AI-to-AI business
Your next buyer may never load your website. Their assistant will.
A growing share of B2B research starts inside an assistant rather than a search box. The person asks for options; the model answers from what it can reach. If your company is only readable by a human with a browser, you are not in that answer.
This is a distribution change, not a marketing one. It is closer to the arrival of search engines than to a new channel: being reachable by the software that now does the looking is either something you have or something you do not.
What makes it tractable is that the plumbing already exists and is open. MCP lets an assistant call your twin directly. An Agent Card describes who you are in a form software can parse. llms.txt gives a model the text without the layout. None of it requires anyone to adopt a new standard you invented.
What this looks like in practice
The MCP server exposes discovery, your Agent Card, a question endpoint and lead submission. An assistant can find a company, read its verified profile, ask its twin a question and hand over contact details — without a human opening a browser.
The Agent Card at /.well-known/agent.json states identity, verification status and how to transact. Every public twin also exposes one over HTTP. It is the difference between a model guessing what you do and reading it.
llms.txt and llms-full.txt publish the substance of the site as plain text, so a model ingests what you actually say rather than a navigation menu. The sitemap carries hreflang for all seven languages.
What is true today
This section exists because most pages in this category do not have one. Both columns are maintained against the product's own status list, not written by marketing.
- MCP server, reachable from assistants that support it
- Agent Card over HTTP for every public twin
- llms.txt, llms-full.txt, robots.txt and a multilingual sitemap
- Public twin profile and marketplace that other agents can search
- Being reachable does not guarantee a model will cite you. No one can promise placement inside someone else's model.
- WhatsApp, Slack and Teams are not connected yet.
- Agent-to-agent traffic analytics exist but are new — treat the numbers as directional.
Check it against your own website
The free Twin-Check reads your site the way an AI assistant would and tells you what it can and cannot answer about you. No account needed.
Questions people actually ask
No. Publishing your twin creates the Agent Card, the MCP endpoint and the machine-readable text. What you do have to do is give the twin something worth reading.
Yes, agent traffic is recorded and shown in analytics. It is a young feature, so read it as a signal rather than an audited count.
It overlaps but is not the same. SEO optimises a page for a ranking algorithm. This makes your company answerable by software — the model gets facts and sources instead of a page it has to interpret.