---
type: WebPage
name: Knowledge is sales.
id: "https://selfe.ai/insights"
url: "https://selfe.ai/insights"
publisher: "did:web:selfe.ai"
description: "What AI agents are actually doing, what the standards are doing, and what either means for somebody running a business."
---

# Knowledge is sales.

> What AI agents are actually doing, what the standards are doing, and what either means for somebody running a business.

What AI agents are actually doing, what the standards are doing, and what either means for somebody running a business rather than building one. Written when something happens, not to a schedule.

Selfe connects what you sell to the AI agents people increasingly use to choose and buy.

**The request behind your booking button is what matters.** A booking button helps a person start a journey. An agent needs to establish the terms of a specific request, know what it may do and carry the relevant context into the next step.

**Someone asks for a dog-friendly stay in October. What happens next?** A dog-friendly stay in October sounds like a simple filter until an AI agent tries to act on it. What happens next depends on details a first-pass search never asks for.

**How much does an agent need to know before it can buy?** Not everything. An AI agent working through a booking request doesn't need a business's full history, it needs the small number of facts that resolve the specific thing the person asked for. A page with forty things to say and none of them the right one loses to a page with three.

**What "a table for four at eight" is actually asking for.** A customer request contains the variables that determine whether an offer is possible. Treating it as a loose search loses the information needed to give a trustworthy commercial answer.

**Five questions an agent checks before it'll recommend a product.** An AI agent shopping on someone's behalf doesn't browse a product page the way a person does. Before it will recommend anything, it needs five separate questions answered: does the exact variant exist, is it currently in stock, will it work for this specific case, what happens if it's wrong, and will it arrive in time. A page built to persuade a browsing human answers few of those directly.

**The agent does not have a favourite.** A shopper who has bought from the same brand for years reaches for it out of habit. An AI agent checking the same purchase has no habit to reach for, so a small business competing against a household name starts the comparison exactly level.

**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.

**The route from a true fact to a completed purchase.** A product page showing live stock is not the same as a product an AI agent can buy. The gap between the two is usually a missing action, not a missing fact.

**What happens when Google moves closer to the transaction?** Google increasingly lets someone research, compare and buy without ever leaving the search results page. An AI agent doing the same thing on someone's behalf needs the same thing Google's own tools do: real, current facts it can act on directly.

**Three questions every business should ask about its AI readiness.** A useful readiness review asks whether an AI agent can find the business, establish what is true and make permitted commercial progress. It should end with the next fix, not a vague score.

**Agent-ready means the business sets the boundary.** Being ready for an AI agent to complete a purchase and letting every purchase complete without a person ever looking are two different decisions, and a business can choose the first without the second.

**Deposits, holds and cancellations: what an agent needs to know.** A hold, a deposit and a cancellation policy sound like one topic to a person booking a stay. An AI agent needs them treated as three separate, precisely stated facts, or it can misrepresent what's been committed.

**The three ways a transaction can still fail after the agent says yes.** An AI agent confirming a purchase isn't the end of the transaction. Three specific failure points sit between "yes" and the order completing, and each needs a different fix.

**Transactions for multi-unit operators: the same rules, every site.** A retailer running several shops shouldn't have to solve agent-ready transactions once per site. The rules an AI agent checks are one business decision, made once, not rewritten at every location.

**What happens when an agent's request doesn't match your rules.** A request that falls outside a business's rules doesn't have to end the transaction, as long as there's a real, bookable alternative behind it that an AI agent can complete.

**What has to be true before an agent presses confirm.** An AI agent completing a booking on someone's behalf needs more certainty than a person browsing does, because nobody is there to notice if something's wrong before it's confirmed.

**Agentic commerce for portfolios: one setup, every property.** A group running six cottages doesn't need to solve this problem six times. Getting one property right the hard way, and then repeating that work five more times, is usually the wrong approach entirely.

**The difference between a mention and a sale.** Most of what decides whether a recommendation becomes a booking happens after the mention, not during it, in the specific checks that either confirm a real booking or quietly let it fail.

