Every B2B pricing page was written for a human. The tiers are simplified because people get overwhelmed. The enterprise column says "contact sales" because the real number is a negotiation. The .99 is there because human brains round down.
A second reader is showing up, and none of that works on it.
The plumbing is already live
This is further along than most GTM teams realize. Ramp issues agents tokenized cards scoped to a single transaction, with human-set spending limits and approval workflows. Mastercard and Google are building an open standard to verify that an agent's purchase was actually authorized. Stripe is six months into an agentic commerce protocol with OpenAI. Vercel sells credits, add-ons, and domains through a CLI command.
That is billing infrastructure for machine buyers, shipping today. The buying side is arriving faster. A meaningful share of B2B evaluation already runs through AI assistants that read your docs, compare your pricing, and shortlist you before a human ever visits the site.
What the machine reader does with your page
A human tolerates opaque tiers. The machine reader runs a structured evaluation. It wants the full pricing detail, room to model scope and usage, and a budget ceiling to reason against. It has endless patience for fine print. Charm pricing and the deal effect bounce off it completely.
Gated pricing fails quietly with this reader. Best case, it guesses your price from stale third-party threads. Worst case, "contact sales" reads as a dead end and you drop off a shortlist you never knew existed. No form fill, no CRM record, no trace. The same way a silent failure in an agent pipeline reports "no results" when the truth is "the lookup failed," a gated pricing page reports nothing at all while quietly costing you the evaluation.
What ten newsletters taught me about the second reader
Every issue I ship carries question-format H2s and an FAQ block. It started as an experiment in writing for the retrieval layer, since answer engines lift structured Q&A almost verbatim, and the format decides whether my thinking shows up when someone asks an AI about eval frameworks or cost telemetry.
The surprise was not that it works. It was what the format revealed. The machine reader rewards structure and punishes cleverness, which is roughly the opposite of what most marketing pages optimize for. Ten issues in, that trade stopped feeling like a constraint and started feeling like the brief.
This blog runs the same experiment on my own domain.
The prep work that matters
Four moves cover most of it.
Run the test. Ask Claude or ChatGPT what your product costs and how it compares to your top competitor. That answer is your pricing page as the second reader currently sees it.
Publish bounds. Full transparency can wait. "Starts at $10K per year" gives the reader a number to reason within, which beats being guessed at.
Ship a machine-readable mirror. Tables, precise product language, a markdown version of the pricing page. Vercel maintains one. Resend reports it makes agents measurably less confused.
Audit the third-party layer. Most of what AI search says about a brand comes from sources the brand does not control. One analysis from AirOps puts third-party share at 85% of brand mentions. If a review site describes your pricing wrong, the second reader inherits the error.
Where this lands
Autonomous B2B purchasing is still early. Today the machine reader acts like an influencer on the buying committee, deciding which products make the evaluation and how they compare. Zero-click buying will start with dev tools and commodity spend, then move upmarket.
The buyers are changing faster than the pages they read.
FAQ
Do AI agents actually buy B2B software today? Rarely on their own. They already shape shortlists, compare pricing, and brief human buyers. The purchasing rails, including Ramp's agent cards, Stripe's agentic commerce protocol, and the Mastercard and Google verification standard, are being built now.
Should we publish exact pricing? Bounds are enough to stay in the evaluation. A starting price, a pricing model, or a cost estimator gives the machine reader something to reason with.
What is the fastest first step? Ask ChatGPT and Claude how much your product costs. Whatever they answer is what every agent-assisted buyer is being told today.
How do you make a page readable for AI agents? Structure over prose. Question-format headings, tables, an FAQ block, precise product language, and a markdown mirror of the pricing page. Retrieval systems lift structured Q&A far more reliably than marketing copy.
Does this replace the sales team? No. It changes which deals ever reach them. The machine reader filters the consideration set before a human books the call.