The stack we built for ourselves first
Get your business found by AI answers and agents.
An agent-ready business can be read, cited and paid by AI. AGMM builds the stack it runs on itself for UK owners of £1m to £50m businesses: a gate that keeps answer engines free and asks training crawlers to pay, a desk that shows who reads you, an MCP server and a content engine. Start with a free fit call.
Or call 0161 543 7189
- People Any browserFree, always
- Search engines Googlebot, Bingbot, ApplebotFree
- AI answer engines OAI-SearchBot, PerplexityBot, Claude-SearchBotFree, so you get cited
- Training crawlers GPTBot, ClaudeBot, CCBot, Meta, AmazonAsked to pay per page. Bulk copies are offered to AI training crawlers for a fee via x402
- Anything unknown No crawler name, or a name we cannot trustFree and watched, never charged on a guess
This is AGMM’s live policy. Read it at agmm.co.uk/x402/policy/.
The problem
AI answers now decide who gets found, and most business sites are shut out without knowing.
Ask ChatGPT, Perplexity or Claude for a supplier in your trade and it answers from the pages it was allowed to read. If your site turns those readers away, you are not in the answer, however good it looks to a person.
The block is rarely a decision. A security setting, a template robots.txt or a bot rule does it, and nobody tells you. It happened to AGMM. Until 6 August 2026 a Cloudflare setting put a hard Disallow for GPTBot, ClaudeBot, CCBot, Google-Extended, Applebot-Extended, Amazonbot, Bytespider and meta-externalagent in front of our own robots.txt. Our file said allow. The site served block. Our own check, which fetches the site as each crawler and fails on anything that is not a 200, caught it, and we fixed it. You only know your own position by running that check.
Signs you have it
You will recognise some of these.
- Ask ChatGPT or Perplexity who you are and what you sell, and the answer describes an older version of the business, or nothing.
- Your robots.txt looks fine, but nobody has fetched a page as GPTBot, OAI-SearchBot or PerplexityBot to see what comes back.
- Your CDN or security tool has an AI bot setting, and nobody remembers which way it is set.
- You hold content people search for or pay for, such as listings, a catalogue, a directory or professional advice, and you cannot say who reads it besides Google.
- Visits from AI assistants are not counted as their own thing in your reports.
- There is no llms.txt and no machine-readable description of what you sell, and your best knowledge sits in PDFs and in people’s heads.
What we build
What does an agent-ready business build include?
An AI crawler gate, with payments
A gate in front of your site that tells people, search engines and AI answer engines (free) from training crawlers, and asks the training crawlers to pay per page with x402, in USDC on Base. Unknown bots stay free and watched, never charged on a guess. If the gate breaks, visitors still get the page.
A crawler desk
A private dashboard showing who reads you, which pages they want, what they were asked to pay and what they paid. Counts are grouped by class and network, and scanners wearing a crawler’s name are separated out.
An MCP server for your knowledge
A Model Context Protocol server, so AI agents can query your own guides, listings or catalogue directly instead of scraping pages. Search can stay free and the structured or bulk data can be paid. AGMM’s own runs at agmm.co.uk/mcp.
Discovery files
An llms.txt, a Markdown copy of each page, an OpenAPI description, a company facts file and a listing where agents search, so a machine can tell what you sell in one request.
A GEO content engine
A pipeline that finds the questions buyers put to AI assistants, drafts answers from your own material, checks every figure against its source and publishes them in the shape answer engines quote: the answer first, dated, with a named author and structured data.
What you can check
Every part of this runs on agmm.co.uk, and you can check it.
The gate, the desk, the MCP server and the policy are live on our own site. Read the policy at agmm.co.uk/x402/policy/, the machine copy at agmm.co.uk/crawlers.json, and the public payment count at agmm.co.uk/x402/stats.json. The first payment settled on Base on 19 October 2026, and its transaction is public.
Window: 15:00 to 21:39 UTC on 19 October 2026 (6 hours 39 minutes), the stretch in which the gate was asking training crawlers to pay and every request was classified.
- Last 24 hours, every request11,181Meta’s crawlers made 3,072 of them, 27%. The 24 hours to 21:39 UTC.
- The window below, every request4,187People and bots together, QA traffic excluded. Most name no crawler.
- One burst at 18:31 UTC345344 came from one Google Cloud network, each with a different crawler’s name, asking for .env files and keys. Left out of the three rows below.
