How AEO Butler defines, measures, and proves data accuracy to AI agents — and why it is the hardest part of agent-ready infrastructure to get right.
An AI agent operating on behalf of a buyer does not guess. It reads what a site declares — prices, availability, policies, product identifiers — and acts on those declarations. If what the site declares is wrong, the agent acts on wrong information. The buyer completes a transaction at the wrong price, or for a product that is out of stock, or under policy terms that do not apply.
Veracity is the property that makes a site's declarations trustworthy. It is not a one-time certification. A site that passes an audit today can fail tomorrow if a price changes in the visible HTML but not in the JSON-LD, or if an availability claim goes stale.
Veracity is not what you claim. It is the continuous, provable consistency between what you claim and what is true.
AEO Butler measures veracity across four data categories. Each has a different rate of change and a different consequence when it fails.
The price visible on the page must match the price declared in JSON-LD. A mismatch of more than five percent triggers a degraded alert. Price data changes frequently and is monitored on a four-hour cycle.
Availability is the highest-weighted dimension because inventory errors cause agent task abandonment more frequently than any other data failure. An agent that reads InStock in structured data but encounters an out-of-stock error at checkout has been misled. Monitored every two hours.
Return policies, shipping terms, and warranty information must be present in machine-readable text. Policy data changes infrequently but matters significantly in high-value transactions. Monitored every twenty-four hours.
Interactive elements must maintain their accessible names and correct element types across all rendering states. An Add to Cart button that loses its accessible name when a product variant is selected is an action veracity failure. Monitored every four hours.
Each dimension produces a score between 0.0 and 0.99. The upper bound is 0.99 rather than 1.0 because no finite sequence of successful monitoring cycles constitutes certainty. The score updates using an exponentially weighted moving average after each monitoring cycle.
The composite veracity score uses a weighted harmonic mean across all four dimensions. The harmonic mean penalises low scores in any single category more strongly than an arithmetic mean would — a site with perfect price veracity but near-revocation availability veracity does not receive a high composite score.
Every site enrolled in continuous monitoring publishes a veracity manifest at /.well-known/veracity.json. This machine-readable JSON-LD document contains current veracity assertions, scores, methodology reference, and last audit timestamp. It is the file an AI agent reads to determine whether a site's data can be trusted before initiating a transaction.
The manifest for this site is at /.well-known/veracity.json. It is a static file served from the site's own infrastructure. Any agent or auditor can verify it independently without contacting AEO Butler.
Between full audits, the monitoring pipeline runs continuous checks comparing the current state of a site's structured data against its certified baseline. Changes are classified into three alert tiers.
A change that blocks agent task completion entirely. An interactive element losing its accessible name. A checkout action becoming unreachable. A price mismatch exceeding five percent on a transactable product.
A change that impairs agent navigation or reduces trust without blocking task completion entirely. A price change within tolerance. A delivery window estimate changing significantly. A moderate availability inconsistency.
A change that diverges from the certified baseline but does not impair agent navigation. A product description update. A minor structural change. Drift alerts are queued for the next full audit cycle.
This page holds a veracity score of 0.99 across all four dimensions. The full veracity manifest is at /.well-known/veracity.json.