Key Takeaways
- Write one sentence that survives being lifted out of the page. Assistants quote fragments, not documents. If your result only makes sense after three paragraphs of setup, it will never be repeated.
- Put the client type first, before the result. The reader has to recognize themselves before the number means anything, and the assistant has to match them to you before it recommends you.
- State the starting point, not just the finish. A result with no before is unverifiable and reads as marketing. Where they started is what makes the change credible.
- Put the case study in text on a page, not in a PDF or an image. The best proof most businesses own is trapped in a downloadable deck or a screenshot of a testimonial, which is the same as unpublished.
- Keep the claim smaller than what you can defend. Assistants hedge around anything that reads as overreach, and a modest specific result outperforms an impressive vague one.
Most case studies are written for a reader who has already decided to trust you. They open with background, build to a result, and end with a quote. That structure works in a proposal and fails completely here.
An assistant does not read your case study the way a prospect does. It looks for a passage that stands on its own, and it repeats that passage or it repeats nothing. Everything before the passage is context it will not carry with it.
So the job is narrower than writing a good story. It is writing one sentence that survives being lifted out of the page, and then earning it underneath.
How do I write a client case study that AI will quote?
Write the result as a single self-contained sentence that names the client type, where they started, what you did, what changed, and how long it took. Put that sentence in plain text near the top of a page, and support it below.
Self-contained is the whole test. Read your best result out loud with no setup. If a stranger could not tell who it was about or what it was measured against, it is not quotable, however true it is.
This is the same principle as what makes an answer quotable, applied to the one kind of page where businesses hold their strongest evidence and publish it in the least usable form.
The five parts of a liftable result sentence
Five segments do the work. Drop any one and the sentence stops being able to travel on its own.
Who
A regional literacy nonprofit with eleven staff
Lets an assistant decide whether the reader is the same kind of buyer.
Starting point
whose monthly giving program had not been reviewed in six years
Gives the result something to be measured against.
What was done
rebuilt its lapsed donor sequence and its ask amounts
Names the work, so the sentence describes a service and not a mood.
What changed
and recovered a meaningful share of donors who had stopped giving
The outcome. Specific enough to matter, honest enough to defend.
Over how long
over about ninety days.
Sets expectations, and stops the claim reading as instant.
The example wording is an illustrative sentence written to show the five parts, not a client result. Any real version carries a real client type, a real starting point, and an outcome you can defend. Every segment and every job described here also appears as text in the section above and as a row in the table below.
Notice how little of that is about you. Four of the five segments describe the client and the situation, which is exactly the ratio that gets a business recommended rather than merely described. The reasoning behind it is in how to make client results legible to AI.
The four case study formats that never get quoted
Four formats account for most unquotable proof. All four are common, and three of them look professional, which is why they persist.
The narrative. Twelve hundred words with the result in paragraph nine. Nothing in it stands alone, so nothing gets lifted. If you like the narrative, keep it, and put the sentence at the top.
The logo wall. A grid of client marks with no words. It says nothing readable at all, which is the argument in turning one client story into proof AI can repeat.
The gated PDF. Your best material, behind a form, in a format that is inconsistently read even when it is public. A form is a wall to a machine as surely as to a person.
The testimonial screenshot. A picture of praise. The words inside an image are not text on a page, and the claim exists only inside the image, which breaks the same rule we hold ourselves to on this blog.
| Segment | What it settles | The version that fails | Why it fails |
|---|---|---|---|
| Who the client was | Whether the reader is the same kind of buyer | "A client came to us" | Matches nobody, so it cannot be recommended to anybody |
| Where they started | What the result is measured against | "They were struggling" | No baseline, so the change cannot be judged or verified |
| What you did | Which service this was | "We worked together closely" | Describes a relationship rather than a service |
| What changed | Whether it worked | "Results were transformational" | Unverifiable language gets hedged around or dropped |
| How long it took | What a buyer should expect | Left out entirely | An undated result reads as either instant or ancient |
What to do when you cannot name the client
Describe them instead. Confidentiality almost always covers the name, not the shape of the work, and the shape is where nearly all the matching value sits.
"A regional literacy nonprofit with eleven staff" does more work than a logo, because it lets a reader at a similar organization recognize themselves. A named client that a reader has never heard of and cannot place does less.
Get permission in writing when you can, and ask for the small version rather than the big one. Most clients who will not agree to a full case study will agree to a two sentence description with no name in it, and two sentences is all the format needs.
Where a real number exists and you can defend it, use it. Where it does not, say what changed in plain words rather than inventing a percentage. Made-up precision is worse than honest description, and it is the fastest way to lose the trust these systems extend to specific claims.
Where to put it, and what to do next
Put it in text, on a page that can be read without logging in, near the top. That single decision does more than any wording change.
A dedicated page per client story works best, with the sentence as the opening line and the supporting detail beneath it. A combined results page works too, as long as each story leads with its own liftable sentence rather than sharing an introduction.
Then let it be found. Google's description of how AI features in Search work makes clear these answers are built per question from retrievable material, and OpenAI's documentation for its crawlers is a reminder that a page has to be both reachable and permitted before any of this applies. A case study on a page blocked from crawling is a case study nobody will ever quote.
Give it time before judging the result. Systems that read the live web can pick up a new page within weeks, and ones leaning on stored copies refresh on their own schedule, so a re-check four days later measures nothing.
One honest limit. A great case study does not make you the recommendation on its own. It makes you the business an assistant can describe accurately, which is the step almost everybody is missing, and the step everything else depends on.
If you would rather have the whole thing run and read for you, the free check asks 25 real buyer questions from your niche through ChatGPT at $0 and emails a plain-English readout within the hour. It runs as soon as you ask for it.
Questions we hear the most
How do I write a case study that AI will quote?
Write the result as one self-contained sentence naming the client type, the starting point, the work, the outcome, and the time it took. Put it in text near the top of a page, then support it underneath.
How long should a case study be?
Long enough to be credible, with the quotable part at the top. Two hundred words of specifics beats two thousand words of narrative, because only the specific part will ever be lifted.
Do AI assistants read case studies in PDFs?
Sometimes, and unreliably. A PDF behind a form is invisible, and a downloadable deck is far less likely to be read than the same words as text on a normal page.
What if I cannot name the client?
Describe them instead. The sector, the size, and the situation carry almost all of the matching value, and an unnamed engagement with real specifics is more quotable than a named one with none.
Are percentages better than plain descriptions?
Only when they are real and you can defend them. An unverifiable percentage invites hedging, while a modest concrete outcome stated plainly gets repeated more often.
How many case studies do I need?
Two or three that are genuinely specific beat a dozen that are not. What matters is coverage of your main client types, so a reader recognizing themselves finds a matching story.
How do I check whether AI repeats my client results?
Ask an assistant about your business and about your specialty, then look for your own phrasing coming back. Repeated wording is the clearest sign your material was read and used.
Does the AnswerHalo free check look at my case studies?
Indirectly. It records who gets named for 25 real buyer questions in your niche and whether you are among them, which tells you whether your proof is reaching the answers. It costs $0.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →