The AI Case Study Generator skill for Claude

The problem the skill solves sits in its own frontmatter, in the words of the people who have it: we do good work but nobody knows it. Our cases read like an invoice. Prospects always ask for examples. Our portfolio is a list of logos with no story. The pattern is always the same: the work was good, the story about it is a list of delivered activities in which the client is at best an extra. A piece like that convinces nobody and gets found by nobody.

This skill turns that around. It writes no project report and no task list, but a story in which the client plays the lead role and the reader recognises themselves in the starting point. Around that it builds the commercial layer that actually makes a case pay for itself: a title and meta for the search results, a heading structure, internal links, a question block, and the short versions for LinkedIn and the sales deck. It is part of the skill library we make available for free from the AI and automation service, with no account and no sales email afterwards.

One thing you should know beforehand: the skill turns a good project into a good story, but it does not turn a project with no result into a case. The first thing it does is check whether there is a measurable result, and if that is missing, it says outright that what you have is a project description, not a case study. That is not a lack of helpfulness, that is the reason the output stays believable.

The agency plays the lead role. The first check is the lead-role check: count how often the agency is the subject of the sentence and how often the client is. In cases that do not work, the agency wins by a mile. We analysed, we implemented, we delivered. Every sentence that allows that gets turned around: not we delivered the reports, but the client's team could run its own reports from week six. Same fact, different lead character, different effect.

The result arrives too late. The second check looks at whether the result sits in the first hundred words. A reader who has to wade through three paragraphs of approach before knowing whether it paid off is already gone. That is why every case from this skill opens with a four to six line summary block with the key figure in it, aimed at readers who do not read on and at AI systems summarising the page.

The figures float free. A rise of thirty per cent says nothing without the questions behind it: measured over what, measured with what, compared to what. The figure check flags every number without a measurement period or source, and every relative rise without an absolute base. In the generated text no figure appears without that context, because a figure without a source is not evidence, it is a claim.

There is no friction in it. The file is explicit about that: also name what disappointed or went differently, because a case with no friction at all reads like an advert. That is why every case has a mandatory block on what we would do differently. In the example: the configurator was built before the price list had been cleaned up, so work had to be redone in week eight. Two lines of honesty that make the rest of the story believable. Someone who never admits anything is never believed either.

In diff form, with lines from the skill's own example: what goes out and what comes back in its place.

Then comes the core: seven fixed steps. First it checks whether there is a measurable result, rewrites the starting point into the client's own language, splits the approach into three to five named phases each with a decision and the reason behind it, carries figures over literally with measurement period and measurement tool, supplies the SEO layer, makes the two derivatives for LinkedIn and the sales deck, and closes with the list of open gaps. Before anything is written, it runs through a mandatory nine-point checklist, from the client's industry and size to the search term and the place of publication, and it only asks about what is missing.

A striking number of rules are about language. It writes in the second person, uses no dashes in running text, always writes AI in capitals and bans itself a whole list of words: seamless, effortless, unlock, take to the next level, in a nutshell. Hollow superlatives such as groundbreaking and incredible are also on the blacklist, as are lists that suspiciously often turn out to contain exactly three things. Concrete over abstract, figure over adjective, short sentences alternated with long ones. And the most important language rule is not a style rule but a structure rule: the client is the subject of most sentences, not the agency.

There are also variants built in for different channels: Before After Bridge for social posts and adverts, PAR for a sales deck or one pager, the narrative variant in which the client's own quotes form the thread, the anonymised variant for clients who do not want to be named, and the multi-part story in which one worked-out main case is backed up by two shorter side cases. Each variant states when you choose it.

It closes with the seven things the skill never does, more on that below, and a source list of five titles: Cialdini, Minto, the Heath brothers, Yin, and the note that the STAR framework comes from behavioural interview technique and cannot be traced to one single publication. That theory is the reason the structure is the way it is.

The social proof comes from Cialdini. Robert Cialdini described in Influence from 1984 why people look at comparable others to decide what is sensible. According to the file, a case study is the most concrete form of that: the reader sees not only that someone chose, but also why. That is why that other person also has to be genuinely comparable, and that is why the skill rewrites the starting point into the client's own language.

The storyline comes from Minto. Barbara Minto's SCQA framework, known from her work on the pyramid structure, orders the story from situation through complication to question and answer. You saw it above in FIG.03. Anyone who wants to use that way of thinking more broadly than cases can turn to the StoryBrand Writer skill, which applies the same principle, the client as the hero of the story, to your whole message.

The completeness check comes from interview technique. STAR, Situation, Task, Action, Result, forces every phase to name what had to happen, what was done and what it delivered. The file is blunt about it: without the R a case stays a description of the work. Anyone who wants to learn to use the framework itself, for cases or for job interviews, finds it worked out in the STAR Method Coach skill NL.

The concreteness comes from the Heath brothers. Made to Stick from 2007 underpins two rules that come back everywhere in the skill: concrete beats abstract, and an unexpected detail makes a story more believable than a string of adjectives. Hence the mandatory block on what went differently, and hence the rule that a figure always outranks an adjective.

And Yin guards the boundary. Robert Yin's Case Study Research from 1984 sits in the source list to mark the difference between academic case research and a commercial case study. This file does the second and does not pretend to do the first: it draws no conclusions the data cannot carry, and if a result could also have other explanations, it names them.

Even without installing the skill you can improve your own case with these principles: put the result in the first hundred words, make the client the subject of your sentences, and give every figure its measurement period and source.

After that, using it is simple: paste your project information and ask for a case study. It runs through its checklist and only asks about what is missing. New to skills, then read what Claude skills actually are NL first and the step by step installation guide NL. Want to write a skill for your own way of working afterwards, then the structure of that kind of file is explained in writing a SKILL.md NL.

There are also situations where you are better off leaving it be, and the file names them itself. Academic case research is one: that is a methodology with its own requirements, and the reference to Yin sits there precisely to guard that difference. Product descriptions with no client are a second: without a client there is no story, only a specification. And press releases a third, because those have a different purpose and a different form. The main boundary, though, is the result: if nothing was measured, there is no case, at most a plan to start measuring after all.

One more honest boundary: the skill makes the story better, not the work. An average project does not become strong evidence through a good structure, and the skill helps you see that rather than dress it up. What it is good at: finally making the strong project that now reads like an invoice sound the way it actually was. Further down the sales chain the case then takes its place: as checkable evidence in the Why us block of your quote, for which the AI Quote Writer skill has set aside a dedicated spot, or as ammunition in the SPIN sales conversation NL that comes before that quote.

What a page like that looks like in real life you can see for yourself with us: on our own cases page we apply the same principles, including naming what has not been measured.

A strong case study is the best sales material you have, because it tells what happened at a real client instead of what you promise. It only earns its place if someone actually lands on that page. If your site gets little search traffic, the work sits before the case, not after it. The free SEO scan shows in a few seconds where your site stands, and if that turns up more than you want to take on yourself, that is the improving SEO service.