Key Takeaways
- Write one claim, not one story. A page of narrative gives an assistant nothing to lift. One sentence carrying who, what changed, how long it took, and where it can be checked gives it everything.
- Name the client type, not the client. You rarely need permission to say a 20-person accounting firm. That is specific enough to be useful and vague enough to stay inside most confidentiality agreements.
- Attach a period to every number. A result with no timeframe reads as a boast. The same result across one quarter reads as a fact, because a reader can judge whether it is plausible.
- Put the claim somewhere you do not own. The fourth piece is verification, and it is the one nearly everybody skips. A claim only you have ever made is the weakest kind of evidence there is.
- Round down and keep the caveats. Understated claims survive being repeated. Overstated ones get contradicted by the next source an assistant reads, and contradiction is what gets you dropped.
You have one client story you tell in every sales call, and it lands every time. The problem is that it lands because you are in the room. Written down, it becomes twelve paragraphs of narrative that nothing can pick up and carry.
Assistants do not read stories, they lift claims. A claim is a compact, self-contained sentence that survives being pulled out of the page it came from. Most case studies contain one somewhere and bury it on the fourth screen.
This post is the extraction job: how to get the claim out of the story, which four pieces it has to carry, and where to put it so something other than you will say it.
How do you turn one client story into proof AI can repeat?
Compress it into a single sentence carrying four things: what kind of client it was, what measurably changed, over what period, and where a reader can check it. Publish that sentence in plain text near the top of a relevant page. Then get one source you do not own to say the same thing.
The compression is the hard part, and it is hard because the story is genuinely more interesting than the claim. All the texture you like about it is exactly what makes it unliftable. A claim has no texture on purpose.
The fourth piece, verification, is where nearly everybody stops. A claim that exists only on your own website is a claim about yourself, and it gets weighted accordingly. The same claim echoed once anywhere else stops being self-report and starts being evidence.
The four pieces a repeatable claim needs
Who, what changed, over how long, and verified where. Drop any one and the claim stops holding, in the sense that a careful reader cannot decide whether to believe it and a cautious assistant will not repeat it.
Each missing piece fails differently. No client type and the result could belong to anyone, including someone nothing like your buyer. No number and there is no result, only a mood. No period and the number cannot be judged. No verification and the whole thing rests on your say-so.
The fourth block is drawn as an outline because verification is the piece most claims leave out, and a chain with a gap in it does not hold. The example wordings are illustrative, written to show the shape of a weak and a strong version rather than quoting any client.
Written out, because nothing here may live only inside a figure. Piece one is who, a named kind of client specific enough to picture: weak version "a client", strong version "a 20-person accounting firm". Piece two is what changed, one outcome stated as a before and an after: weak "great results", strong "cut month-end close from 9 days to 4". Piece three is over how long, so the result has a rate attached: weak "quickly", strong "across one quarter". Piece four is verified where, a place a reader can go to see it is real: weak "nowhere", strong "a named review or the client saying it publicly". All example wordings are illustrative.
How to write the sentence itself
Put all four pieces in one sentence and stop. If it needs a second sentence to make sense, it will not survive being lifted, because the second sentence is the one that gets left behind.
Assembled, the pieces above read as one line: a 20-person accounting firm cut its month-end close from nine days to four across one quarter, and said so in a review you can read. That sentence is dull, and dull is the point. It carries everything and needs nothing around it.
Write it in the buyer's words rather than yours. Month-end close is what the client calls it. Financial process optimization is what a consultant calls it, and nobody types that into anything. Google's guidance on creating helpful content makes the same point about writing for the person with the problem, and it applies with more force when a machine is doing the reading.
| How it is written | Pieces present | What a reader can check | Can it be lifted |
|---|---|---|---|
| Transformational results for our clients | None | Nothing | No |
| We helped a client speed up their close | Who, loosely | Nothing | No |
| Cut a client's month-end close by more than half | What changed | The size of the change | Weakly |
| A 20-person accounting firm cut its close from 9 days to 4 | Who, what changed | Client type and both numbers | Yes |
| The same, across one quarter, with the firm saying so in a review | All four | Everything, including who said it | Yes, and it holds up |
Notice that the fourth row is already liftable. The fifth row is what stops it being contradicted later. Those are two different jobs, and skipping straight to the fifth is why so many small firms have nothing quotable on their site at all: they are waiting for permission they could have worked around three rows earlier.
