PL / EN Book an audit

AI readiness Your customer has stopped searching. Soon an AI agent will decide who gets their money.

Today people ask AI whom to recommend. Tomorrow they will tell AI what to do: book, order, buy. The first requires a model that knows you. The second, one that can get something done with you.

Two levels: visibility and transactabilityA step list you can start todayA conversation with no commitment

By , updated 12 November 2026

We optimize visibility in:

What it looks like in ChatGPT

The same request, two different windows.

On the left, a hotel the agent cannot serve. On the right, a hotel with an MCP server that answers specific functions.

A hotel with no connection
A hotel with an MCP server

The windows are an illustration, not production screenshots. Function names and data are examples, and the look of the button depends on the client the agent runs in.

Scenario

Tuesday, 8:40 a.m. You ask your assistant for a hotel in London.

You don't open a browser. You say one sentence and put the phone down.

You to your assistant“Book me a hotel in London for 12 to 14 December, near Paddington, under £120 a night, with free cancellation.”

  1. Task split into conditions: city, dates, area, budget, cancellation.
  2. Found 37 hotels nearby (search, directories, direct sources).
  3. Dropped 34: no current price (14), cancellation terms in a PDF or by phone only (9), booking only through a contact form (11).
  4. 3 left. Availability and price checked for the chosen nights.
  5. Chosen: Westbourne Rooms, about £108 a night, cancellation up to 48 hours. Waiting for your yes.
  6. Booking confirmed. Paid with a token capped at £250. Confirmation by email.
Hotel A

Westbourne Rooms, visible and transactional

  • Prices and availability in a form a machine can read
  • Cancellation terms as data, not a PDF
  • Booking and payment with no human in the middle
  • A consistent description across several independent sources

It gets a £216 booking. Nobody called, nobody visited its website.

Hotel B

Grand Paddington, a great website, absent for the agent

  • A video on the homepage and design awards
  • Price: “request a quote”
  • Booking through a form and a call back
  • Cancellation terms in a PDF

Dropped within two seconds. It won't even see a visit in its reports, because the agent never opened the site.

Illustrative scenario. The hotels and numbers are made up, the mechanism describes what shopping agents do today.

Hotel B did not lose a tender. It was never invited to one. That is the change: visible to a human does not mean selectable by an agent.

The state of play

This is not science fiction. Check the dates.

Four signals, each with a source. Some solutions have already changed shape, the direction has not.

9 December 2025

MCP moves under neutral governance

The Model Context Protocol, an open standard for connecting agents to tools and data, became a project of the Agentic AI Foundation under the Linux Foundation. The foundation was co-founded by Anthropic, Block and OpenAI and is supported by Google, Microsoft and AWS, among others. Before that, function calling (a model invoking functions, OpenAI, 2023) had become the de facto standard.

Linux Foundation: formation of the Agentic AI Foundation (9 Dec 2025)
29 September 2025

Buying in the chat, then a turn back to the merchant

OpenAI and Stripe launched Instant Checkout and the open Agentic Commerce Protocol, first for Etsy sellers in the US. In 2026 OpenAI moved away from payment inside the chat itself toward a model where the purchase ends on the merchant's site or app. The form changed, the agent's role (discovery and decision) stayed.

Checkout.com: OpenAI's change of approach to shopping in ChatGPT (20 Apr 2026)
January 2026

Google standardises the shop for agents

The Universal Commerce Protocol was built with Shopify, Etsy, Wayfair, Target and Walmart. It lets agents in Gemini and in Google’s AI Mode search, build a cart and complete a purchase, and supports MCP and AP2 (payment authorisation), among others. A merchant needs a Merchant Center account, and the documentation covers lodging too.

InfoQ: Google Universal Commerce Protocol (24 Jan 2026)
October 2025 onwards

Card networks pay for agents

Mastercard said in October 2025 that the first agentic transaction had taken place on its network. Visa and Mastercard took part in Google's standard, and Visa's Trusted Agent Protocol is set to align with OpenAI's protocol. McKinsey forecasts that by 2030 agents could handle about a trillion dollars of transactions in the US.

Digital Commerce 360: Visa and Mastercard in agentic commerce (2 Apr 2026)

The scale is still small, the standards compete, and some experiments end. That is why it does not pay to build for one protocol. It pays to take care of what is common to all: data, availability and secure authorisation.

What it means for a business

The risk is not a lower Google ranking. It is being absent from the channel that decides for the customer.

There are two levels of readiness. They need different work and are easy to confuse.

01

Level one: visibility

Does AI know who you are, what you do and who you do it for?

