Increasingly, prospects aren’t starting their journey on Google or your homepage. They’re opening tools like Gemini, ChatGPT or Perplexity and asking a simple question: “Who should I be talking to?” Gartner puts GenAI use at 45% of recent B2B purchases; G2 found AI chatbots are now the single biggest influence on software shortlists. The challenge for marketers now isn’t just about getting found. It’s making sure AI can find enough evidence to recommend you in the first place!
Why LLMs need your customer evidence specifically
An LLM answering “which supplier should I consider” doesn’t rank pages, it retrieves information, weighs it and synthesises a response. It needs something to weigh. A vague claim gives it nothing to work with. However, a named customer, a measurable result and an attributed quote give it a fact it can extract, compare and cite. That’s the whole game: stop writing marketing copy and start producing material a machine can lift and repeat with confidence.
Four things make a customer story usable this way.
1. Numbers are what get extracted
“We significantly improved performance” can’t be quoted by anything. “Qualified opportunities increased by 42% in six months” can. Academic research into generative engine optimisation found that adding relevant statistics, quotations and citations to source material improved its visibility in AI-generated responses, with some approaches producing relative gains of around 30-40% on one visibility measure.
As always with AI, there’s a nuance here. This doesn’t mean adding a percentage to a webpage suddenly makes you visible. But it does reinforce a broader pattern we’re already seeing: AI tools prefer, and prioritise, specific, verifiable facts over vague marketing language.
2. Named entities are what get connected
LLMs work by linking entities: your business, the customer’s business, their sector, their problem, your solution. A logo with no name attached gives a machine nothing to connect. Name the customer wherever permission allows it, put the name in the copy itself rather than trapping it inside a logo, state the sector and the problem in plain text, and use the same terminology everywhere.
Consistency is also key: using “customer advocacy” on one page and “customer success marketing” on another splits what should be one connected fact into two the machine can’t reconcile.
3. Attributed quotes are what get cited
The same GEO research found that relevant quotations performed strongly for visibility. Why? Because a quote is inherently attributable: it comes from a named person with a title and a company, which is exactly the kind of source-backed claim generative systems favour when constructing an answer.
“Great agency, enjoyed working together” is unusable. A Marketing Director explaining precisely what changed, with their name and company attached, is something an AI system can cite as evidence rather than paraphrase as opinion.
4. Third-party repetition is what gets crawled
Your own website is a minority source. AirOps analysed over 21,000 brand mentions across ChatGPT, Claude and Perplexity for commercial discovery queries and found 85% came from third-party domains, not the brands’ own sites, with real variation between platforms. A result that only exists on your case studies page is invisible to most of what an LLM actually reads. The same result covered by a trade publication, referenced by a partner, cited by an analyst, entered for an award or discussed at a conference gives the machine multiple independent sources agreeing on the same fact, which is corroboration, and corroboration is what these systems are built to weigh. Semrush’s analysis of 1,000 domains also found stronger backlink authority correlated (moderately) with greater AI-answer visibility, another reason the story needs to leave your own domain.
Build once, feed it everywhere
A case study that lives only as a page in your resources section is doing a fraction of its job, and the practical implication is bigger than better formatting: it needs to stop being a one-off asset and start being the output of a proper customer advocacy programme. That distinction matters for every B2B marketer, not just those with a dedicated advocacy function.
An isolated case study asks one question: can we get a story from this particular customer? A programme asks four: where are our happiest customers, what can we prove, who is willing to advocate for us and how do we get that evidence out everywhere an LLM might be reading. The second set of questions produces a steady supply of named, quantified, quotable proof. The first produces one page that ages the moment it’s published.
Most B2B marketing teams already do the hard part: they interview customers, capture results and write the story up. The only thing missing is treating that work as a repeatable system rather than a single deliverable. Capture the number, the name and the quote once, store them as a single source of truth, and push the same fact into every channel an LLM might actually be reading from: your sales deck, your customer’s own channels, press coverage, award entries, analyst briefings, partner content.
Same customer, same number, same quote, everywhere. One inconsistent figure between your website and a press release is a fact machines can no longer trust; ten consistent mentions across independent sources is exactly the corroboration they’re built to surface. That consistency doesn’t happen by accident: it happens because someone owns the evidence bank and reuses it on purpose.
Your next prospect might not start on your homepage. They might start by asking a machine who’s worth talking to. Give it a specific number, a named customer and an attributed quote to find, and repeated in enough places to trust.