- Search engines185Googlebot, Bingbot, Applebot, Baiduspider. All free.
- AI answer engines and assistants98OAI-SearchBot, PerplexityBot, Claude-SearchBot, ChatGPT-User and others. All free, none asked to pay.
- Training crawlers53890 asked to pay (Meta’s crawler 87, Amazon’s 3). 55 asked for the whole-site exports, which the rule at the time refused; bulk copies are now offered to AI training crawlers for a fee via x402. 393 read the free front-door pages.
- Payments settled on Base1In the window: 19 October 2026, 20:13 UTC. A second settled at 22:18 UTC, after it. Public, and checkable on-chain.
Counted from AGMM’s own production database. Classes, operators and counts only: no IP address and no user-agent string is published.
Check the public payment count at agmm.co.uk/x402/stats.json. See how to read a report like this one and how we run the rest of our own business. AGMM’s MCP server is at agmm.co.uk/mcp, and our listing in Coinbase’s Bazaar, where agents with wallets search for paid services, was indexed on 19 October 2026.
Who it is for
Who is an agent-ready build for?
Owners of established UK businesses with something worth reading. The more of your value sits in pages, records and knowledge, the more it matters who is reading them.
Listings
Property, vehicles, jobs, stock. Agents ask what is available now, so fresh and structured is what gets quoted.
Catalogues
B2B products and parts. Specifications and part numbers are what an assistant quotes when someone asks which supplier.
Directories and registers
The whole dataset is the asset, and a bulk copy is exactly what training crawlers want. This is where a gate pays for itself first.
Publishers and trade bodies
Articles and guides get quoted by answer engines and copied into training sets. You want the first and a say over the second.
Professional knowledge
Advisers, consultancies and specialist firms. Your explanation is the answer, and being cited is the enquiry.
The process and the cost
What does it cost? The audit is £2,000; the build is proposed after it
- The AI Constraint Audit
- £2,00014 days, fixed
- The build
- By proposalPriced against what the audit found
- Ongoing support
- By proposalOnly if you want it
One price is published on this page and it is the diagnostic: the AI Constraint Audit, £2,000, fourteen days, three sixty-minute working sessions. The build is proposed after it, in its own document, priced against what the audit found and agreed before anyone writes code. We do not quote ranges, because the gate, the desk and the engine each depend on your stack and your content. The audit fee is not credited against a build, so we have no reason to recommend one that the evidence does not support.
What does the £2,000 cover before anything is built?
Four documents you own and can take to any other supplier: a business-wide constraint map, an AI opportunity map, an economic case for each option and a ranked build proposal. Where your site holds content worth reading, the audit includes a read of your crawler exposure: what AI systems can reach today, what they are turned away from, and which of your pages are worth gating, opening up or leaving alone.
The whole ladder, including what we do not bill for, is on what it costs.
How long
How long does it take?
Fourteen days from the first working session to the written recommendation. A build runs in stages, and the first is usually the gate in watch-only mode, which charges and blocks nothing and counts who reads you. AGMM’s own ran watch-only from 18 October 2026 and began asking training crawlers to pay on 19 October. Dates for yours go in the proposal, once we know your stack.
- Day 1First working session, after the free fit call
- Day 14Written recommendation, yours to keep
- ThenThe gate in watch-only mode, plus discovery files
- ThenCharging, the MCP server and the content engine, once you have seen who reads you
Not in the price
What this does not include
A promise of income from crawlers
No named training crawler is known to carry a wallet today, and the x402 market is small. The first value is control and visibility. Payment is a position for the day crawlers pay, not a forecast.
A guaranteed place in ChatGPT’s answers
Nobody controls what an answer engine says. We make you easy to read and quote, then measure what happens. We guarantee no ranking and no citation count.
Hosting, CDN and wallet fees
We resell nobody’s services. The gate runs on infrastructure in your own accounts, and any fees are paid direct.
Tax treatment of what you receive
Payments arrive as USDC, which your accountant should record as income. The desk exports each payment for them. We set up the records, not the treatment.
People and machines
Where does a person stay in the loop?
You decide what is free, what is asked to pay and what is refused, and the gate does exactly that and nothing else. Anything unknown stays free. A person reads the desk, and a person approves any change to the policy.
Proof it worked
How do we know it worked?