**What 'agentic commerce' means, in one paragraph.** The phrase gets used loosely enough that it's worth pinning down properly, once, rather than half-defining it on every other page that uses it.

**Why price alone doesn't win an agent's recommendation.** An AI comparing two nearly-identical products defaults to the one it can be most confident about. Price is only one part of that confidence, not the whole of it.

**Your customer's agent, not your customer: who's doing the shopping now.** A growing share of shopping isn't a person browsing a site at all. It's a standing AI agent, already trusted with the job, checking and buying on someone's behalf without them watching.

**How long it actually takes to go from unready to bookable.** The honest answer to "how long will this take" depends far less on how hard the technical work is than on how quickly a business can find the one fix an AI agent is genuinely waiting on.

**Readiness for portfolios: one score, many properties.** Three shops running the same till system don't have three separate readiness problems, if one has a stock sync gap, they very likely all do. The real work is telling that shared gap apart from what's genuinely different at each site.

**The one fix that moves the readiness needle fastest.** Most businesses try to fix everything a readiness check flags at once. The faster route is finding the single fix that's genuinely blocking an AI agent, and doing that one first.

**The readiness score, explained: what each number actually measures.** Two shops can land on the exact same readiness score for completely different reasons, one undiscoverable, the other unable to let an AI agent act once it's found. Knowing which of three separate things is weak is worth more than the number itself.

**What a failed readiness check actually looks like.** An AI agent quietly giving up or guessing is what a failed readiness check usually looks like, harder to notice than an obvious error because nothing about it looks broken.

**Why "we're on Google" isn't the same as being agent-ready.** Ranking well in a search engine and being ready for an AI agent to book you are answering two different questions, and a business can pass one completely while failing the other.

**Multi-brand retailers: keeping every label distinct when an agent is comparing.** A retailer selling forty brands under one site template often ends up saying the same thing about all of them. An AI agent comparing two products from two different brands needs the fact that's true of each one, not a house style that's quietly smoothed the real differences away.

**Returns and sizing: the two questions agents ask that catalogues rarely answer.** Sizing and returns are the two questions that decide whether an AI agent will risk a recommendation at all, and most catalogues answer neither with any real specificity. "True to size" and "easy returns" aren't answers, they're the absence of one.

**Variants, stock and delivery: what an agent checks before it recommends.** Three facts gate whether an AI agent will recommend a product at all, regardless of anything else about it: does the specific variant exist, is it there right now, and will it arrive in time. Get any one wrong and the rest of the listing doesn't matter.

**What a product feed needs that a product page doesn't.** A product page is written for a person looking at one item. A product feed is what an AI agent reads to compare many at once, and it needs a different kind of precision: structured, decision-relevant fields, not the same copy reflowed into a spreadsheet.

**What makes one listing easier for an agent to sell than another.** Given two nearly identical products, an AI agent recommends the one with fewer unresolved questions, not the one with better marketing copy. The gap between them is almost always one missing, narrow fact, not a missing feature.

**What "buyable" needs beyond stock.** An AI agent has already checked that a product is in stock. Buyable asks more of it than that: is the price the one that'll actually be charged, and can this specific item reach this specific buyer at all. Miss either and "in stock" stops meaning anything useful.

**The next twelve months in agentic commerce.** Most writing on where AI agents are heading falls into one of two camps: everything changes overnight, or nothing really changes at all. Neither is honest. Some things about agentic commerce are already settled enough to build for. Others genuinely aren't decided yet, and pretending otherwise doesn't help anyone plan.

**The businesses that will be ready before anyone asks them to be.** Being ready for AI agents isn't a deadline to hit when the volume arrives. An AI agent already checks the same facts today that it will at scale next year, just less often. The businesses ahead of the curve are doing the unglamorous work now, before anyone's asking them to.

**The standards still being written, and why that's not a reason to wait.** The specifications an AI agent uses to connect to a business, prove who it is, and get authorised to spend are all still being actively revised by the organisations building them. That's genuinely unsettled, not a reason to wait. The work that needs doing doesn't depend on any of it finishing first.

