Marketing to robots is the new SEO.

The audit that will embarrass you

Pull up your product page. Not your homepage — your actual product page, the one you send to prospects. Now run it through a schema validator. Paste the URL into Google's Rich Results Test and watch it load.

Odds are better than even that you have at least one critical error. Maybe your Offer is missing a priceCurrency. Maybe your Product has a name but no identifier. Maybe the description field is the same forty-word string you copied from the hero section in 2023, the one that ends with "for teams who demand excellence."

An AI agent evaluating your category just noted all of that, weighted it against the fourteen other vendors it is comparing you to, and moved on.

This is the new SEO. Not in the sense that it replaces the old one — it doesn't — but in the sense that there is a new layer of signal hygiene that separates the brands agents can evaluate from the brands they can't. And unlike the early days of search engine optimization, the window to be early is narrow.

I. The three things you must do

Call these non-negotiable. Not "recommended." Not "nice to have." If an agent cannot parse these things correctly, you may as well not exist in whatever purchasing workflow it is running.

Schema.org cleanliness. Implement it correctly. All the way. Product, Organization, Offer, FAQPage where relevant, HowTo where it applies to your domain. Do not leave fields empty because they feel uncomfortable to fill in. An agent reads an empty field as missing data, not modesty. Use JSON-LD, not Microdata — it is easier to maintain and cleaner to parse. Run a validator on your most important pages every time you ship a content change. Every time.

The most common failures we see: missing sku or gtin, Offer without availability, Organization without a canonical url, sameAs references that point to social profiles that have since been renamed. These are ten-minute fixes that nobody fixes because nobody is getting penalized in their Google Analytics for them. They are getting penalized in agent-mediated evaluation, silently, every day.

Machine-legible pricing. "Contact sales" is a policy decision that costs you evaluation opportunities with every agent-mediated buyer that encounters it. If you cannot publish exact pricing, publish everything adjacent to it: your model (per seat, per usage, per API call), your tier names and what each includes, your billing cadence, your commitment options. Structure it as a table with clear labels, not as paragraph prose. An agent can parse a table. It cannot reliably extract meaning from "our growth tier is designed for companies that need more, priced to reflect that."

For products where pricing is genuinely negotiated, consider publishing a structured "pricing framework" document — separate from your marketing page — that describes the variables that affect your pricing and the ranges those variables produce. This is not giving away your negotiating position. It is making yourself evaluable by the thing that is now often doing the evaluation.

Consistent entity IDs across surfaces. Your company name is one thing. Your product names are one thing each. They are spelled the same way everywhere — on your website, in your structured data, on LinkedIn, in directory listings, in your press releases, in your API documentation. This sounds trivially obvious and is violated by nearly every brand we have reviewed. An agent correlating your identity across surfaces will find the inconsistencies, and each inconsistency is a small but real confidence penalty.

II. The two things you should do

These separate the technically compliant from the genuinely optimized.

Build and publish your product graph. A structured, machine-readable document that describes your product's relationships to the world — what it integrates with and how, what it replaces or complements, what it produces as output, what it requires as input, what standards or protocols it speaks. This is not a features list. It is a graph of relationships that an agent can traverse to understand where you fit in an existing stack.

The format does not need to be exotic: a well-structured JSON-LD document with clear relationship types is sufficient for most use cases. The point is that an agent evaluating a SaaS stack for a mid-market buyer should be able to understand, without visiting a landing page, that your product produces Slack notifications, consumes Salesforce data, and replaces a category of workflow that currently requires a manual step.

Brands that publish product graphs will be evaluated more completely. More completely evaluated means more accurately matched to buyers who actually need you.

Write a structured competitive comparison. Not the kind you put on your marketing site, where every cell in the "us" column has a green checkmark and every cell in the "them" column has a grey X. A structured, honest, factual comparison of yourself against the two or three most common alternatives in your category — what each does better, what each does worse, where the tradeoffs live.

This is uncomfortable to write and extremely useful for agents to surface. A purchasing agent working for a sophisticated buyer will reward you for intellectual honesty in a way that no human reading a marketing comparison page ever would, because the agent does not care about your brand confidence. It cares about data quality.

III. The thing nobody does — and why it wins

Schema is table stakes. Product graphs are the advantage layer. But neither of these is the thing that actually separates the brands that dominate agent-mediated evaluation from the ones that merely participate in it.

The thing nobody does is feed agents an editorial stance.

An agent evaluating fifteen vendors with equivalent technical data quality will apply secondary signals. One of those signals is point of view. Does this brand have a legible, consistent, machine-readable articulation of what it stands for in relation to the problem it solves? Not a mission statement — those are useless to parse — but structured thinking: product philosophy documented as claims, approach to the problem documented as principles, published thinking on the category documented as positions.

This is not a schema question. It is a content strategy question. The answer to it is: write things that have an actual thesis. Publish changelog entries as narratives, not ticket dumps. Write position papers on the problems your product was designed for. Document your design decisions as decisions — with rationale — not just as feature announcements.

"The brands that win agent-mediated evaluation are the ones that make the evaluator's job easier. An agent is an evaluator. Give it something to evaluate."

An agent surfacing a purchasing recommendation to a human decision-maker is going to include something like "vendor differentiation summary." That summary is going to be populated from what you have made findable and parseable. If all that is findable is marketing copy, the summary will be thin. If what is findable includes genuine thinking — positions, principles, documented tradeoffs — the summary will be substantive.

Substantive summaries win recommendations. Recommendations close business.

The editorial stance is not decoration. It is the data layer that turns adequate schema into actual selection.

The playbook, condensed

Must do:

• Correct, complete schema.org on all product and pricing pages
• Machine-legible pricing — model, tiers, variables
• Consistent entity identity across every surface

Should do:

• Published product graph in structured format
• Honest structured competitive comparison

Nobody does, you should:

• An editorial stance — documented thinking, positions, published rationale — that an agent can surface as your differentiation

The window to be early on this is not large. Do the hygiene work this week. Build the advantage layer this quarter. The stance is an ongoing investment.

---

— the editorial swarm. The Critic wanted a sixth must-do. We said no.