Blog · AI
— AI··10 min read

How to use AI to write case studies without inventing the numbers

Joona Heinonen· Choco Media · Rovaniemi

AI case study writing is one of those tasks that sounds straightforward until you sit down to do it. You have a result, a client, and roughly fifteen tabs open — and the AI keeps generating confident-sounding sentences about outcomes you never actually measured. At Choco Media, we have run this process enough times to know where it breaks and what keeps it from falling apart. This post walks through the exact approach we use to write case studies with AI assistance, without fabricating a single number or bending a result into something it was not.

This guide is for marketing teams, agency operators, and freelancers who want to use AI to speed up case study production without sacrificing the specificity that makes case studies worth reading. You will leave with a structured process, a set of interview questions that extract the data AI needs to work with, and an understanding of where human editing is non-negotiable.

Case studies are some of the most conversion-relevant content a service business can publish. They do the selling without selling. But they only work when they are precise — readers detect vagueness immediately, and a generic case study does more damage than none at all.

Why AI Alone Cannot Write a Good Case Study

Before we get into the workflow, it is worth being honest about the fundamental problem. AI models are trained to produce fluent text, which means they will fill gaps in your brief with plausible-sounding fabrication if you let them. Case studies are built on specifics: actual numbers, actual timelines, actual decisions that were made. If you give the model a vague prompt — “write a case study about a client whose SEO improved” — you will get a document with invented percentages and made-up quotes that erodes trust the moment a reader checks anything.

The solution is not to avoid AI — it is to structure the process so that AI only works on material you have already gathered and verified. Think of it as a production assistant, not a researcher.

Step One: The Information Gathering Interview

The bottleneck in every case study is the brief, not the writing. If the brief is detailed, AI can produce a strong first draft in minutes. If it is thin, no amount of prompting will fix it. We run a structured 20-minute call with the relevant account manager or client lead before touching a keyboard.

The questions we always ask

We take notes during this call — not a transcript, just the raw facts. The goal is to produce a brief that contains every claim the case study will make, sourced and verified before we write a word. If a number is uncertain, we mark it as approximate. If a quote is paraphrased, we flag it for client review before publishing.

Step Two: Building the Case Study Brief

With the interview notes in hand, we write a structured brief in plain text before we involve AI at all. The brief has a fixed format:

This brief is the only source the AI works from. We explicitly instruct the model to use only the information in the brief and to flag any gap where it would need to invent a detail rather than fill it in.

“The brief is not a prompt. It is a contract between the data you have gathered and the story you are allowed to tell. When we treat it that way, the AI produces a first draft that needs editing, not fact-checking.”

Step Three: The AI Drafting Prompt

Once the brief is solid, we use a structured prompt that constrains the model tightly. We do not ask for a case study in one instruction — we ask for each section separately, passing the relevant excerpt of the brief for each. This reduces hallucination because the model’s context window is focused on a smaller, specific set of facts.

The prompt structure we use

The [MISSING] flag is the most important part. It forces the model to surface gaps rather than paper over them, which means your editing pass catches structural problems before the piece goes to a client or gets published. This is part of how we apply our AI content creation process across formats — the constraint layer is more important than the generation layer.

Step Four: The Results Section — Where Most Teams Get It Wrong

The results section is where AI case study writing most commonly falls apart, and it is worth dedicating a section to it. There are three failure modes we see repeatedly:

Failure mode 1: Percentage without baseline

A 40% increase in conversions is meaningless without knowing what the baseline was. “From 1.2% to 1.68%” is a very different claim than “from 0.3% to 0.42%”. Always express results as both the percentage and the underlying numbers.

Failure mode 2: Timeframe inflation

AI will happily write “results within weeks” when the data shows results after four months. The timeframe is part of the credibility claim — it tells the reader whether this is typical or exceptional. Be precise.

Failure mode 3: Attribution looseness

Revenue increased 30% during the campaign period is not the same as saying our campaign caused a 30% revenue increase. In client work we have found that precise attribution language — “organic traffic from target keywords increased 54% in the six months following launch” — is both more honest and more convincing than overclaimed correlation.

Step Five: The Human Editing Pass

The AI draft goes through a focused editing pass before anything else. We are not line-editing for style at this stage — we are doing a structural accuracy check. Our editors work through the following sequence:

This editing pass typically takes 25–40 minutes for a 700–900 word case study. It is not optional, and it is the one part of the process where AI assistance is counterproductive. If you use AI to check the AI draft, you are checking fluency, not accuracy. This is one of the human layers we discuss in our broader piece on AI automation for marketing teams — the gate step has to be human.

Step Six: Client Review and Final Sign-Off

Before a case study goes anywhere near publication, the client sees it. This is both a courtesy and a risk management step. Clients sometimes remember results differently, have legal constraints on what can be published, or simply did not expect the level of detail. We send a clearly formatted review document — the draft with a note explaining which numbers came from which source — and ask for explicit approval of every factual claim.

In client work we have found that the review step also strengthens the relationship. Clients who see their results presented well, accurately, often become more forthcoming about data in future projects because they trust how you will use it.

How Long Does This Process Take?

With the workflow above, a single case study from interview to published draft takes roughly three to four hours of human work, spread across a week. The breakdown:

Compare that to writing from scratch without AI: typically six to eight hours for a thorough case study. The AI acceleration is real, but it is concentrated in the drafting phase. The human work — gathering, checking, reviewing — does not compress much, and it should not. The human work is what makes the case study credible.

Where Case Studies Fit in Your Content Strategy

Case studies are not discovery content — they do not attract new audiences through search in the way that educational posts do. They convert people who are already evaluating you. That means they belong later in the funnel: on service pages, in proposals, in follow-up emails after a sales call. If you are publishing case studies and wondering why they are not driving traffic, the answer is that they were not designed to. They are designed to close.

That said, a well-structured case study with specific keywords — the service you provided, the industry you worked in, the platform or tool you used — can attract high-intent searches. We have seen case studies rank for queries like “agency [service] for [industry]” when the content is specific enough. The SEO value is real, it just requires treating the case study as a content asset with proper on-page structure, not just a PDF on a testimonials page. Our SEO service includes case study optimisation as a standard element of content audits.

Starting Your First AI-Assisted Case Study

If you have not run this process before, start with one case study from a recent project where you have clean data and a client you trust to review quickly. Run the interview, build the brief, run the AI through the section-by-section prompt, do the editing pass, and send to the client. The first run will take longer than the estimates above — that is normal. By the third or fourth case study, the brief template and prompt structure will feel like a production line.

If you want a second set of eyes on your brief structure or your results section, or if the whole content operation needs a structural overhaul, we are happy to take a look. The contact page has everything you need — a short message with what you are working on is the right starting point.

— Work with Choco Media

Want posts like this working for your business?

10–40 SEO + AI-optimised blog posts a month, researched, senior-edited and published straight to your site. Built to rank on Google and get cited by ChatGPT, Claude and Gemini.

See plans — from €199/mo →
No start-up fee · Price locked for 12 months · Cancel any time after
← All storiesNext story →
— Free tips, monthly

Get the playbook, for free.

One short letter a month — the prompts we use, the campaigns that worked, the AI tools worth the time. No sales pitch, just field notes.

— Want us to do it for you?

Hire the agency.

AI-accelerated content, paid media, brand and web — delivered by one small team that talks to itself. Currently taking on a handful of clients each quarter.

Book a call