aeo
Case Study for AI Citation: The Full Playbook.
A case study for AI citation fails for the opposite reason most people expect. It is not too plain, it is too vague: hedged percentage ranges, unnamed clients, three paragraphs of philosophy before a single number, and a PDF behind a form an AI crawler will never fill out. Building a case study an engine will actually lift into an answer means writing the proof the way you would write a data point, not the way you would write a brochure.
A generic guide can explain what answer engine optimisation is. A comparison page can explain why a category exists. Only a case study answers the question a buyer close to a decision actually asks an AI engine: has this worked for a company like mine. That is a proof-stage question, and most B2B sites have nothing built to answer it in a form an engine can extract.
What Makes a Case Study for AI Citation Different From a Sales Case Study?
A case study for AI citation is written to be quoted by a model, while a sales-enablement case study is written to be read by a prospect a rep has already warmed up. The sales version can lean on tone and a logo wall, because a human reader supplies the trust a cold page has not yet earned. An AI engine supplies none of that context. It scans the page for a specific, verifiable claim it can attribute confidently, and if that claim is buried under brand language or hedged into a vague range, the model has nothing solid enough to repeat.
Why Do Most B2B Case Studies Never Get Cited by ChatGPT or Perplexity?
Most B2B case studies never get cited because they are optimised for a human reader's trust, not a model's extraction step. Three failures show up repeatedly during a SourceRank AI audit: the result is stated as a range ("20 to 50 percent improvement") instead of a figure, the client is anonymised as "a leading logistics provider" with no way to corroborate the claim, and the document sits behind a gated PDF form no crawler will complete. Any one of those three is enough to keep an otherwise strong result out of a generated answer.
How Should You Structure a Case Study for AI Citation?
Structure a case study for AI citation so the proof sits in the first hundred words, not the last. Use this five-part order, which mirrors what an engine's extraction step is actually looking for:
- 1. Lead with a fact box. A short, self-contained summary at the top: client, category, one headline metric, timeframe. This is the block most likely to be lifted whole into an answer.
- 2. State the challenge in one sentence. What specific problem existed before the engagement, named plainly rather than dressed up as an industry trend.
- 3. State the action in one or two sentences. What was actually done, specific enough that a reader could not confuse it with a competitor's approach.
- 4. State the result with a real number. An exact figure or a clearly labelled indexed value, never a vague qualifier like "significantly" or "substantially" standing in for a number you have but will not print.
- 5. Attribute it to someone who can be checked. A named client contact, a dated publication, or a linked source, so the claim reads as corroborated rather than self-reported.
That order also matches how a person skims a case study, which is not a coincidence. A structure built to be extracted by a model is usually also the structure that respects a human reader's time.
What Metrics Belong in a Citable Case Study, and Which Ones Hurt It?
Exact figures, percentage changes, and indexed values belong in a citable case study; vague qualifiers and undisclosed ranges hurt it. A 2024 Princeton and Georgia Tech study on generative engine optimisation, published at ACM SIGKDD, tested nine content optimisation methods against roughly 10,000 real queries and found that adding concrete statistics and citations to a passage lifted its visibility in generative engine answers by up to 40 percent, more than any other tactic tested. A specific number is the single highest-leverage edit most teams can make to an existing case study.
| Metric style | Example | Citation behaviour |
|---|---|---|
| Exact figure | "Reply rate rose from 1.1% to 2.4% over 90 days" | High. Specific, dated, and checkable against a stated baseline. |
| Percentage change | "Reply rate improved 118% over 90 days" | Medium-high. Citable, but weaker without the baseline figure alongside it. |
| Indexed or masked value | "Reply rate rose from an index of 100 to 218" | Medium. Useful when the raw number is confidential, still specific enough to quote. |
| Vague range | "Reply rate improved by 20 to 50%" | Low. A model treats a wide range as low-confidence and usually skips it. |
| Unquantified claim | "Reply rate improved significantly" | Very low. Nothing here is extractable as a fact. |
Should a Case Study Live Behind a Gated PDF?
