We Sent an AI Buyer to 10 Cybersecurity Sites. It Couldn't Answer a Single Question.

Open ChatGPT and ask it to compare three vendors in your category. Watch what it does. It visits their websites, reads what it can, and builds a shortlist out of whatever it finds. If it can't find your pricing, your certifications, your integration list, it doesn't email you to ask. It moves on to the vendor whose site answered.

So we ran the experiment properly. We took 10 cybersecurity companies and pointed an AI buyer-agent at each one with the same 10 questions a real security buyer asks before a first call. Pricing. SOC 2 status. Integrations. Deployment time. The competitive question. 100 questions in total.

One number stopped us: zero. Not a single one of the 100 questions got a complete, act-on-it answer from the vendor's own website. Not low. Zero.

44 questions got a partial answer, the kind where a human would still be guessing or stitching together three pages to be sure. The other 56 were dead ends. The answer simply wasn't on the site, and a buyer would have to book a sales call to get it. The median site scored 23 out of 100 on agent-readability.

We're not naming the 10. I'll explain why below, and why anonymizing actually made the data more honest.

This isn't a fluke in our sample

We already knew the shape of this before we ran the 10. Our broader benchmark this year scored more than 200 B2B SaaS vendors across 10 verticals on the same question: can a buyer's agent find answers on your site, or does it leave empty-handed? Across 2,331 buyer questions, 37.4% were flat-out unanswerable on the vendor's own site. Only 2.3% produced a direct, confident, usable answer.

Cybersecurity scored worse than the average. 42.4% of cyber buyer questions had no answer on the vendor's site, dragged down by enterprise-gated pricing and shallow compliance disclosure. Our 10-vendor cut came back harder still, at 56% dead ends. Same pattern, sharper, because we hand-checked every answer instead of sampling. We ran an earlier version of this on EOR vendors and got the same result: agents are the new top of funnel, and no vendor was ready.

The scorecard

We scored each company on a simple rule. A fully answered question is worth a full point, a partial answer is worth half, a dead end is worth nothing. Ten questions, so a perfect score is 100. Here's where the 10 landed, names withheld.

Every single company scored zero in the answered column. The whole spread you see is just the difference between sites that gave an agent something to work with and sites that gave it almost nothing.

The findings that are hard to unsee

Vendor J, the worst in the batch, sells browser security. Its own website is built so an AI agent reading it through a browser comes away with almost nothing. A company whose whole pitch is "we see what happens in the browser" is close to invisible to the thing now doing the browsing. You don't need us to point out the irony.

One vendor sells compliance automation. Audit-ready reports, SOC evidence on demand, the whole story. Its own site never states which certifications the company itself holds. The agent asked the most basic trust question a security buyer asks, and the compliance vendor couldn't answer it about itself.

Another vendor raised nine figures last year. Real money, real team, presumably a real website budget. The agent got partial answers to 3 of its 10 questions and nothing on the other 7. Funding doesn't buy agent-readability. Apparently nothing on the to-do list did either.

The last one reframes everything. The single company that publishes actual prices on its site, real per-system numbers, still landed near the bottom at 15. Agent-readability is the whole site or it's nothing. One published price can't save you when the agent hits a wall on the other nine questions.

Where the agent kept hitting walls

The dead ends weren't random. They clustered on the questions buyers care about most.

• Pricing: a dead end on 8 of 10 sites.
• Security and compliance: 7 of 10.
• The competitive question: 9 of 10.
• Support and SLA terms: 8 of 10.

Integrations and data-privacy questions fared better, dead-ended on only 2 sites each, because most vendors at least keep an integrations page. But pricing and compliance are the two questions a security buyer will not move forward without. When the agent can't find them, the evaluation stalls before you're even in the running.

Why we didn't name names, and why the data is still solid

We anonymized on purpose. These companies didn't sign up to be graded in public, and we're after the pattern, not the scalps. The pattern holds whether or not you can see the logos.

What we won't hide is the method, because that's where the credibility lives.

We asked each company only its own category's questions, the ten a cybersecurity buyer actually asks. Out-of-category questions would inflate the failure rate, so we didn't ask them. The classifier was tuned to be conservative: when an answer was ambiguous, it scored the lower outcome. So if anything, these numbers are the optimistic read.

The crawler also doesn't execute JavaScript. It reads the HTML, the meta tags, and the structured data, which is roughly what a less-capable AI agent sees, and exactly what many of them still see today. A site that hides all its content behind client-side rendering scores worse here than it might inside a top-tier model with full browsing. That floor is the point. It's what your site looks like to every agent that isn't the most advanced one, and most aren't.

You can argue the scores should be a few points higher for the JavaScript-heavy sites. You cannot argue your way to a passing grade from a starting line of zero answered questions.

The part most teams still get wrong

Your website now has two kinds of visitors, and they read it nothing alike. A human skims, infers, fills gaps, gives you the benefit of the doubt. An agent does none of that. It needs the answer stated plainly, on the page, in text it can parse. Partial doesn't pass. The buyer's agent treats a half-answer the same way it treats no answer: it goes looking somewhere else, or it drops you.

This is already happening, not coming. On the B2B web, 83% of content-page traffic now comes from AI agents, not humans. That's from more than 3 million agent events Salespeak has logged across our monitored network this year. The agents go straight for the deep pages, the comparison page, the security brief, the pricing calculator, the exact pages that decide your shortlist.

And there's no single thing called AI traffic. It's four channels that behave nothing alike. ChatGPT is the volume, 84% of all agent crawls, reading everything. Claude is the buyer, it clicks pricing pages at roughly five times ChatGPT's rate, and its traffic grew 26 times in five months. Perplexity is the citer. Gemini is a phantom that sends referral traffic with almost no visible crawl. When your pricing page is the dead end, it's Claude, the agent most likely to be carrying a buyer with budget, that hits the wall.

The catch: almost none of this shows up in your analytics. AI agents fetch a page in milliseconds without ever loading the JavaScript your analytics stack waits for. So the first-pass evaluation that decides whether you make the list is happening on your site, every day, and you have close to no record it occurred.

What to do this week

Pick your three hardest buyer questions, the ones your sales team gets on every first call. Pricing, compliance, and the competitive comparison are a safe bet. Then open ChatGPT or Claude and ask it those exact questions about your company. Read what comes back. If the agent hedges, guesses, or sends you to a competitor, that's the gap, and it's the same gap that sat under all 10 of these vendors.

Every CMO is under the same board pressure right now: prove marketing still moves the number. Finding out what an AI buyer sees on your own site is one of the cheaper, faster ways to start. If you want to see exactly where your site stands, come talk to us. We do this for a living, and we'll show you what the agents already think before your buyers' agents decide for them.