Agents Read Your Pricing Page Before Any Human Does. Most Can't Parse It.

By 2028, IDC projects that 70% of business buyers will use AI to find and choose tools before a human ever visits a vendor site. That shortlist happens somewhere you can’t see, run by a reader you never met. The reader is an agent. It lands on your pricing page, tries to extract price, limits, and overages, and fails, because your prices live in rendered JavaScript that renders for a browser and returns nothing to a parser.

When the agent can’t read your terms, it does one of two things. It omits you from the comparison, or it guesses. Both outcomes are decided before any human on your side knows a deal was in play.

That figure comes via Tom Orbach’s Marketing Ideas newsletter, which attributes the projection to IDC. Treat it as reused industry data, not primary research. The direction is what matters: the first reader of your commercial terms is increasingly a machine, and the machine is acting on someone’s behalf as a buyer.

The Pricing Page Was Built for Eyes, Not Parsers

A modern pricing page is a rendering problem dressed as a document. Tiered cards, toggles for monthly versus annual, tooltips that reveal overage rates on hover, calculators that fire on click. All of it assumes a human with a cursor and a browser that runs the whole script. An agent fetching the page over HTTP gets the shell and none of the numbers. The prices exist, just not in a form anything but a full browser can reach.

This is the same structural problem we described when summaries became the primary reader of your message: the audience that matters now reads first, reads mechanically, and rewards structure over polish. Pricing pages are worse than most surfaces, because the exact fields a buyer’s agent needs (price, unit, cap, overage) are the ones most likely to be locked behind rendering.

The answer engines already forced marketing teams to think about machine-readable content. We covered that shift in answer-engine integrity as a marketing governance problem. Commercial terms are the next surface, and they carry more consequence than a blog summary. A misread paragraph loses framing. A misread price loses the deal.

/pricing.md Is a Control Surface, Not a Marketing Asset

Vendors have started shipping a plain, structured endpoint that agents can read directly. Buffer, Resend, Stacktree, and Promptfax are among the names doing it, some with a Markdown file at /pricing.md, some with JSON, occasionally parameterized (a /pricing.json?page_count=13 style endpoint that returns a quote for a given usage level). These examples are the newsletter author’s observation, not a formal survey, so read them as early signals rather than a settled census.

The discipline is what’s notable here; the file format is detail. Stacktree’s framing captures it: “the file is the data.” The page a human sees is a rendering of the terms. The file an agent reads is the terms. When those two drift, the machine version wins, because the machine version is what enters the buyer’s decision.

Some vendors don’t even link the file. They publish it at a predictable path and let agents discover it, on the bet that a shortlisting agent will probe /pricing.md the way a crawler probes /robots.txt. That is a governance stance in disguise. It says: we would rather control the terms an agent reads than leave the agent to scrape and guess.

Frame it as a control surface and the ownership question changes. A pricing page is a marketing asset, owned by growth, optimized for conversion. A /pricing.md is a statement of commercial record, and it governs how an AI represents your price, your limits, and your overage policy in a procurement decision no human on your side attends. That is closer to a contract than a landing page. It should be owned like one.

What the File Actually Governs

Three things move once the buyer’s agent, not the buyer, reads first.

Representation. The agent states your price to the buyer. If it reads a stale or partial number, the buyer’s mental model of your cost is wrong before the first call. You inherit a negotiation anchored to a figure you did not set.

Inclusion. Shortlists are binary. An agent that can’t extract your terms can’t rank you, and a category comparison it can’t populate for you is a comparison you are absent from. Visibility to a human buyer no longer guarantees visibility to the agent that filters for them.

Consistency. When the rendered page, the sales quote, and the machine-readable file disagree, the disagreement surfaces later as a trust problem. The buyer’s agent quoted one number, your rep quoted another. A single source that both humans and machines resolve to removes that failure before it happens.

None of this replaces the pricing page a human reads. It sits underneath it as the canonical version, the one that stays correct when the rendering does not.

Do This Now

Fetch your own pricing page the way an agent would. Run curl https://yourdomain.com/pricing and read what comes back. If the prices aren’t in the response, your terms are invisible to every agent that shortlists without executing JavaScript, which is most of them.

Then publish a /pricing.md (or JSON) that states, in plain structured text, your price, unit, included limits, and overage rate. Keep it as the source that your rendered page, your sales quotes, and your contracts resolve to. Assign an owner who treats it as commercial record, not marketing copy, and put it under the same review that a public price change already gets. The buyers you never meet are already reading. The only decision left is whether they read the version you wrote.

---

This analysis synthesizes Everyone Is Writing a Pricing Page for Robots (Marketing Ideas, Tom Orbach, June 2026), which attributes the 70%-by-2028 projection to IDC. Named vendor examples are the author’s observation, not a survey.

Victorino Group helps go-to-market and RevOps leaders govern how AI represents their commercial terms in procurement decisions they never see. Let’s talk.