Build an AI Case Study Generator

When a project ends, write a short wrap memo covering the results and key KPIs. It does not need to be polished; a few bullet points is enough. If you are pressed for time, record a quick voice memo instead.

This memo is the raw material the system runs on. The better the wrap memo, the better the case study, but even a messy memo can produce something useful. Use the example wrap memo in the resources as a reference.

Go to Claude, click Projects in the left sidebar, then choose New Project. Name it “Case Study Generator.”

Open the project instructions field and paste in the full contents of the “Claude Project Instructions: Case Study Generator” file. This tells Claude how to structure each case study, including the seven-section framework, tone rules, what to do with missing data, and how to handle different types of raw input.

Upload the “Case Study Examples and Structure Reference” and “Case Study Do’s and Don’ts” files as project files. These give Claude annotated examples to calibrate against and a quality control checklist to run on every output.

You only do this setup once. Every case study you generate from here on should live in a new chat inside the same project.

Pro tip: Name each chat after the client and project, such as “Vanta Health, Q4 Campaign.” This keeps your case study library organized and searchable inside the project.

Start a new chat inside the project. Upload your project documents directly into the chat.

Generate a publication-ready case study from these documents. Lead with the strongest result in the headline. Use the challenge, solution, results framework. If any section is missing data, flag it with [NEEDS INPUT] rather than inventing information.

Claude will read what you uploaded, cross-reference the project instructions and examples, and produce a full draft. With extended thinking on, this takes about 60 to 90 seconds.

Pro tip: If the PDF formatting looks strange, don’t worry. It will look fine when you download it or export it to Google Docs.

Read through Claude’s output against the checklist in the Do’s and Don’ts file. Verify the main quality markers before you publish or share it.

• Does the headline include a specific number or outcome?
• Is the “before” state clearly established?
• Are there at least three metrics in the results section?
• Is there a client quote, or a [NEEDS INPUT] flag where one should go?

If you do not have a client quote, ask Claude to draft two or three options based on the results. Send them to your client and ask if they can put their name on one.

Once you are happy with the case study, Claude can also generate a slide deck version or a social media carousel from the same case study.

Save the final case study as a PDF and drop it into a shared folder. Label it clearly with the client name and primary result.

Upload the final PDF back into your Claude project as a context file. This is how the system gets smarter over time: each new finished case study becomes an example Claude can reference when generating the next one.

Only upload cases you have already edited and are happy with. If you feed Claude unedited drafts, it will optimize for the patterns in those drafts, including the bad ones.