Your buyers' AI agents are reading your website

They are, and they're reading it for facts. A buyer's agent wants to know what you cost, what you connect to, what you don't do, how long it takes to get live and how you answer the security questions, and if your site doesn't say, it fills the gap from somewhere else or moves on to a vendor whose site does.

That changes what a startup's website is for. For years it was a brochure with a "Book a demo" button. Now one of its most important readers is a machine with a shortlist to fill and no patience for adjectives.

What the IDC numbers say

The figure that did the rounds in August was that eight in ten B2B tech buyers now use AI agents to buy. I went back to IDC's own slides, and the wording is a little different. In IDC's 2025 B2B Technology Buyer Survey of 406 tech buyers, 80% said they'll use AI guided selling assistants as a go-between with vendors, for comparisons, configuration, RFPs and quotes. And 74% said they'll hand buying tasks to AI agents with set parameters such as price, availability and fit. That's stated intent rather than measured behavior, which makes it a forecast with a short fuse.

Since then, the behavior has started showing up. IDC's September 2026 data says 68% of buyers use AI always or often to compare products and features across vendors during evaluation. And the share using vendor websites as an evaluation source has fallen from 66% to 37%.

That second number looks like a reason to stop caring about your website. I read it the other way. Fewer humans are reading your site, and more of them are reading a summary of it written by a machine. You aren't in the room when that summary gets written, so the pages it's built from have to do the selling.

What an agent is looking for

Put yourself in the agent's position. It's been told something like "find three tools for a firm our size, within budget, that work with HubSpot and can be live this quarter." Your tagline is no use to it. It wants numbers, names and yes-or-no answers, and it'll take them from wherever it finds them: your site, your docs, review sites, an old forum thread.

IDC's advice to vendors points the same way. AI agents favor content that's well structured, machine-readable, cited by trusted third parties and consistent across credible channels. In plain English: write the facts down, in public, and make sure everyone else is saying the same thing.

The checklist

This is what I'd want on a B2B startup's site before the next agent comes calling. Most of it is a page or a section, and none of it needs a redesign.

1. Pricing, with numbers. A starting price, what pushes the price up, and what costs extra. In TrustRadius's 2025 buyer research, transparent pricing was the top thing software buyers wished they could change. If your deals vary, publish a range and the factors behind it. "Contact us" gives an agent nothing to compare, and it won't wait for you to call back.
2. Integrations, named and explained. Every integration by name, what data moves and in which direction, whether it's native or through a connector, and which plan includes it. "Integrates with 200+ tools" is an adjective dressed up as a fact.
3. What you don't do. Who you're not a fit for, limits on users, data volume or regions, and the features people ask about that you don't have. It feels risky to publish. I think it's the most commercial page you'll write, for the reasons in why honesty is the most commercial move in sales: a buyer who finds your limits after signing becomes a churn number.
4. Implementation time. How long a typical customer takes to go live, what they have to do themselves, and what pushes it to the long end of the range.
5. Security answers, in public. Your SOC 2 or ISO 27001 status (including "audit in progress" if that's the truth), where data is stored, SSO, sub-processors, and how you use AI on customer data. A trust page behind a form doesn't exist as far as a machine is concerned.
6. Proof of fit by segment. Customer examples grouped by size and industry, with results the customer has agreed you can publish. Buyers and their agents both look for "companies like us".
7. Text a machine can read. Answers in plain text on public pages, not in PDFs, images of tables or gated decks. I once uploaded a polished one-pager to a content library and ended up emailing the PDF three days later because I couldn't find it. If I can't find my own PDF, a buyer's agent has no chance with yours.
8. The same facts everywhere. Your site, docs, G2 or Capterra profile and LinkedIn page should agree on pricing, integrations and who you're for. When they disagree, the agent picks one, and you don't get a vote.

Test it the way a buyer would

Once a quarter, open ChatGPT, Claude, Gemini and Copilot and ask the question your buyer would ask. Name your category, describe a company like your best customer, and ask for a comparison that includes pricing, integrations and time to go live. Then read what comes back about you. Whatever's wrong or missing is your to-do list, and it usually points at a page you haven't written yet.

If the answers show nobody can tell who you're for, that's a positioning problem, and editing the website won't fix it on its own. That's revenue consulting territory. If the answers are fine but your team doesn't know what to do with buyers who arrive this well briefed, that's where our AI for revenue teams work comes in.

One more item if you've put an AI chat agent on your site and you sell into Europe. The EU AI Act's transparency rules have applied since August 2, 2026, and people talking to a chatbot have to be told it's AI unless that's obvious from the context (Goodwin's summary is a clear read). I'm all for it. A bot that says "I'm an AI assistant, here's our pricing page" will do better than one pretending to be a person.

Where the humans come back in

All of this gets you onto the shortlist. Winning from there is still selling. More and more buyers now arrive already introduced to you by a machine, carrying a summary of what you do that you didn't write and didn't get to shape. The facts on your website are the best say you have in that summary.

What happens next is still a conversation between people. The agent can tell a buyer what you cost and what you connect to. It can't tell them which of their problems to fix first, and the buyer often can't either until someone asks better questions than the RFP did. I wrote about that in what "adding value" means when buyers ask AI first, and at book length in Curiosity.

On the bright side, the agent is the first visitor in years who'll read your website all the way to the bottom.