A case page that an AI assistant cites answers one question, and it answers it out loud. It names the sector, the starting point, the approach, and the result in plain language, without the marketing jargon. Models like ChatGPT, Perplexity, and Gemini pull facts from several sources at once. The clearer your fact stands, the better the odds it ends up in the answer.
Most B2B case pages aren't written that way. They open with a bit of scene-setting, hide the result in the third paragraph, and use words like "leading" where a number should be. We see a growing number of marketers wondering why their best case never shows up in an AI answer.
Sound familiar? Then the structure is the problem, not the content.
An AI assistant doesn't read your case page the way a human does. It looks for facts it can lift into an answer to a business question. If the answer sits there implicitly, buried between mood and superlatives, the model grabs a different source.
Writing case pages for AI is about extractability. Large language models process text in tokens and perform better with explicitly structured information than with implicit context. A sentence like "we helped an MSP with lots of employees streamline their processes" holds not a single hard fact to cite.
Our experience is that vague marketing language and superlatives without concrete facts are harder to cite than neutral, verifiable phrasing. The AI findability of your content ties directly to that. No fact, no citation.
An assistant picks a handful of sources per answer and summarizes them. There's no second page. If you're not in that handful, you don't exist for that question.
The big assistants increasingly show where they got it. Perplexity displays explicit source citations with clickable links to the pages it quotes. ChatGPT has had a search function since 2024 that cites current web sources. Google Gemini points to underlying web sources in its AI answers.
That means something for your case page. The question is no longer just whether you rank high in Google. The question is whether your case delivers the fact the model needs.
AI assistants often pull those facts together from several sources at once. Explicitly stated facts are easier to extract than facts you have to infer from context. That's the core of Perplexity source citation and Gemini content ranking: the model rewards the page that serves up the answer ready to go.
We see that case pages with a clear problem-approach-result structure line up better with how assistants build answers. Makes sense. That order is exactly how a human asks the question too.
How do you write a case study an AI can cite? You build it in three blocks, each with its own heading. Every block contains facts, not mood.
The citable structure looks like this:
Our experience is that a case page answering one central question gets cited more easily than a page that touches ten topics at once. So pick one question per case. "How do you shorten the sales cycle of an ERP implementation?" is a question someone actually types into ChatGPT.
The case page structure then follows the logic of the answer you want to win. Question up top, fact underneath. Nothing hidden.
Structured content for B2B starts with sentences that stand on their own. A model that plucks one sentence from your page has to get that sentence without having read the previous three paragraphs.
Compare these two.
The second sentence holds a fact, a time frame, and a direction. The first holds air. For making citable cases the rule is: write it so a journalist would repeat it without checking.
Avoid words that point back to earlier sentences. "This delivered" or "that caused" forces the model to reconstruct context. Repeat the subject instead. It reads a little clunkier for a human, but much cleaner for generative engine optimization.
Cut the superlatives too. "Leading", "revolutionary", "the best". A model can't do anything with them, and a reader doesn't believe them anyway. A number convinces, an adjective doesn't.
Structured data helps search engines and AI systems interpret the meaning of your page better. For a case page you put that context explicitly in the code, so a crawler doesn't have to guess what's what.
The schema.org types that matter here:
Use Review and AggregateRating only for real reviews. Never invent a rating in your markup. It falls apart and works against you. A GEO content strategy for B2B leans on honest, verifiable signals.
Schema is no magic wand. It won't make your fact citable if the fact isn't there. It does reinforce what's already there.
Three mistakes keep coming back when writing case pages for AI.
Mistake one: the result only at the end. Structuring it like a suspenseful story, with the outcome as the closing scene. A model doesn't read to the end. Put the result at the top of the result block.
Mistake two: not a single number. A case without a number is an anecdote. "Client was very happy" gets cited by nobody. If you have no hard figure, name the concrete change: a process that disappeared, a step that fell away.
Mistake three: ten topics on one page. A case that covers strategy, tech, culture, and budget all at once wins no question at all. For making citable cases you trim it down to one question and one answer.
Sounds strict? It's mostly honest. A page that wants to say everything says nothing a model can use.
Run through these points before you publish. Every check raises the AI findability of your content.
That last point gets forgotten a lot. AI crawlers often don't run JavaScript. If your case page only loads the text client-side, an assistant sees a blank page. All your work on the structure is then for nothing.
The case page structure above works for generative engine optimization because it follows the logic of the answer. Question, context, fact. In that order.
A citable case page answers one question with explicit facts: sector, starting point, approach, and a concrete result. It avoids superlatives and back-referencing words, so a model can lift individual sentences without reading the rest.
Pick the question your ideal client would type into ChatGPT or Perplexity about the problem you solved. One question per case. Phrase that question in the heading and answer it in the first paragraph.
Schema isn't a requirement, but it helps. Structured data makes the meaning of your page more explicit for AI systems. It reinforces a good fact, but it never replaces one. Without a fact on the page, markup does little.
Want to know whether your current case pages already show up in AI answers? With our free AI visibility scan you see per assistant where you're mentioned now and where you're not. If you'd rather walk through your cases together, an audit call can be a good starting point.