When a buyer asks an AI assistant “Has anyone used this agency for a hospital website?” or “What results do companies get from switching to this software?”, the assistant goes looking for evidence. Case studies should be that evidence. Too often they are not: vague success stories with no numbers, anonymous clients and outcomes like “significant growth” give AI nothing it can confidently repeat.
A case study that AI can cite reads more like a short research note. It names the client (with permission), sets out the starting point, explains what was done and reports specific, dated results with a clear method behind them. That same structure also makes the story more persuasive to people.
This guide covers what makes a case study citable, a reusable structure, how to present numbers honestly, the legal limits on results claims, and how to get your stories found.
Key Takeaways
- AI assistants look for specific, attributable evidence; vague success stories are rarely cited.
- Every case study should state the context, baseline, actions, results, time frame and measurement method.
- Put the headline result in the first lines and in a summary table so it can be quoted on its own.
- In the US, the FTC expects advertisers to make clear what results are generally expected when testimonials describe exceptional outcomes.
- Publish case studies as indexable HTML pages, link them from relevant service pages and keep them updated.
Why case studies matter for AI answers
Google’s guidance on helpful, people-first content asks: “Does the content provide original information, reporting, research, or analysis?” A real case study is one of the few types of content almost every business can produce that genuinely is original. Google’s generative AI guide goes further, encouraging content that:
“provides unique expert or experienced takes that go beyond common knowledge and the ordinary.”
— Google Search Central, Google's guide to optimizing for generative AI features on Google Search
Answer engines also cite pages they retrieve. Perplexity says each answer “includes numbered citations linking to the original sources”, and ChatGPT search shows citations for web results. When a prompt asks for proof, a page with clear outcomes and context is a natural candidate. That is especially true in B2B, where buyers use AI to shortlist vendors — see AI search for B2B.
Citable vs non-citable case studies
| Element | Hard for AI to use | Easy for AI to use |
|---|---|---|
| Client | “A leading retailer” | “Northfield Garden Supply, a 12-store UK retailer” (named with permission) |
| Problem | “They wanted to grow” | “Online orders had been flat for 18 months and checkout abandonment was rising” |
| Actions | “We implemented our proven strategy” | “Rebuilt category pages, fixed site speed issues, added local inventory feeds” |
| Results | “Huge increase in traffic” | “Organic sessions rose from 21,000 to 34,000 per month between March and September 2026” |
| Method | Not mentioned | “Measured in GA4, same months year on year, excluding paid campaigns” |
| Attribution | Anonymous quote | Named quote from the client’s marketing director |
The example values above are illustrative, not real client data. The point is the shape: names, numbers, dates and method.
A structure AI (and buyers) can follow
Use the same template every time. Consistency helps readers compare stories and helps AI find the same fields in the same place.
- Headline with the outcome. “How a 12-store retailer grew organic revenue 41% in six months” beats “Northfield case study”.
- Summary box. Three to five lines: client, industry, location, challenge, main result, time frame.
- Context and baseline. What the situation looked like before, with numbers where possible.
- Constraints. Budget, timeline, technology or regulatory limits that make the result meaningful.
- What was done. The main actions in order, with enough detail that a peer could understand them.
- Results table. Metric, before, after, period and data source.
- Client quote. Named, with role and company, and approved in writing.
- What did not work. One honest lesson adds credibility.
- Method and caveats. How results were measured and what else might have contributed.
For layout tips, see our guides to AI-friendly tables and lists and answer-first writing.
Presenting numbers honestly
Numbers are what make a case study quotable, and also what makes it risky if they are cherry-picked. A good results table looks like this:
| Metric | Before | After | Period | Source |
|---|---|---|---|---|
| Organic sessions / month | Baseline value | New value | Same months, year on year | GA4 |
| Leads from organic | Baseline value | New value | Quarter vs quarter | CRM |
| Average position for target pages | Baseline value | New value | 90-day average | Search Console |
| Revenue attributed | Baseline value | New value | Six months | Ecommerce platform |
- Give absolute numbers as well as percentages — “up 300%” from a tiny base is misleading.
- Use consistent periods and say whether figures are seasonally comparable.
- Note other factors (a price change, a new product, paid campaigns).
- Date the case study and update it if results change significantly.
If you publish aggregate data across many clients, consider a separate methodology page; our guide to statistics pages for AI citations explains how.
Legal and ethical limits
Case studies are marketing, so advertising rules apply. In the US, the FTC’s Endorsement Guides Q&A says that when an endorser describes results that are not typical and the advertiser cannot show most people get similar results, the ad “must make clear to the audience what the generally expected results of following that same regimen are.” Material connections — for example, a client who received a discount for taking part — should be disclosed clearly. Always get written permission to name clients, share data and quote people, and be extra careful in regulated sectors such as health and finance. In healthcare, patient privacy rules (such as HIPAA in the US) apply, so never publish identifiable patient information without proper authorisation.
Getting case studies found
- Publish each case study as its own HTML page with a descriptive URL; avoid PDF-only downloads.
- Link it from the matching service and industry pages, and from related blog posts.
- Add author or team details so expertise is attributable (see author signals for AI search).
- Ask the client to reference the project on their own site or LinkedIn where appropriate.
- For software companies, connect case studies to feature and integration pages — our SaaS AI search guide covers this.
Common mistakes
- Anonymous everything. If you cannot name the client, at least specify industry, size and region.
- Percentages without baselines. AI may repeat an impressive-sounding number without context.
- Gated case studies. Content behind a form cannot be crawled or cited.
- No dates. Undated results look stale and are hard to trust.
- Overclaiming causation. Say what you did and what changed; do not claim sole credit when other factors were involved.
- One giant “results” page. Separate pages per story are easier to retrieve for specific prompts.
Related Guides
- Brand Facts Page: A Single Source of Truth for AI Assistants
- Quora, Stack Exchange and Niche Forums: Community Mentions Beyond Reddit
- LinkedIn Posts and Articles as AI Sources
Frequently Asked Questions
Do I need to name clients for AI to cite a case study?
It helps a lot, because named and attributable evidence is more credible. If naming is not possible, give specific industry, size, location and dated results.
How long should a case study be?
Long enough to cover context, actions, results and method — often 600 to 1,200 words. Google says there is no ideal page length, so write for the reader.
Should case studies use structured data?
Google says structured data is not required for its AI features. Clean headings, a summary box and a results table matter more.
Can I publish results that are better than average?
Yes, but in the US the FTC expects you to make clear what results are generally expected if the featured result is not typical. Check local rules elsewhere.
How often should I update case studies?
Review them at least yearly. Update numbers if results change significantly and add a ‘last updated’ date.
Turn wins into evidence
A case study with names, numbers, dates and method is useful to every reader — human or machine. Start by rewriting your two or three strongest stories with the template above. For more ideas see our guide to case study marketing and social proof marketing. If you want our content writing team to help, request a custom quote.
