Customer stories AI assistants actually cite

Someone new reads your customer stories first now, and it isn't a person.

We've always assumed that the first reader was a person; hopefully a buyer, skimming, deciding whether to keep going, and (even better) sharing. That person still matters as much as ever, but increasingly there is a 'reader' before them: an AI assistant.

A buyer asks ChatGPT, Claude, or Perplexity "who's good at this, and who really uses them?", and the assistant answers by pulling from whatever it can find and trust. If your customer stories are among its trusted sources, happy days. If not, your company might as well be invisible.

We all already know how to write a good customer story. This guide is about what is new: how this new reader behaves, what it changes. This is one part of the customer storytelling job that did not exist until recently.

Two bits of jargon to wrap our heads around first: AEO (answer-engine optimisation) is being the answer an assistant gives. GEO (generative-engine optimisation) is being the source it cites. You do not need the acronyms but everyone is bandying them around. You need the single idea underneath both.

So we need to think about reframing what we do.

When an AI assistant decides whether to cite a story, it is making the same judgement you would make about any source: can I trust this, and can I attribute it? A named advocate at a named company with a concrete result is safe to quote. "A leading enterprise saw significant improvements" is not; there is nobody to attribute it to, so the assistant drops it and cites someone who was specific instead.

So the specificity we have always brought for the human reader now does a second job: it decides whether the machine reader puts you in front of the buyer at all.

That is the whole reframe. Everything you know about good storytelling still holds; it simply matters in a new place, for a new reason, with a new cost for getting it wrong. We don't need to learn to write differently. We all just need to be aware that the machine is now watching (eeek!), and that it rewards exactly the discipline good customer storytelling already has. But there's more, and some of this is tricky.

1. Answer the questions buyers actually ask a machine

Buyers ask assistants in plain language, and differently from how they'd phrase a search: "does anyone in healthcare use this?", "how long until it pays off?", "what did it replace?" These are not keywords; they are questions, and an assistant answers questions by finding content that already contains the answer.

So a story that explicitly answers the real questions; ideally in a short FAQ at the end; hands the assistant a ready-made response in your advocate's own experience. This is new. You were never writing for a reader who asked questions out loud before; now you are, and the ones who anticipate the question get quoted.

2. Permanence is now a distribution decision

Of course a story should be findable. What is new is how unforgiving that has become. A story trapped in a PDF, behind a form, or on a page that moves every site redesign is not merely inconvenient now; it is invisible to the machines doing the finding. An assistant cannot cite what it cannot reach, and a growing share of your buyers reach the assistant before they reach you.

The practical consequence: every story needs a permanent, public web address that does not change, so that when an assistant finds it, cites it, and a buyer clicks, it is still there months later. Where a story lives has become as much a craft decision as how it is written.

3. The part you cannot do in a Word document

This is the new skill; normally beyond the limits of a traditional editorial checklist.

AI assistants read a layer of the page you never see; a machine-readable summary sitting underneath the words, which states plainly: this is the advocate, this is their role, this is the quote, this is the result, this is when they approved it. When that layer is present and correct, an assistant can cite your story accurately and confidently. When it is missing, the assistant has to guess, and usually it just moves on to a source it trusts more.

Writing that layer by hand is a technical job; properly technical, not "learn a new tab in your CMS" technical. It is not a good use of a customer marketer's afternoon, and it is not something your years of craft prepared you for, because it did not exist until recently. This is simply a new requirement of the job that sits outside the job's traditional skill set. If you want to learn JSON-LD and MCP then go for it!

It is also where a purpose-built platform earns its keep. Proofpoints publishes every story with that machine-readable layer built in as standard; the named advocate, their approval and its date, the quotes, the results, all marked up so assistants can read and cite them accurately. It also publishes a plain, public account of what a Proofpoints story guarantees; a real named person, a recorded approval, a permanent home; so buyers and machines alike know what the "customer-approved" mark means. If you're interested, that standard is at proofpoints.com/standards.

You already own the first half. The second half is new, technical, and worth handing to something built for it.

You do not need to become a different writer. The named advocate, the concrete result, the clear sentence; the things you have always done; are exactly what the machine reader rewards, and now they decide whether you are cited at all. What is new is three-fold: write to the questions buyers ask a machine out loud; treat permanence as a distribution decision, not an afterthought; and accept that there is now a technical layer beneath the page that you cannot type in Word, and probably should not try to. Do the half you are expert at. Let a platform do the half that is new.

What's the difference between AEO and GEO? AEO (answer-engine optimisation) is about being the answer an AI assistant gives. GEO (generative-engine optimisation) is about being the source it cites. They overlap heavily, and the same qualities; specificity, structure, real attribution; serve both.

I've been doing customer stories for years. Is any of this actually new? The craft is not; you already write specific, named, results-led stories. What is new is that a machine now reads them first and decides whether to put you in front of the buyer, and that there is a technical layer beneath the page that determines whether it can cite you accurately. The skill is the same; the second reader, and the new layer, are what changed.

Do I need a developer for this? For the editorial choices, no. For the machine-readable layer beneath the page, yes; which is where a purpose-built platform saves you the job entirely.

Will writing for machines make my stories worse for humans? No; the opposite. Everything that makes a story citable by a machine (real names, concrete results, clear sentences, straight answers to real questions) is what makes it convincing to a person. You are not writing twice.

How do I know if assistants are citing my stories? Ask them. Put a category question to ChatGPT, Claude, or Perplexity as a buyer would, and see what surfaces and what gets cited. Do it periodically; it is the simplest way to tell whether your customer evidence is in the room when buyers ask.

Robin Hamilton is the founder of Proofpoints and writes Proof Positive, a fortnightly publication on the editorial judgement underneath customer storytelling. Twenty-five years in B2B customer advocacy, both sides of the fence, dozens of programs.