AI customer stories, without the slop

We all now have access to the same models. That's a blessing and a problem. When the whole profession can generate a competent customer story in ninety seconds it's worth very little because it reads like every other one, and buyers have already learned to skim past it.

This is slop: AI output mistaken for substance. Fluent, grammatically correct, on-topic, and somewhat hollow. Many AI-assisted customer stories being published right now are slop, or slop-adjacent, and their authors often cannot tell. Slop reads fine until you have read a hundred of these.

This guide is about creating great stories in the age of AI. Not by avoiding AI; that argument is over, but by using it the way a good editor uses a fast, tireless, slightly unreliable junior writer.

Seven things separate a story worth reading from slop.

AI can generate a story from almost nothing, and that is precisely what makes it dangerous: the results are enchanting. When I was testing Proofpoints processes I got bored halfway through filling in a story form and just answered "I like cheese". It generated an 1,100-word story that looked great for all of five minutes. Fluent, structured, plausible - and made of nothing. Scary.

That's the trap. Ask a model to create rather than shape - "write a customer story about a logistics firm that improved efficiency" - and you get the same thing at production scale: invented quotes, invented numbers, perhaps even an invented person. It looks great at the outset, and it is a disservice to everyone.

So every great story starts from something real: an interview, a transcript, a recorded call, a set of notes, a campaign response in the customer's own words. The model's job is to shape that raw material, never to conjure material that never existed - and holding that line takes both discipline and brutal editing.

A model with no brand context defaults to the average of everything it has ever read; which is the very definition of generic. If you want output that sounds like you, the model has to know what "you" means: your voice, your terminology.

This is the difference between pasting a transcript into a chatbot and running it through a system that is grounded in your actual brand guidelines. The first gives you 900 words of slop with your topic in it. The second gives you something that at least starts in the right register.

A single model has a single set of habits; the same sentence rhythms, the same structural tics, the same reach for the same adjectives. Read enough output from one model and you can spot it, which means so can everyone else.

Generate more than one draft, ideally from more than one model, and choose. Not because one is "right," but because seeing three versions of the same story shows you the shape of the thing underneath; the parts that survive all three are the real story, and the parts that differ are where your judgement is needed.

One rule matters more than all the others: the AI draft is the beginning of your work, not the end of it.

The moment we treat the first fluent draft as finished, we are probably publishing slop, or at least something that's not as good as it could be.

What a human editor adds, and a model cannot:

1. ·Knowing which moment IS the story. A model will faithfully summarise the whole interview. It can't tell that the throwaway line in minute nineteen is the actual story and everything else is scaffolding.
2. ·Cutting what is true, but boring. Models are reluctant to throw things away. Good editors are ruthless.
3. ·Hearing when a quote sounds wrong. The model will happily smooth a real person's words into corporate mush, or will leave it as broken English. You have to hear the difference between what they said and what they meant.

This is what Editor-Enhanced Content means in practice: AI handles volume and structure; a human handles judgement. Skip the judgement and all you have done is automated the production of forgettable content.

This is a tightrope to walk. Often there is a temptation to let the model "improve" the advocate's language until it no longer sounds like this specific living breathing human being. Real people say slightly awkward, memorable things. Slop replaces those with smooth, quotable-looking sentences that no actual person has ever said aloud.

When you edit, guard the specifics to keep the balance of eloquence and authenticity. Keep the odd phrasing when it's the right thing to do. A quote that reads a little rough and real beats a quote that reads polished and invented, every time.

Now it seems that the first "reader" of your customer story is not a person; it is an AI assistant answering a buyer's question about who to trust. If your story is going to be cited rather than skimmed past, it has to be legible to that machine: a clear named customer, a real result stated plainly, structured so the key facts are extractable rather than buried in prose.

This is not the same as writing for the old SEO. It means being specific and being structured; the named person, the named company, the concrete outcome, in a form a model can quote accurately. Vague, hedged, everyone-sounds-the-same slop is exactly what these systems cannot use. Specificity is now a distribution strategy, not just a craft and voice preference.

The final test is the simplest, and the one everyone skips because they are tired, it's week nine and they cannot 'see' the story anymore. Try to read the finished story as a buyer, not as its author. Would you read past the first paragraph? Would you believe it? Would you remember it tomorrow?

Does this do more harm than shipping nothing? Probably not; with a caveat. Things have moved on from when the business of case studies was a volume game. Clients would need 20 4-pagers and 10 2-pagers and success was measured in the stack of stories they had on their desk at the end of the year. Perhaps it's still like this in some orgs. Yet we need to be careful; in 2026 volume is easy, value less so, and every hollow story you publish has the potential to teach your audience to ignore the next one.

AI has not replaced the customer-story craft. It has removed the excuse of not having time for it. The drafting is now cheap, which means the judgement is the entire job; and judgement is the difference between great stories and slop.

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.