No, not if the goal is AI citation rather than lead capture. A beautifully designed PDF gated behind a form contributes nothing to citation, because a crawler cannot fill out a form, and most will not index a document sitting behind one regardless. The fix is not to abandon gated PDFs, since they still serve a purpose further down a sales cycle, but to publish an open, crawlable web page version first and offer the branded PDF as a secondary download. This mirrors the logic in content types for AI citation: the format that gets cited and the format used in a sales deck do not have to be the same file, but the citable one has to exist somewhere a crawler can reach it.
Does Schema Markup Help a Case Study Get Cited?
Yes, structured data reduces the ambiguity a model has to resolve before attributing a claim to your company. Article or Report schema with a clear author, date, and organisation field tells an engine who is making the claim and when, which matters because an undated case study reads as potentially stale. Consistent naming of the client, your company, and the category across the page also feeds the entity resolution problem covered in entity disambiguation for AI citation: a model that cannot confidently identify who did what to whom will not risk repeating the claim. The mechanics of markup itself are covered in schema markup for AI citation.
How Do You Stay Specific Without Breaching Client Confidentiality?
Use an indexed or percentage figure instead of a raw number, and name the category precisely even when you cannot name the client. "A mid-market industrial distributor" with a specific, dated percentage change is far more citable than "a leading company" with a vague range, because the model still has a real figure to anchor to. Where a client permits it, a named quote from a real contact adds corroboration an anonymised case study cannot match. The goal is to move as far toward specificity as the agreement allows, not to treat every constraint as an excuse for vagueness.
How Does a Case Study for AI Citation Differ From a Comparison Page?
A case study answers a narrower, later-stage question than a comparison page, and the two should not compete with each other. A comparison page wins the definitional and evaluation-stage fan-out ("what should I look for," "how does X compare to Y"). A case study wins the proof-stage fan-out that follows ("has this worked for a company like mine"). SourceRank AI audit data shows the average B2B company is cited in fewer than 5% of relevant AI prompts, and a meaningful share of that gap sits at the proof stage, where a strong comparison page exists but no citable case study backs it up.
Frequently asked questions
What is a case study for AI citation?
A proof asset structured so an AI engine can extract and attribute a specific, verifiable claim, typically led by a fact box with a client, metric, and timeframe, rather than a narrative built to persuade a human reader.
Do case studies actually get cited by ChatGPT or Perplexity?
Yes, but mainly for proof-stage questions like "has this worked for a company like mine," and only when the result is stated as a specific figure with a checkable source rather than a vague range or an anonymised claim.
Should I name the client in a case study built for AI citation?
Naming the client where permitted strengthens the case study, since a named, checkable source is more citable than an anonymised one, but a precise category description paired with an indexed figure remains reasonably citable even without a name.
Does a gated PDF case study hurt AI citation?
Yes, because most crawlers cannot reach content behind a form, and a static PDF rarely carries the structured data an engine relies on anyway. Publish an open web version and keep the branded PDF as a secondary asset.
How specific do the numbers need to be?
As specific as the confidentiality agreement allows. An exact figure is strongest, an indexed value is a reasonable substitute, and a wide range or a word like "significantly" is close to worthless for citation purposes.
Does schema markup make a real difference to case study citation?
It reduces ambiguity rather than adding persuasive weight: consistent Article or Report schema with a clear author, date, and organisation field helps a model confirm who made the claim and when, for a document type that is often left undated.
How many case studies does a B2B site need for AI citation?
Fewer than most content calendars assume. One well-structured, specific, publicly accessible case study per core buyer question usually outperforms a large library of vague, gated ones.
How do I know if my existing case studies are citable?
Run them against the five-part structure above and check whether a stranger with no context could lift the headline result as a standalone fact. A free AI visibility audit also shows whether your proof content is already being picked up across ChatGPT, Perplexity, Gemini, and Copilot, or whether a competitor's case study is winning those questions instead.
Rebuilding a case study library for AI citation is one part of a broader AEO programme. See how the process works, check current plans, or talk to the team about auditing your existing proof content before you commission anything new.