Where the claim has to live
Near the top of a page about that kind of work, in plain text, not inside an image or a slide deck or a video. A claim rendered as a graphic is invisible to the thing you are trying to reach, which is a mistake we see constantly on otherwise well-made sites.
Then get it said once somewhere else. A review naming the outcome, a client mentioning it on their own site, a podcast description, an association case note. Reviews are the most common route and the most misunderstood one; the mechanics of asking are in how to ask for reviews AI can find and trust.
If you want the claim to be machine-readable as well as human-readable, the review vocabulary at schema.org exists for exactly that, and labels what is already on the page. It is a small improvement on top of the writing, never a substitute for it, and the broader case for making results legible is in making client results legible to AI.
Doing this without overstating anything
Round down, keep the caveats, and never write a number you could not defend if the client were reading over your shoulder. Understatement is not modesty here, it is durability. A claim that survives scrutiny keeps getting repeated; one that does not gets contradicted and dropped.
Be careful with reviews specifically. The Federal Trade Commission's guidance for marketers on soliciting reviews is direct about what is and is not allowed when you ask for them. Asking a happy client to describe what actually happened is fine. Steering the wording toward a number they did not give you is not, and it is the kind of thing that ends up costing more than it ever earned.
Keep the qualifier if there is one. If the result depended on the client having a decent finance team already, say so. Conditional claims are more credible than absolute ones, they are more useful to a buyer deciding whether they are the same kind of client, and an assistant repeating the condition alongside the number is doing your qualifying for you.
One last discipline: date it. A result from three years ago is fine as long as it says three years ago. A result with no date attached silently ages into something that looks stale or, worse, looks like it was written to avoid the question. What a run of test questions tends to surface about which claims get repeated is collected in what 25-question test runs keep showing us.
Our free check runs 25 real buyer questions from your niche at $0 and emails a plain-English readout within the hour, including who got named instead of you and what was said about them. It runs as soon as you ask for it.
Questions we hear the most
How do you turn a client story into proof AI can repeat?
Compress it into one sentence carrying four things: what kind of client it was, what measurably changed, over what period, and where a reader can verify it. Then publish that sentence somewhere plain, and get one source you do not own to say the same thing.
What if my client will not let me use their name?
Name the type instead. A 20-person accounting firm, a regional manufacturer, a solo therapist scaling to three. Category plus size plus context is usually specific enough to be useful and almost always inside what a confidentiality agreement allows.
Do case studies work better than testimonials?
They serve different jobs. A case study carries the numbers and the sequence; a testimonial carries a human saying it happened. Assistants lift compact factual claims most readily, so the sentence matters more than the format around it.
Can I use a result I cannot prove?
You can state it, but expect it to carry no weight and to create risk. An unverifiable number that another source contradicts is worse than no number, because contradiction is one of the fastest ways to get left out of an answer.
Should I put numbers on my website if they are small?
Yes. A modest, specific, checkable result outperforms a large vague one every time. Four days saved on a monthly close is repeatable. Transformational results are not, because there is nothing in them to repeat.
How many client stories do I need?
Fewer than you think, done properly. Three claims that each carry all four pieces do more than a dozen narrative case studies, because an assistant can lift a complete claim and cannot lift an incomplete one.
How do I know whether AI is repeating my proof?
Ask the questions your buyers ask and read what the answers say about you, not just whether you appear. If the wording of your own claim comes back, it is being lifted. If you are named with no detail, it is not.
What does the AnswerHalo free check include?
It costs $0. We ask ChatGPT 25 real buyer questions from your niche, then email a plain-English readout within the hour: how many answers named you, who got named instead, and the first fix. It runs as soon as you ask for it.
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 →