**The research an agent does before you ever open a search box.** A growing share of product research now happens inside an ongoing conversation with an AI agent someone already trusts, rather than a fresh search each time. Nobody has a precise figure for how much yet, and any business that says otherwise is guessing. The direction, though, is real enough to plan around.

**What changes for discovery in the next twelve months.** Not a sweeping prediction. Over the next year, the practical change for discovery is narrower and more useful than "everything will be different": more of the requests reaching a business will already carry real constraints attached, rather than arriving as a vague, browsable query.

**What changes for transactions once agents hold payment credentials.** Once an AI agent can hold a payment mandate and complete a purchase on someone's behalf, it's tempting to assume checkout itself transforms. For most businesses, what changes is narrower than that, and what doesn't change is worth stating plainly.

**What happens to reviews once agents are the ones reading them.** A star rating tells a person something useful at a glance. It tells an AI agent almost nothing it can act on. As more of a property's reviews get read by agents rather than skimmed by people, what makes a review useful is shifting, and a 4.6 average isn't going to be the part that matters.

**Being found vs being bought from: what's the difference?** Being found means an AI agent knows a business exists and mentions it. Being bought from means the agent can complete a booking or purchase there. The first doesn't guarantee the second.

**What is agentic commerce?** Commerce where an AI agent, acting on a real person's goal, does some or all of the finding, deciding and buying that a person used to do by hand.

**What counts as a verified fact?** A fact an AI agent can trust because it traces back to a real source, a business's own booking system, a supplier feed, a confirmed document, rather than to unverified marketing copy or a crawled page.

**What does "agent-ready" mean?** A business is agent-ready when an AI agent can find its real facts, trust them, and complete a booking or purchase, without a person having to fill in the gaps by hand.

**What is a buyability check?** A quick, honest read on whether an AI agent could complete a booking or purchase with a business today, checked against its real, current facts.

**What is a Curator?** The part of Selfe that checks a business's facts against real evidence before they're trusted, weighing a claim differently depending on whether it comes from a verified system or an unverified source.

**What is a readiness score?** A single number summarising how ready a business's facts are for an AI agent to find, trust and act on, measured against the same checks an agent would run.

**What is a Wallet, in the Selfe sense?** The record of what a specific AI agent is currently authorised to spend, on whose behalf and up to what limit, checked before Folio settles anything against it.

**What is A2A?** Agent2Agent protocol, a standard for how one AI agent hands a task to another agent clearly enough that the second one can trust it and act on it.

**What is ACP?** Agentic Commerce Protocol, an open checkout standard, associated with OpenAI and Stripe, for an AI agent to complete a purchase directly inside a conversation rather than sending the shopper off to a separate website.

**What is "agentic"?** A description of behaviour, not a fixed category of product: something is acting agentically in a given moment when it's pursuing a goal on its own initiative, rather than only responding to what it's just been asked.

**What is an agent credential?** A short-lived, scoped proof that an AI agent is who it claims to be, checked before a business trusts a request enough to act on it.

**What is an AI agent?** An AI system that's given a goal and works toward it on a person's behalf, rather than one that just answers whatever's typed into it, turn by turn.

**What is an Echo?** Selfe's structured representation of a business, the specific, verified facts an AI agent checks, kept separate from the website copy written for a browsing person.

**What is AP2?** Agent Payments Protocol, a standard for proving that a real person authorised an AI agent to spend a specific amount, so a business can verify that before it charges anything.

**What is ARD?** Agentic Resource Discovery, Selfe's own standard for how an AI agent finds a business's Echo in the first place, complementary to UCP rather than competing with it.

**What is Folio?** Selfe's payment layer, the part that confirms an AI agent is authorised to spend on someone's behalf and settles the transaction with the business once it does.

**What is Bootcamp?** Selfe's eight-week, hands-on programme: Selfe's own team working directly with a business to get its whole Selfe properly configured, on the business's existing systems.

**What is MCP?** Model Context Protocol, an open standard that lets an AI agent call into a business's own systems, checking live data or triggering a defined action, the way it would check availability or pull a menu.

