Search tools are now being built for AI agents as well as people. If a buyer hands the first round of supplier research to one, what it can establish about your firm depends on the evidence you’ve published.
Your next prospective client might not start on your website. They might start with a brief:
‘Find three engineering consultancies with experience of decarbonising occupied industrial facilities. Explain their relevant experience, flag any gaps in the evidence and give me sources.’
Nobody is claiming procurement has gone autonomous. But the tools needed to run that kind of brief exist today. Exa, a company building search for AI, offers a connector that lets a connected assistant search the web, read pages and run multi-step research, including building lists of companies with sources attached.
So the question for a firm that wants to be considered now has two parts. Will the buyer like your website? And can the tools working for that buyer establish why you belong on the list?
Search is being built for a different user
In May 2026, Exa raised $250 million at a $2.2 billion valuation, in a round led by Andreessen Horowitz. At the time, the company reported that more than 5,000 companies and 400,000 developers used its services.
Those are Exa’s own figures, and a funding round tells you nothing about how your buyers behave. It does show investor confidence in a different use of search: results a piece of software can act on, rather than only a page of links for a person to browse.
For expert-led firms, the valuation matters less than what it funds. A buyer can now delegate more of the work of finding and comparing suppliers. Your firm may be assessed before anyone on the buying team has opened your homepage.
An answer is one thing. A brief is another
‘Explain industrial decarbonisation’ is a question. ‘Find consultancies with relevant experience, compare them and show me the evidence’ is a research brief. The second means identifying firms, finding material about each one and testing it against criteria.
The output isn’t due diligence. People still need to check sources, question omissions and apply judgement. What changes is the potential starting point: the buyer reviews a reasoned longlist instead of building one from scratch.
We covered the wider shift in our article on AI search visibility for B2B firms. This piece is about the next step, when the tool actively investigates which firms fit.
What MCP changes, and what it doesn’t
Model Context Protocol, or MCP, is an open standard for connecting AI applications to external tools and data.
In one possible setup, a buyer gives a connected assistant a brief, the assistant calls a search tool through an MCP connection, the tool finds and reads public material, and the assistant writes up a comparison with sources. Exa’s MCP server supports the search and research capabilities needed for this kind of workflow.
Two points get lost in the excitement.
First, you don’t need your own MCP server to be researched this way. The connection sits between the buyer’s assistant and the search service. Your public web presence is what gets read.
Second, MCP is plumbing. It gives an assistant access to a research tool. It can’t supply evidence about your firm that you haven’t published. ‘Optimise your website for MCP’ is advice that misses the point.
Building your own MCP server is a separate decision about product, data and security. It’s worth exploring if you want a connected assistant to use a particular service or access data you make available, as a software business might. MCP servers expose defined capabilities to AI applications; they are not simply another form of website optimisation.
A consultancy shouldn’t build one because the acronym came up in a board meeting.
The real gap is evidence
A project page can show the finished work while leaving out the reasoning that made it possible.
Take two hypothetical consultancies, equally capable. The first describes ‘a collaborative team delivering sustainable solutions’. Its project pages offer strong photography, a client name and a paragraph on the result. The second explains the type of facility, the client’s problem, the firm’s precise role, the constraints it worked within and the outcome it can substantiate.
The second has given a researcher, human or software, much more to judge against a brief. What that looks like depends on the firm:
Architecture practice: explain the decisions behind a project, as well as showing the building.
Engineering consultancy: separate your contribution from the wider project team’s.
Deep-tech company: tie technical performance to a defined operating environment and commercial use.
Specialist adviser: say when your approach applies, who delivers it and what experience backs it.
None of this means publishing confidential detail or turning every engagement into a long case study. It means making relevant experience explicit. That is the argument behind an authority SEO strategy: tactics don’t compound without demonstrated expertise. Delegated research raises the stakes on the gap between what your firm knows and what an outsider can verify.
We’ve seen an early version of this. An executive search firm we worked with reported that prospective clients and new contacts had discovered it through podcast conversations and AI-assisted research, then made contact already familiar with its perspective and expertise. Our executive search marketing case study describes the integrated work behind that visibility: a clearer specialist position, a rebuilt website, a podcast and published insight.
That is not evidence of autonomous procurement. It is an example of a firm’s published expertise helping people discover and assess it before making contact.
Your website is part of the evidence
A research brief can ask an agent to distinguish substantiated experience from marketing claims and find independent sources. Exa’s research tools support multi-step investigation and checking findings against criteria.
So the rest of your public presence deserves attention too: professional profiles, speaker biographies, interviews, directory listings and old service descriptions still in circulation. Do they describe the firm you are now?
Consistency doesn’t mean identical copy everywhere. It means one recognisable account of what you do, with the distinctions a researcher needs: company experience versus individual experience, completed work versus ambition.
It also means being specific about whose expertise supports a claim. A senior hire’s experience may be highly relevant without being a project the firm itself delivered.
The aim is not to make every source sound the same. It is to give an unfamiliar researcher an accurate picture.
Where llms.txt fits
llms.txt is a proposed Markdown file that gives agents a short guide to a site, with links to more detailed information. It is intended to help them locate and use relevant content.
It remains a proposal, and Google says its Search systems, including their generative AI features, do not use llms.txt. That does not establish how other agents use it.
A signpost also can’t improve what it points to. If a case study doesn’t explain your contribution, directing an agent to it does not fill the gap.
Treat llms.txt as an optional aid, never as the strategy.
Three questions for your leadership team
Start with a brief you’d want to win, then test your public evidence against it.
Can an outsider establish that you fit?
Are the sectors, services, project types and circumstances named? A broad capability statement can be accurate and still leave a stranger unable to judge.Can they substantiate it?
Is there evidence of the work: the challenge, your exact role, the constraints, the result and what is attributable to your team?Can the material be reached and read?
Is the important information available as readable page text, with related pages connected and no unintended restrictions blocking the tools you want to reach it?
If an agent leaves you out, don’t read one generated shortlist as a verdict. Repeat the exercise, inspect the sources and look for the cause: a missing project explanation, an outdated description, an unclear specialism or a limitation of the tool.
The useful outcome is not a screenshot showing that an assistant mentioned you once. It is a clearer understanding of what your public evidence does and does not establish.
Which expertise is visible outside your firm?
It’s too early to call agent-led supplier research standard buying behaviour in the markets we work in. It is a scenario worth preparing for because the preparation also helps human buyers: clearer positioning, accessible information and evidence that supports your claims.
Consider how much of the reasoning that makes your firm the right choice lives in proposals, conversations and senior people’s heads. Inside the firm, that is the problem AI brand intelligence addresses. This is the same problem facing outward.
You don’t need to redesign the business around a protocol. You need to decide which expertise matters commercially, what evidence supports it and how clearly that evidence is published.
If a prospective client asked an agent to find a business like yours, what evidence would it have to choose you?
If you’d like a second view on what a buyer’s agent could establish about your firm, contact us to arrange a free 15-minute introductory conversation.
