AI Has Joined the B2B Buying Committee

B2B software is bought by committees. Six to ten people, several functions, most of whom never speak to a vendor. Marketing has known this for a decade and built around it — content for the champion, security documentation for IT, an ROI calculator for finance.

What has changed is where those people now go first. A security reviewer asked to sanity-check three vendors, an ops lead told to “look into” a category, a finance partner checking whether a price is normal — these are exactly the questions people now ask an assistant, because the questions are broad, low-stakes and faster to ask than to research.

That produces a specific and underappreciated situation: your shortlist status is often decided by people who never visit your website, using a summary you did not write.

Which committee roles this actually displaces

Not all of them, and the distinction matters for where to spend effort.

The champion still visits you. Someone who will own the tool and stake credibility on it does real research — your site, your docs, a trial, a call. AI is a starting point for them, not an endpoint. Traditional content still works here.

The evaluators frequently do not. The security reviewer, the finance partner, the adjacent-team lead asked for an opinion — these people have a day job that is not this purchase. They need a defensible view in fifteen minutes. An AI answer is exactly the right tool for that, and whatever it says about you is very likely the whole of their input.

The blockers are the sharpest case. Someone looking for a reason to say no gets a synthesised answer that may include a “though some users report…” clause traceable to a single stale forum thread. That clause is now an objection in a meeting you are not in, sourced from something you could have addressed and did not.

So the displacement is uneven: the deepest researcher is least affected, and the people who most influence elimination are most affected. Which is the wrong way round from where most B2B content budget goes.

What that means for the questions worth tracking

The instinct is to track the category head term — “best CRM for enterprise” — and it is the least useful prompt in the set. It is contested, generic, and mostly reached by people at the very start of a process.

The prompts that decide B2B shortlists are narrower and much more specific:

• Fit constraints. “CRM for a 200-person sales team in the EU.” Buyers state constraints conversationally, and an assistant matches on them. This is where specialists win against larger incumbents, because the constraint disqualifies the generalists.
• Security and compliance. “Is [vendor] SOC 2 compliant?” “Does [vendor] support SSO and SCIM?” Checkable facts, asked by someone who will not dig. A wrong or absent answer here removes you silently.
• Head-to-head. “[You] vs [competitor] for [use case].” The comparison is now generated rather than read, and it is generated from third-party sources as much as from either vendor’s site.
• Objection-shaped. “Problems with [vendor],” “[vendor] alternatives,” “is [vendor] worth it.” Nobody enjoys tracking these and they are consistently the highest-value prompts in a B2B set.
• Pricing. “How much does [vendor] cost?” Especially painful if your pricing is gated, because the model will answer anyway, from whatever secondhand figures exist.

Note how many of those are branded. B2B AEO skews far more toward branded and comparative prompts than consumer categories do, because the buying process is longer and involves more people checking a name someone else supplied. Branded vs unbranded AI queries covers reading the split.

The gated-content problem

This is where B2B practice collides with the mechanism most directly, and it deserves stating plainly.

A large amount of B2B substance sits behind forms or logins: pricing, security documentation, implementation detail, integration lists, the technical depth that would actually answer an evaluator’s question. That was a defensible trade when the alternative was a prospect leaving without converting.

It is a worse trade now. Content an engine cannot fetch cannot inform an answer. If your security posture is in a gated PDF and a competitor’s is on a crawlable page, the assistant answering “which of these is SOC 2 compliant” has one source and it is not you. You have not protected the information; you have delegated the answer.

The resolution is not to ungate everything. It is to recognise that facts and assets are different things. The fact that you hold a certification, support an integration, or start at a given price point is not a lead magnet — it is a qualification signal, and hiding it disqualifies you from consideration by people who will never fill in the form. Publish the facts; gate the deep artifacts.

What an evaluator can actually verify

Worth being concrete, because it determines what survives into an answer.

Unverifiable, and largely discarded: “enterprise-grade,” “best-in-class,” “trusted by leading teams,” “seamless integration.”

Verifiable, and retained: a named certification, a published integration list, a stated data-residency region, a specific team-size fit, a numbered SLA, a real price. Several of these have structured-data equivalents worth marking up — Google’s organization structured data reference covers the identity half, and the same discipline applies to the rest: if a fact matters to a buyer, it should exist somewhere a machine can read it without a form in the way.

This is the same filter described in agentic commerce applied to a longer sales cycle. The intermediary is not persuadable and cannot see your design; it matches stated constraints against checkable claims. Most B2B homepages are written almost entirely in the first register.

What this does not mean

Two overclaims to avoid, since this argument is easy to push too far.

It does not mean AI decides B2B purchases. It influences shortlist formation and elimination. The purchase itself still runs through evaluation, procurement, security review and negotiation — none of which an assistant conducts. Anyone claiming AI visibility drives enterprise revenue is asserting a chain nobody can trace.

It does not mean you can measure the effect. An evaluator who read an answer and quietly declined to advocate for you leaves no trace anywhere in your funnel. No visit, no form, no data. That absence is the entire problem and it is also why the impact cannot be quantified honestly. What you can measure is presence — whether engines name you for the questions your buyers ask, and how you are framed when they do. Measuring GEO/AEO ROI is careful about the boundary.

The counter-argument

The strongest objection: enterprise buyers are professionals spending large sums, and the idea that a security reviewer would form a view from a chatbot rather than reading a SOC 2 report understates their diligence.

Partly right, and the correction matters. The person who owns the decision does diligence. But committees are mostly not decision-owners — they are people asked for input on something adjacent to their actual job, working to a deadline, and “I looked into it, they seem fine, though there were some complaints about onboarding” is a completely normal contribution to a vendor discussion. It always was. The change is only in where the fifteen minutes of looking-into-it now happens.

The honest version is therefore narrower than “AI buys software”: for the shallow-research members of a committee, the substrate of their opinion has moved from a mix of search results and colleague hearsay to a single synthesised paragraph. That is a smaller claim and a more defensible one.

Where to start

1. Track your branded and comparative prompts before the category ones. Include the objection-shaped ones you would rather not read.
2. Audit what an evaluator can reach. Take your five most common security and pricing questions and check whether a crawler can get to the answers at all.
3. Rewrite one page in verifiable register. Replace the adjectives with certifications, numbers, regions and named integrations.
4. Read what engines say when asked to compare you to your main competitor. That paragraph is running in meetings you are not in.

AEO for B2B SaaS covers the category, and how AI recommends products covers the selection mechanics underneath.