**What is structured data, for AI?** A business's facts held as distinct, checkable fields, availability, price, terms, rather than as prose an AI agent has to read and interpret to extract the same information.

**What is UCP?** Universal Commerce Protocol, the checkout protocol that completes a purchase once an AI agent has already been connected, identified and authorised, handing a business back a confirmed order instead of an unverified promise.

**What is WebMCP?** An emerging, still-early standard that would let an autonomous AI agent discover and call actions directly on a website it visits on its own, without a business or developer configuring a connection ahead of time.

**Activities and tours: what "suits who" actually means to an agent.** An activity that's technically available isn't automatically suitable, and an AI agent checking a booking for a family with young children needs to know the difference before it recommends anything.

**Cottages, hotels, and the difference agents don't care about.** An AI agent comparing a self-catering cottage against a small hotel isn't weighing the category, it's weighing the same handful of underlying facts wherever they happen to sit.

**Restaurants, pubs and the booking that isn't just a table.** A table booking for twelve people celebrating a 40th birthday is not the same transaction as a table for two, and treating it that way is where the booking usually goes wrong.

**Theatres and attractions: showtimes, casts and who a show is for.** A theatre's listing for a touring show can be accurate on Monday and wrong by Thursday, and an AI agent booking a specific performance needs to know which facts are stable and which change performance to performance.

**What a room rate needs to say that a listing page doesn't.** A room rate on its own is just a number, and an AI agent needs to know what that number includes before it can recommend it with confidence.

**How to check what AI is already saying about you.** Being found by an AI agent is the easy half of the job. Here's how to check where you stand today, and why the check itself only ever tells you half the story.

**Structured data, explained without the jargon.** A returns policy that says "just get in touch if there's a problem" and one that says "returns accepted within 30 days, unworn, tags attached" can describe the same shop. Only one of them can be checked by something that isn't a person using judgement.

**What changes the moment you go from mentioned to shortlisted.** Being mentioned by an AI agent and being one of the two or three options it puts in front of someone are different achievements, and the second one is decided by a narrower set of facts than the first.

**What it means for an AI to know a place, not just describe it.** A product description tells an AI agent what a shop says about itself. Being known is a different, higher bar: whether an AI can trust what it's told enough to act on it.

**What it means to be found by an agent that already knows the person asking.** A standing AI agent that already knows its person skips the broad question a search engine expects entirely. It arrives already narrowed, checking a shop's facts against one specific person's specific needs.

**Why two businesses with the same facts get different recommendations.** Two shops can publish exactly the same facts about a similar product and still get recommended unevenly. The difference is rarely the facts. It's how confidently an AI agent can act on them.

**One setup, every surface: what that actually requires.** A canal boat hire company doesn't need a separate connection for every AI agent that asks about its boats. What keeps one live setup actually one setup, rather than three quietly disagreeing copies, is a daily discipline most businesses never name.

**Personal agents and what changes when your customer's agent already knows them.** A ceramics studio investing in being found on every surface a shopper might browse has nothing to offer a customer whose personal AI agent already knows exactly what they want and isn't browsing anywhere at all.

**The surfaces where agents choose and buy.** Customers increasingly begin in places that help them research, compare and progress a purchase. The durable task is to make a business understandable and actionable across those surfaces without rebuilding everything for each one.

**The other kind of agent: when there's no chat window at all.** A walking-tour company checking its visibility in ChatGPT and Google is checking the wrong thing entirely for a growing share of its potential bookings, the ones arriving through an AI agent nobody ever sees typing a question.

**What a standard actually buys you over a one-off integration.** A one-off integration solves the surface one AI agent is asking from today. An open standard solves the shape of the question, which is what every future surface will keep asking in some form.

**What a standing agent checks that a one-off chat session never does.** A one-off chat session checks a business's facts once and moves on. A standing AI agent checks the same facts again next week, and the week after, which changes what "current" needs to mean.

**What changes when ChatGPT can check out directly.** Being mentioned favourably by an AI agent used to be the whole test. Once that same agent can complete a purchase without leaving the conversation, it needs a live, checkable answer, not just a good description.

