Shopping agents do not browse. They fetch, parse, compare, and discard. When a brand loses a recommendation, it is rarely because the product was worse, it is because the product was harder to read.
The agent's actual input
Strip a product page of its design and what remains is a title, a price, a variant matrix, a returns policy, some availability signal, and a wall of marketing prose. An agent asked to compare three options weights the fields it can trust and ignores the ones it cannot parse.
Ambiguity is not neutral, it is a penalty. A size chart rendered as an image, a shipping timeline buried in a collapsible tab, or a returns window described as 'hassle-free' all reduce to missing data, and missing data loses comparisons to a competitor who stated a number.
Three failure patterns we see repeatedly
First, variant collapse: an agent cannot tell whether a colour and size combination is genuinely purchasable, so it recommends a product where availability is unambiguous.
Second, prose-only differentiation: everything that makes the product better lives in copy written for humans, with no structured attribute an agent can compare against a rival's.
Third, policy opacity: returns, warranty, and delivery are the tiebreakers in agent-mediated purchases far more often than they are for human browsers, and they are usually the least structured content on the site.
What 'agent-ready' means concretely
It means structured product data that matches what the page says, machine-readable policies, and an MCP endpoint exposing discovery, evaluation, and transaction as explicit capabilities rather than as pages to be scraped.
It also means evaluation: a golden set of realistic shopping questions, run against your catalog on a schedule, so you can see whether an agent recommends you today and whether last week's merchandising change quietly broke that.
How we measure it
Our audit scores a sample of your catalog on parse-ability, attribute completeness, policy clarity, and head-to-head recommendation win rate against named competitors. The output is a ranked fix list where each item carries an estimated effect on that win rate.
Most brands find that a small number of structural fixes move more agent traffic than a full replatform would.
The takeaway
Agent-readiness is a data-quality problem wearing a marketing costume. Fix the structure and the recommendations follow.