A fair chunk of enterprise buyers now open a chatbot before they open Google. A 2025 Responsive survey of 350 B2B buyers put it at about a quarter who lean on generative AI more than search when they’re vetting vendors. (Responsive sells RFP software, so take it as one vendor’s read.) If your team spent the last two years using AI to pump out three times the content, here’s the uncomfortable follow-up: would any of it get quoted back to that buyer?
Probably less than you’d hope. That’s partly why so many software firms have started hiring a GEO agency to check how they surface in ChatGPT, Perplexity and Google’s AI Overviews. The audits rarely blame the tech stack. More often they point at the writing, and the cure turns out to be surprisingly old-fashioned.
Remember when volume felt like a strategy?
Writing tools landed, and overnight every sequence had nine steps and every calendar had a post a day. The dashboards loved it for a while. Impressions went up. Pipeline meetings got cheerful.
Buyers caught on fast, though. A VP at a large firm learns the tells within a month: phrasing that’s a bit too smooth, a first name dropped awkwardly into line one, and not a single detail you’d only know from having done the job. Give every vendor the same tools and they all end up sounding like one slightly over-caffeinated vendor.
You’ve likely deleted a couple of those this morning.
The research on what AI answers pick up
Nobody has to guess about this part. A team from Princeton and IIT Delhi presented a paper on generative engine optimization at KDD 2024, and they ran the experiment properly. Pages that added statistics, direct quotes and cited sources gained up to 40% more visibility in generated answers on their benchmark. A live test on Perplexity showed gains of up to 37%. Keyword stuffing? It actually scored worse than doing nothing at all.
Notice what those winning ingredients have in common. Someone had to go and get them. A real deployment figure, a customer’s own words, a source a reader can click. A language model can restate the general case all day, but when it needs a specific, it borrows one from whoever bothered to publish it.
A quick, made-up example
Say an IT director at a freight company asks an assistant how mid-market ERP vendors cope with legacy data migration. The reply leans heavily on one competitor’s implementation guide. That guide names the migration tooling, spells out the cutover order, and quotes a customer describing the Saturday night it nearly fell over.
Your forty-odd blog posts about “seamless transformation” never come up. They weren’t filtered or penalized. They just didn’t contain anything worth repeating, which goes a long way to explaining why buyers are asking AI first and still never bumping into some very well-funded brands.
Go where the messy detail lives
Broad efficiency claims bounce off a buying committee. What lands is friction they recognize: the migration from a twenty-year-old system, the compliance review that stalls in week three, the regional team that quietly refuses to log in.
Answers to that live in odd places. Support tickets. The recording of a renewal call that went badly. The product standup where an engineer explains, with some exasperation, why a feature behaves the way it does. Sitting in on raw customer feedback can feel a bit exposing, honestly. It’s still where your best paragraphs are hiding.
Keep using the tools for what they’re good at. Let them transcribe the interviews, sort a thousand tickets into themes, and hand you an ugly first draft. Then do the human bit: rewrite it with the detail only your people know, and make sure every claim traces back to a person or a dataset. Format matters a little too. Headings phrased as buyer questions, named customers (with sign-off), numbers with a date beside them, and one clear answer per page all make it easier for a person, or a model, to lift the useful part.
An outside partner can show you which questions you’re missing and whether you’ve started turning up. They can’t sit in your customer calls for you.
Say the quiet part about your limits
Enterprise sales has always run on trust, and right now buyers assume most of what reaches them was written by a machine. Responsive’s study also found 52% of buyers weigh industry expertise heavily when they make the final call. Expertise is hard to prove in a tagline. It’s easy to prove in a candid paragraph about trade-offs.
Talk about the edge cases. Mention the rollout that took twice as long as planned and what you changed afterwards. The department head reading it is protecting their team’s sanity and their own reputation, and they’ll trust a vendor who admits where the product struggles far sooner than one who claims it never does.
There’s a quieter search benefit as well. When someone asks an assistant what’s wrong with your product, it helps a lot if your own honest limitations page exists for it to quote, rather than a cranky forum thread from 2023.
Half your readers might not be human
Finance turns up in month one, security in month three, and engineering wants a call whenever. A tidy nurture sequence doesn’t survive that. What does hold up is plain, open material: implementation docs, architecture diagrams, unedited reviews.
More and more, the visitor reading those pages is an assistant doing homework for a buyer, and it can’t fill in your form. Keep that in mind when you split your demand-gen budget between gated assets and open ones.
So what’s actually scarce now?
Not efficiency. Every competitor bought the same tools you did. The scarce thing is the engineer who sat through last Tuesday’s escalation and can explain exactly what broke and why.
Getting what she knows onto a public page, in plain language, with names and dates attached, is slower than generating another post. Whose story on your team still hasn’t been written down?
Related: Technical SEO in 2026: Is Your Website Invisible to AI Search?