**When a direct platform relationship is still worth building.** A distillery that keeps its real facts in an open standard is free to build a genuine relationship with a specific AI platform without repeating its old mistake, because the two are doing different jobs: one keeps the facts readable everywhere, the other earns visibility somewhere specific.

**"Add an llms.txt file and you're SEO-optimised for AI." There's no ranking signal involved. What it's for is narrower and more specific.** An llms.txt file isn't a ranking mechanism, and treating it like a new kind of SEO misreads what it's for. What it can genuinely help with is narrower and more useful: telling an AI agent what a business can do, not making its content rank better anywhere.

**"AI agents just means ChatGPT." It's a much wider, more personal category than that.** Most people picture one thing when they hear "AI agent": a chat window they open and type into, the way they'd open a search tab. That's one shape among many, and probably not the shape that ends up mattering most for buying and booking.

**"AI agents will replace your website." No, and here's what actually happens to it.** An AI agent shopping on someone's behalf doesn't need a website the way a person does. That's not the same as the website disappearing. Its job changes, from the primary place a sale happens to one surface among several reading from the same underlying facts.

**"AI can just read the page, you don't need structured data." It's solving the wrong layer of the problem.** The usual fix people reach for is extending the markup already on a page, more tags, richer fields, describing more of the business in the same place. That's still a page-shaped answer. The facts an AI agent needs don't have to live on a page at all, marked up or not.

**"Being mentioned by ChatGPT means you're winning." It doesn't, and treating it that way is costing you bookings.** In the search-results era, a mention at least came with a link, something a person could click and act on. A lot of today's chat mentions carry neither the link nor any path to book. That's not just as weak as old-style visibility. It's often hollower.

**"Bigger platforms will sort this out for us." Why we don't think that's coming, or coming for everyone.** The search era consolidated around a handful of gatekeepers, so it's a reasonable guess that agentic commerce will too, and that waiting for whichever one wins is the safe move. Genuinely autonomous, personal AI agents don't obviously converge that way, and waiting assumes a consolidation that might just not happen this time.

**"Just add a chatbot" was never a strategy. Here's the argument for why.** A chat widget is a human-interface metaphor, a text box someone types into, bolted onto a business that hasn't changed underneath. It answers the question "how does a person talk to us" when the real question is whether the facts behind the widget are the ones an AI agent can act on.

**A lot of what you've heard about agentic commerce isn't quite right.** Every myth in this hub comes from the same mistake: judging the future by how well it fits today's tools, a page, a widget, a chat window, a markup file, instead of asking what would make sense if you were starting clean. The actual destination underneath all of it is simpler than any of the myths: AI agents talking directly to data, not reading pages built for people.

**"This is years away." Say a competitor already connected its facts. It's checkable now, not years from now.** This isn't a gradual interface upgrade a business adopts on its own schedule, the way a redesigned website rolls out over a quarter. It's a distribution shift that happens the moment an AI agent can check a business's facts and finds them connected, which is exactly why it can look sudden to one that hasn't done that work and unremarkable to one that has.

**"WebMCP is the real answer for autonomous agents." Maybe eventually. Worth asking why it's needed at all.** WebMCP is genuinely early, worth flagging as unsettled rather than treated as decided either way. Even taken on its own terms, it's answering a narrower question than it sounds: how an AI agent copes with a web built for people, not what agentic commerce looks like if nobody had to cope with that at all.

**"We've set up MCP, so we're agent-ready." That covers your own team's tools. It's a different job from the one a stranger's agent needs done.** MCP is genuinely useful, just not for the case most businesses think it covers. It's strong for an operator's own tools, a chat assistant working against your booking system, less so for a stranger's AI agent trying to book or buy from you. Both are real, but only one of them is what "agent-ready" usually means.

**Your local knowledge is a sales asset.** The best local knowledge lives entirely in conversation, asked for and answered in person. That's exactly the knowledge an AI agent can't reach, however good the website is.

**The things an agent cannot guess.** Getting found and getting matched are two different jobs. A business can nail the first and still lose the second, because an AI agent only knows what it's been told, never what it could have worked out by being there.

