The New Visibility: Why Professional Firms Must Optimise for AI Agents, Not Just Search

Collecting client documents is a critical step for any professional firm, whether you are an accountant, a lawyer or a broker. But before you can collect a client's documents, that client has to find your firm in the first place.

For years, being found meant ranking on the first page of search results. You optimised for keywords, hoped the person reading your page found it convincing, and waited for the phone to ring. That is changing. More and more of the searching is now done by an AI agent working on the client's behalf, which is why the shift has picked up a name of its own: Business-to-Agent, or B2A.

When a potential client wants to know whether a local accountant handles cross-border VAT, they increasingly ask ChatGPT, Claude, Gemini or Perplexity. The agent does the research, compares the options and comes back with a short list. If it cannot read your firm clearly, you are not on that list.

Three things change as a result.

1. The Machine Needs Proof, Not a Pitch

A well-designed page and a confident tone go a long way with a person. A language model reading the same page looks for something else: facts it can check, stated the same way everywhere it finds them.

For a professional firm, those facts are plain ones. Which services you offer and which you don't. Your fixed fees, or how fees are calculated. The jurisdictions and regions you cover. The languages you work in. The professional body you are registered with, and your registration number. Who the partners are and what each of them handles.

If your pricing page says one thing and a PDF price list from two years ago says another, the agent has two answers and no way of telling which one is current. It may quote the wrong one, or pass over your firm for one whose information agrees with itself. The same goes for an old service page that still ranks, a former partner's profile that was never taken down, or a directory listing with last year's address.

Keeping what search engines and AI crawlers can reach limited to what is true today is sometimes called index hygiene. The machine layer of your site (robots.txt, sitemap.xml, structured data such as Organization, Person and AccountingService or LegalService, and newer files like llms.txt) has to tell exactly the same story as the pages people read.

2. Agentic Commerce Is Already Here

The bigger shift is what happens after the agent finds you. In agentic commerce, the client's agent doesn't stop at a recommendation: it places the order and pays.

Picture a client who needs a first consultation with your brokerage. Instead of browsing your site, filling in a form and waiting for a reply, their agent talks directly to your firm's agent-facing system. It reads what you offer and at what price, picks a slot, books it and pays the deposit, within limits the client set in advance.

This already runs in production for service businesses. Brand Design Ltd., a branding agency in Varna, sells its own services this way and has documented how it works. Its site publishes a machine-readable catalogue with prices and terms. An agent can register, place an order and pay with nobody on the seller's side, while the payment itself stays under the buyer's approval. Services without a fixed price wait for a person to quote them. On 21 September 2026, an outside participant gave their agent only a service code and a budget of one euro. The agent found the offer on its own, paid EUR 0.99 for it and passed a public verification.

That split between fixed-price and quoted work maps neatly onto professional services. A standard tax return, a company formation or a first consultation can be sold end to end. Anything that depends on judgement, such as a dispute or a complex restructuring, still needs a person to look at it before a price exists. The agent can gather the facts and open the conversation; the professional decides.

And once the engagement is agreed, the work still starts where it always has: with the client's documents.

3. Measuring What the Machine Sees, and What It Says

How do you know whether an AI recommends your firm? Search Console reports clicks from traditional search. It says nothing about how often a model names you in an answer, or which of your pages it uses as a source.

Two separate things need checking. The first is the state of your own site: whether it is structured, consistent and easy for a model to read. Brand Design's free AI Brand Audit checks this for any address you enter. It scores structured data, how readable the content is to a model and how consistent the brand signal is, then emails a report of what to fix.

The second is what the models actually say. That means asking the questions your clients ask, across several assistants and repeatedly over time, and recording who is named, in what context, and which pages are cited. Brand Design tracks this for its clients as AI share of voice: how often a brand is named or cited, against its competitors, for the same set of questions. A small firm can start by hand. Write down ten questions a new client would ask ("Which accountants near me handle VAT for Amazon sellers?"), put them to three or four assistants once a month, and keep a simple log of who comes up.

Where to Start

• Make your facts agree everywhere. Services, fees, areas, languages, registrations and people. Remove or redirect whatever is out of date.
• Mark them up. Use Organization, Person and the schema type for your profession, so an agent doesn't have to guess.
• Decide what an agent may buy without you. Which fixed-fee services can be booked and paid end to end, and which always wait for a person.
• Check both sides. The state of your site, and what the assistants actually say about you.

Agents already do the searching and the chasing. The firms they recommend will be the ones whose information is the clearest and the easiest to verify.