  • Content answering real customer questions and prompts
  • A clear page and heading structure
  • Structured data: Organization, Product, Service, FAQPage
  • Consistent company data in listings and directories
  • Mentions in rankings, media and reviews

Result: you land on the candidate list.

02

Level two: transactability

Can an agent get something done with you, with no human in the middle?

  • Prices, stock and availability in machine form (feed, API)
  • Terms as data: returns, cancellation, delivery
  • schema.org actions, e.g. ReserveAction and OrderAction
  • An MCP server or API on top of your source system
  • Secure authorisation: OAuth, amount limits, confirmations

Result: you win from the candidate list.

One without the other does not work. Visibility without transactability gets recommendations that end at a competitor. Transactability without visibility gets a great API nobody knows about.

How to prepare

Where to start this week.

Ordered from the simplest to the hardest. Steps 1 and 2 need no developer. The times are our estimates, not measurements.

This explains mechanics, not promises. None of the large engines guarantees it will read schema.org actions or an llms.txt file. They are cheap hygiene that does not replace available data and real integration.

  1. Visibility audit

    visibility · 1 day

    Ask ChatGPT, Gemini, Perplexity and Claude ten questions your customers ask. Record whether you appear, at what position, whom they recommend instead and which sources they cite. A ready list of questions is in our 15-minute test.

    Result: you know where you stand and what is missing.

  2. Content and structured data

    visibility · 2 to 4 weeks

    Describe the company unambiguously and answer the audit questions with separate pages: price, comparison, who it is for, FAQ. Add JSON-LD (Organization, Service or Product, FAQPage), complete your Google profile and directory data. We described the mechanism in the piece on query fan-out.

    Result: the same test a month later shows the change.

  3. First simple actions for agents

    transactability · 1 to 2 months

    Expose prices, availability and terms as data: a product feed or Merchant Center, Offer and availability markup, potentialAction (ReserveAction, OrderAction). Add OpenAPI if you have an API. Treat llms.txt as an extra, since evidence that large engines use it is lacking.

    Result: an agent can read your offer without guessing.

  4. Full transactional integration

    transactability · a quarter and longer

    Put an MCP server or API on top of your source system (bookings, stock, CRM) with functions: check availability, quote, book, cancel. Secure it: OAuth, narrowly scoped tokens with an amount cap, human confirmation above a threshold, logs, rate limits. Delegate payments through a provider that supports agent protocols (ACP, AP2, UCP). Do not tie yourself to one.

    Result: an agent can order from you, and you stay in control.

AI readiness audit

Let’s see if an AI agent can choose you today.

Leave your details and book an audit. We’ll check visibility in four AI assistants, assess transactability (data, actions, API, authorisation) and give you a step list in implementation order.

In the audit you get

  • A visibility map: what ChatGPT, Gemini, Perplexity and Claude say about your business
  • A transactability assessment: what an agent can check and do with you
  • A step list from the simplest to the hardest
Kacper Bartoszak

Kacper Bartoszak

Marketing and AI search optimization expert

Owner of the PEOPLE HATE ADS* agency and SEO for AI. Has worked with brands such as Lech Poznań, Binance and Danone. Board member of the Startup Community Poznań Foundation.

LinkedIn → kacperbartoszak.com →

Prefer to write directly?
kontakt@seodlaai.pl

Book a free consultationNo obligation · online
What is your request about?

By sending the form you share data we will use only to contact you about the consultation. Details and your rights: Privacy policy.

FAQ

Questions about AI readiness

What is AI readiness?

It is a company's readiness for customers handing search, comparison and purchases to AI agents. It has two levels: visibility (AI knows who you are) and transactability (AI can get something done with you).

Are AI agents really buying for users already?

Yes, but so far at limited scale and in selected countries and categories. The standards (MCP, ACP, UCP, AP2) appeared between 2024 and 2026, and some deployments, such as payment inside the ChatGPT chat itself, changed shape. So the safest investment is in data, availability and authorisation, not in one protocol.

Do I need an online shop?

No. A transaction can be a booked slot, a quote request or a consultation sign-up. What matters is that an agent can complete it without a phone call or email on the way.

Do I need to build an MCP server?

Not at first. Visibility and data come first. An MCP server or API makes sense when you have a system an agent can realistically pull availability from or place an order into.

Is an llms.txt file enough?

No. It is a proposal, not a standard, and there is no evidence that large engines use it. It can be an extra, not a replacement for content, structured data and integration.

How long does it take?

A visibility audit is one day. Content and structured data usually take 2 to 4 weeks. Transactional integration depends on your systems and takes from several weeks to several months. These are estimates, not guarantees.

Let’s see if an AI agent can choose you today.

The AI readiness audit shows where you stand on both levels and what to do first.

Book an AI readiness audit