**What 'agent-ready' means.** "Agent-ready" gets used to mean almost anything. Here's what it requires, in one chain, and where most businesses are further along than they think, and further behind than they hope.

**Can an agent buy from you?** Being chat-visible and being buyable are different tests. Most shops pass the first without ever finding out whether they'd pass the second.

**Being found is not the same as being bought from.** This is the distinction the rest of this hub, and most of this site, keeps coming back to. Being found by an AI agent is real progress. It is not, on its own, a booking, a sale, or a reason to stop.

**The first thing an agent gets wrong about your business.** An AI agent rarely gets a business wrong out of carelessness. It gets it wrong because the fact that would have changed the answer was never written down anywhere it could check.

**Nobody asked for you by name. You got the sale anyway.** The strongest position in agentic commerce isn't brand recognition. It's being genuinely, checkably right for the request, even when nobody's ever heard of you.

**A2A, explained: how one agent proves what it's asking another to do.** A shopping AI agent rarely finishes the whole job itself. It hands specific tasks to other, more specialised agents along the way. Agent2Agent protocol is how one agent describes a task to another clearly enough that the second agent can trust it and act on it.

**Agent credentials: the token that says "yes, actually let this agent buy."** Before an AI agent can act on anyone's behalf, something has to vouch for it. An agent credential is the short-lived, scoped proof a business checks before it trusts a request enough to act on it.

**AP2 and agent payments: how an agent proves it's allowed to spend on someone's behalf.** A credential proves an AI agent is who it says it is. AP2 (Agent Payments Protocol) proves something narrower and more specific: that a real person authorised this exact spend, up to this exact limit, and a business can verify that before it charges anything.

**Holds, mandates and spend limits: how a business keeps control of an automated payment.** A mandate only says what an AI agent was allowed to try. What protects a business is what happens on its own side of the transaction: the hold it places, the ceiling it enforces, and the mandate scope it insists on before a booking or order is allowed to proceed.

**MCP, explained: what it actually connects, and what it doesn't.** Model Context Protocol gives an AI agent a standard way to call into a business's own systems: read live data, run a search, trigger a defined action. It has nothing to do with payment or identity. That's a separate job, done by separate protocols.

**The plumbing behind every agent purchase.** Every purchase an AI agent completes, whatever it's buying, passes through the same four jobs: proving who the agent is, proving what it's allowed to spend, connecting to the business's own systems, and completing a transaction the business can settle. The protocols in this hub each do one of those jobs. None of them does all four.

**UCP and the checkout handoff: what changes when a purchase completes inside the conversation.** Every other protocol in this hub gets a purchase ready. UCP is the one that completes it: the point where a business hands back a confirmed order without sending the customer off to finish the job on a website.

**What a checkout API call looks like from the other side.** Strip away the protocol names and an AI agent completing a purchase is, from a business's own system, just another API request arriving with a specific, checkable shape. Here's roughly what lands, and what a business's system has to do with it.

**What makes information safe enough for an agent to act on?** An agent should not turn a confident sentence into a commercial promise. Information becomes actionable when its source, scope, timing and next permitted use are clear.

**Trust for DMOs: what it means to vouch for someone else's business.** A destination marketing organisation is not the source of any single business's facts, it is a curator standing between hundreds of them and an AI agent, and that middle position changes what "trust" has to mean.

**What happens when your data disagrees with itself.** A product can be in stock on the website, sold out on the warehouse system, and back-ordered on the marketplace listing, all at the same moment, and an AI agent checking any one of the three has no way to know it is only getting a third of the picture.

**What "verified" actually means, source by source.** "Verified" gets printed on all kinds of listings as if it were one fixed guarantee, but it can mean three very different things depending on where the fact came from, and an AI agent needs to know which kind it is looking at.

**Why a five-star review isn't evidence an agent can use.** A five-star rating tells an AI agent that other people were happy, which is a real signal, but it is not the kind of fact an agent can act on the way it can act on a stated return window or a confirmed size.

## 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.