--- type: Article name: What an AI agent checks before it trusts you. id: "https://selfe.ai/insights/trust-and-commerce-data/what-an-ai-agent-checks-before-it-trusts-you" url: "https://selfe.ai/insights/trust-and-commerce-data/what-an-ai-agent-checks-before-it-trusts-you" publisher: "did:web:selfe.ai" description: "Trusting a business is not one decision an AI agent makes once, it is four smaller checks made every time: where a fact comes from, what it applies to, how current it is, and what happens when one of those is missing." datePublished: "2026-09-15T05:32:19.119Z" author: The Selfe team --- # What an AI agent checks before it trusts you. > Trusting a business is not one decision an AI agent makes once, it is four smaller checks made every time: where a fact comes from, what it applies to, how current it is, and what happens when one of those is missing. ## In short - Trust breaks down into four separate checks: source, scope, freshness and what happens when one of them is missing. - Source distinguishes a fact the business controls from someone else's secondhand mention of it, even when both currently agree. - The four checks can pull against each other, passing three and failing the fourth is a real, useful outcome, not a technicality to wave through. Passing three of the four checks and failing the fourth is not a technicality an agent waves through. A claim that is sourced correctly and scoped correctly can still fail on timing alone, and when that happens, the honest outcome is a pause, not a booking confirmed on hope. ## The first check is source: where did this fact come from? A guest wants a dog-friendly cottage near the Peak District for the last weekend in October, three nights, budget around £450. Their agent is not forming an opinion about whether it likes the place, it is running a short, specific check against exactly that request, and any cottage that fails a step drops out however good the photos are. The four checks behind that judgment are not abstract categories, they are the same ones this site's own guide to safe commercial information sets out. Run against one real booking, including what happens when two of them pull in different directions, they look like this. A price quoted on the cottage's own booking calendar and the same price mentioned in passing on a review site three years ago are not equally trustworthy, even if they happen to match today. The booking calendar is the business telling the agent something it controls and can be held to. The old review is someone else's memory of a fact that may since have changed. An agent that cannot tell the two apart has no way to prefer the one worth acting on, and for this particular guest, the calendar is what holds the answer. ## The second check is scope: what does this fact cover? "Dog-friendly" on a cottage's page might mean one well-behaved dog under 20kg, or it might mean up to three dogs of any size. Both are honest uses of the same two words. This guest is travelling with one medium-sized dog, so the scoped version is what matters, not the headline claim, because a fact that is true in general but wrong for this exact request does no better than one that is simply false. ## The third check is freshness: is this still true right now? Availability for an exact three-night window can change hour to hour. A cancellation policy is far more stable and can safely be treated as current for longer. Treating both the same, either re-checking a fixed policy needlessly or trusting a live calendar that has not been touched in days, wastes effort in one direction and creates risk in the other. ## The checks do not always agree with each other, and that matters too. Say the scope check passes cleanly, the cottage takes one dog under 25kg, and the source check passes too, the claim comes straight from the owner's own page. The freshness check is where this particular booking stalls: the calendar was last updated four days ago, and a three-night window this close to the date needs confirming within hours, not days. Passing three checks and failing the fourth is not a technicality, it changes a booking an agent can complete into one it has to pause on, and the pause is the correct outcome, not a failure of the process. ## The fourth check is what happens when something is missing. Sometimes nothing resolves as cleanly as that. The dog-friendly policy might exist but not specify a size limit at all. A good outcome here names exactly what is unresolved, the site confirms dogs are welcome but not a size limit, rather than inventing a plausible-sounding number and hoping it holds. The agent then either asks the question directly or hands it to a person who can confirm it. None of these four checks are unique to hospitality, the same shape governs whether an agent trusts any commercial claim it might act on, and the site's own piece on why being found by an agent is only half of what matters covers the wider version of the argument. What is specific to a cottage booking is how often the underlying facts move, and how much a wrong one costs a guest who has already planned a weekend around it. Selfe connects exactly that: a live link to a business's actual booking and policy systems, which is what makes the four checks possible in the first place, so "dog-friendly, one dog under 25kg, confirmed this morning" is a fact an agent can act on, not a page a person has to interpret. ## Why would the same fact be trusted from one source and not another? Because the source determines who is answerable for it. A business's own live system can be held to a fact in a way a passing mention elsewhere cannot. ## What happens if an agent cannot confirm something important? A well-built agent says so and either asks a follow-up question or hands the gap to a person, rather than filling it with an assumption. ## Act on this - [Check whether an agent can buy from you](https://selfe.ai/agentic-commerce/buyability-check) — the free readiness scan. - [Discover venues](https://selfe.ai/api/registry/discover) — `POST`, semantic browse across the registry. - [Match a bookable answer](https://selfe.ai/api/registry/match) — `POST` with dates and party size. - [Verify Selfe's identity](https://selfe.ai/.well-known/did.json) — `did:web:selfe.ai`. - [Agent card](https://selfe.ai/.well-known/agent-card.json) — how to connect over A2A or MCP.