Make the interview a standing agenda item in the client relationship, not a separate ask
- Use async voice notes if the client is pressed for time — a 10-minute voice note covers the same ground as a 25-minute call and is easier to schedule
- Keep a running folder of approved client quotes — these can be used in case studies, on service pages, and in sales materials without requiring a new interview each time
Once the process is routine, each case study takes about 60 minutes of real work: 25 to 30 minutes for the interview, 5 minutes to prep the prompt, 15 minutes for the AI draft and initial review, and 15 minutes of editing across two passes.
The compounding value of documented proof
A single case study has limited reach. Ten case studies across different sectors, service lines, and company stages create a body of evidence that does selling work continuously. Prospects who arrive sceptical leave with specific answers to the questions they were going to ask on a call.
In client work we have found that agencies with five or more published case studies close proposals at a measurably higher rate than those relying on credentials alone. The case study is not just marketing content — it is the signal that you take results seriously enough to document them. That signal matters to the kind of clients worth working with.
If you want to build this process into your agency operations or have us handle case study production as part of a broader content engagement, get in touch and we can talk through what that looks like in practice.