On counts you can check. How many AI crawler requests get a 200. How many answer-engine fetches reach your pages each week. Which of a fixed list of buyer questions name you when we put them to ChatGPT, Perplexity and Claude each month. How many payments settle, if you chose to be paid.
Known risks
What tends to go wrong on this kind of build?
Three things. A gate that treats an answer engine as a training crawler and costs you citations, so answer engines sit on a named free list. A gate fault that blocks people, so ours fails open and a broken gate serves a free page. And spoofed names: any request can wear a crawler’s name, and 344 did in five seconds on 19 October, so we never charge or block on a name alone.
Fit, stated plainly
What do we need from you?
- Access to your DNS and CDN settings, or the person who has it.
- One person who can say which content is yours to gate and which should stay free to read.
- Your last 30 days of server or CDN logs, if you keep them.
- A wallet address on Base, only if you want to be paid.
Who this is not for
- A site of a few brochure pages with nothing worth reading or citing.
- Anyone who wants guaranteed revenue from crawlers.
- A business that wants every AI system blocked. That is a settings change we will explain on the call, not a build.
Questions
What do people ask before they start?
- What is an agent-ready business?
- A business whose website and knowledge can be read, cited and paid for by AI systems on purpose. Answer engines like ChatGPT and Perplexity can reach every page you want quoted. Training crawlers are counted and asked to pay. AI agents can query your knowledge through a documented route instead of scraping.
- Will charging AI crawlers hurt my search visibility?
- Not if the gate is built the way ours is. Search engines and the AI answer engines that fetch pages to cite you stay free, because a citation is a visit you want. Only named training crawlers are asked to pay, and anything unknown stays free. OpenAI says a site that opts out of OAI-SearchBot is not shown in ChatGPT search answers, so we never charge it.
- Do AI crawlers actually pay?
- Not at any scale yet. The first payment settled on AGMM’s gate on 19 October 2026, and the running count is public and checkable on-chain. No named training crawler is known to carry a wallet today. We build the gate for control now and for the day they do, and we say so before you buy.
- What is an MCP server, and why would my business want one?
- MCP, the Model Context Protocol, is a standard way for an AI agent to ask a service questions and get structured answers. An MCP server over your own guides, listings or catalogue lets agents query what you know instead of scraping pages. AGMM runs one at agmm.co.uk/mcp, where search is free and the structured data is paid.
- What is GEO, and how is it different from SEO?
- GEO is generative engine optimisation: making your pages easy for AI answer engines to find, understand and quote. SEO gets you ranked in a list of links. GEO gets you named inside the answer. They share foundations, such as clear pages, structured data and crawl access, so work on one helps the other.
- How do I know whether my site is blocking AI crawlers now?
- Fetch your pages with each crawler’s name and read the status codes. A 200 means it got in. A 403 or a challenge page means something at the edge is turning it away, whatever your robots.txt says. Our free AI visibility check does the first pass for any domain, and the exposure report shows the ten-minute manual version.
- What does it cost?
- The audit is £2,000, fixed: 14 days, three 60-minute working sessions and four documents you keep. The build is custom, priced by proposal after the audit and before anyone writes code. We do not quote ranges, and the audit fee is not credited against a build.
- Do I need a crypto wallet?
- Only to be paid. The gate can run in watch-only mode first, which charges and blocks nothing and counts who reads you. To be paid you need a wallet address that receives USDC on Base. Receipts are income, so your accountant should set how they are recorded, and the desk exports each payment for them.
- Who is this for?
- Owners of established UK businesses with something worth reading: listings, a catalogue, a directory, price or availability data, or professional knowledge people search for. If your site is five brochure pages, the audit will probably tell you to fix other things first.
- How long does it take?
- The audit takes 14 days. A build follows in stages, usually starting with the gate in watch-only mode, so you see who reads you before anything is charged or blocked. Charging, the MCP server and the content engine come after that. Dates go in the proposal, once we know your stack.
Free to read first
Would you rather work it out yourself?
Everything below is free, with no account and no paywall. If it solves the problem without us, that is a good outcome.
Find out who is reading your business.
A free 30-minute call. We look at what AI systems can reach of your site today and tell you whether the £2,000 audit is worth doing. Nothing is sold on the call.
- Before day 1Fit callFree, 30 minutes
- Day 01Map the businessConstraints and weak points
- Day 07Price the opportunitiesWhat could save or earn
- Day 14Rank the investmentFour outputs you keep
Then the proposal. The fuller audit is always part of it.