Case Studies for AI Products: A Template Founders Can Copy
A case study structure that works for a solo founder with one or two users: the six sections, what evidence each needs, and how to get permission.
Published 2026-10-14 · 4 min read
A case study does not require a large customer. It requires one person with a specific problem, a documented change, and permission to describe it. Founders with two users often have everything needed and assume they do not.
This is the structure, the evidence each section needs, and the process for getting it published without damaging the relationship.
The six sections
| Section | Length | What it must contain |
|---|---|---|
| 1. Who they are | Two sentences | Role, company size, and the situation — enough for a reader to recognise themselves |
| 2. The problem, in their words | One short paragraph | Their phrasing, not yours. Include what it cost them in time or money |
| 3. What they tried first | Two or three sentences | The alternative and why it did not work. Without this the case study reads as fiction |
| 4. What changed | The longest section | The specific workflow change, with numbers where they exist |
| 5. What it did not fix | One or two sentences | The honest limit. This is the section that makes the rest believable |
| 6. Who else it suits | Two sentences | The profile of a similar buyer, and who should look elsewhere |
Section by section
- 1
Get the identity right, then anonymise if needed
"Head of operations at a 20-person bookkeeping firm" is nearly as useful as a name and much easier to obtain. A named case study is stronger, but an attributed-role one is not worthless — and refusing to publish anything because you could not get a name is the common mistake.
- 2
Record the problem before you solve it
This is why the best case studies come from design partners and early users you spoke to before building. Ask for the exact phrasing they used. "I spent every Friday reconciling" is worth more than any summary you write.
- 3
Document the baseline
What did the task cost before? Time, money, errors, or workarounds. If you do not have a baseline number, ask for an estimate and label it as one.
- 4
Describe the workflow, not the features
Readers want to see themselves doing it. "She imports the statement, the tool matches transactions, she reviews the eleven it flags" is a workflow. "Our AI matching engine" is not.
- 5
Measure the change, then verify it
Ask the person whether your numbers are right before publishing. They often are not, and being corrected before publication is far better than after.
- 6
Add the limitation
Every case study has one. Stating it is the difference between a document and an advertisement, and it is the part readers remember.
The permission process
- Draft it first, then send it. Do not ask "can we write a case study about you?" — send the finished draft. It is less work for them and they can see exactly what will be public.
- Offer specific redactions. Numbers they do not want public, the company name, the person's name. Ask which, rather than assuming all or nothing.
- Give a deadline that is easy to miss safely. "If we do not hear back by Friday we will hold it" — not publish it.
- Send the live URL once published. Most people appreciate seeing it, and it makes the next one easier to ask for.
- Include a withdrawal route. One line saying they can ask for it to come down. Almost nobody does, and offering it removes the main hesitation.
Where a case study goes
| Placement | Purpose |
|---|---|
| Its own page on your site | The canonical version, linkable and indexable |
| Condensed into a testimonial | One or two quotable sentences for the homepage and listings |
| Referenced in comparison pages | Evidence for a claim you would otherwise be asserting bare |
| Sent to prospects mid-conversation | The single most effective sales asset a small team has |
| Summarised in a community post | The finding, not the pitch — "here is what changed for one bookkeeper" |
| Used in directory listings | Where a listing allows a longer description, one specific result beats feature copy |
What to do with only one
One case study is enough to start, and it should be used in six places rather than sitting on a page nobody visits. The mistake is treating it as a document; it is a source of evidence for everything else you write.
A single well-documented user also tells you who your actual buyer is, which is usually not who you expected. That insight changes your listing categories, your one-liner and the communities you spend time in — which is worth more than the case study itself.
A case study is not proof that your product is good. It is proof that one specific person's problem got solved, which is the only version of that claim anyone believes.
More in Content & building in public
- How to Get Your First 10 Newsletter MentionsNewsletters are a reachable audience for a new AI product. How to get mentioned in ten of them: finding the right ones, becoming a source, and pitching.
- Repurposing Your AI Product Launch Across 6 ChannelsOne launch, six surfaces: how to cut the same material for a blog, a thread, LinkedIn, a community, listings and a pitch without repeating yourself.
- How to Write a Changelog People Actually ReadA changelog is the cheapest trust signal a software product can publish, and most are unreadable. Structure, voice, and what to leave out.