B2A marketing: the five rungs of AI visibility for professional firms

Somewhere in Coral Gables tonight, a person whose father just died is going to open ChatGPT instead of Google. They will type the whole situation: the house, the trust, the brother who is already arguing. They will get back a short answer and two or three firm names. They will call one of them. They will never see a search results page, and they will never see the website of any firm that was not in that answer.

What is B2A marketing?

B2A stands for Business to Agent. The term started circulating in agency circles this year (51blocks published a short guide on it in early 2026, and the label is a good one). It names something most professional firms have felt but not framed: there is now a second audience for your marketing, and it does not read like a person.

The AI agent between you and your prospect does not care about your hero image or your tagline. It cares whether it can fetch your pages, whether it can tell who you are and what you do, whether other sources agree with you, and whether there is a sentence on your site specific enough to quote. It rewards clarity, structure, consistency, and depth. It skips vague, promotional, or incomplete content because that content is hard to use.

B2A does not replace what you already do. It adds a layer underneath it. Most of that layer is structural, which is the good news: you rarely need more content. You need the content you have to be readable, specific, and backed up elsewhere.

Why a ladder instead of a checklist?

Every guide on this topic hands you a checklist. Checklists are fine. They do not tell you where your firm actually stands, and they do not tell you what to fix first. Most owners read one, feel behind on everything, and do nothing.

So we built it as a ladder. Every firm sits on one of five rungs right now. Each rung has a plain test you can run yourself in ten minutes, in ChatGPT, Perplexity, Gemini, or Google AI Mode. Run the test. If you pass, move up. The first rung you fail is where your work is. That is the entire method.

Rung One: Found. Can AI read your site at all?

More firms sit below this rung than you would expect. Their sites look fine to a human and are close to empty to a machine. The usual causes: the text is rendered by JavaScript instead of sent as HTML (common in single-page apps built on React and similar frameworks), a security plugin is blocking GPTBot, ClaudeBot, PerplexityBot, or Google-Extended in robots.txt, or the facts that matter live inside images and scanned PDFs.

The test: ask Perplexity, or ChatGPT with browsing on, to summarize what your firm does using only its website. A pass is an accurate, specific summary. A fail is a refusal, a vague guess, or a confident summary that is wrong. Wrong is worse than nothing, because it means the system is filling gaps with assumptions. Then view your homepage source and search for a sentence you know is on the page. If it is not in the source, your text is being rendered by script.

Rung Two: Understood. Does AI know who you are, what you do, and where?

Being readable is not the same as being understood. At this rung, AI can build an accurate model of your firm: the entity, the services, the audience, the place. It builds that model from structure, not from adjectives.

"We provide exceptional, client-focused legal solutions for individuals and businesses" contains no information a system can use. "We handle probate, trust administration, and estate litigation for families in Miami-Dade County, with an office in Coral Gables" does. Add schema markup (LegalService, MedicalBusiness, FinancialService, or RealEstateAgent, plus Person for each named professional and FAQPage where you answer questions), give each core service its own page, and open every page with a direct 40 to 60 word answer to the question a client would actually ask.

The test: ask what your firm specializes in, who its clients are, and where it is located. Compare the answer to the truth. Then ask what questions your firm answers on its site about your main service. If the system can list specific questions and answers, you are understood.

Rung Three: Trusted. Do other sources say the same thing?

Your own website is one witness. AI systems look for corroboration before they put a firm in front of a person. Your name, address, phone, and services need to match across your Google Business Profile, your bar or licensing profile, directories, review platforms, and LinkedIn. Named professionals need real, verifiable credentials on the page. And someone other than you needs to have said something about you.

This is where regulated practices need to be careful. Florida's Bar rules, the FTC's endorsement guides, and HIPAA all have something to say about reviews and claims. Never pay for reviews, never fabricate them, and never respond to one in a way that confirms a client or patient relationship. "Thank you for the note. We are glad we could help." is enough.

The test: ask whether your firm is a reputable practice in your city and what reviews and other sources say. Then search your firm name plus your city and read the knowledge panel. Does the address match? Are the people current? Are the reviews answered?

Rung Four: Cited. Does AI name your firm in the answer?

This is the rung most people mean when they say AI visibility. Everything before it is preparation. The single biggest blocker here is that firms optimize for keywords while clients ask questions. "Probate attorney Coral Gables" is a keyword. "My father just passed and my brother is already arguing" is a prompt. AI answers prompts.

The fix is a prompt bank: ten to twenty real client questions per practice area, written the way clients type them at 11pm. Each one becomes a page heading, an FAQ entry, and a monthly test. Answer each directly in the first 40 to 60 words, then make the answer specific enough to quote: timelines, thresholds, steps, jurisdictions, dates, and the statute or rule where one exists. Sign and date the work with a named author.

The test is the most valuable ten minutes in the whole method. Run all ten prompts in ChatGPT, Perplexity, Gemini, and Google AI Mode. Note whether your firm was named, whether it was linked, and which competitors appeared. Save the baseline. Run it again next month. No one can promise you a citation, and anyone who does is selling something else. What you can control is whether you are readable, understood, trusted, and measuring.

Rung Five: Chosen. Does the citation become a client?

Being on the shortlist is not the win. The consultation on your calendar is. The client, or increasingly the client's AI agent, has to get from the answer to a booked appointment without friction. The page they land on should continue the conversation they were having, not switch to "comprehensive legal solutions." The next step should be an action with a time: "Book a 20-minute consultation," not "Contact us." And someone should respond fast, because the client now has two other names in the same answer.

AI agents are beginning to act, not just recommend. They read hours, dial numbers, and fill forms. Hours that live only in an image, a form that breaks without script, a phone number that is not marked up in schema: each one is a door an agent cannot open.

The test: ask an AI system how to contact your firm and book a consultation. Then do exactly what it says, as a client would, and time how long it takes a human to respond.

Where to start

Do not fix the whole site. Pick one service page, the one that brings in the clients you most want more of, and walk it up the ladder. Run the five tests and write the baseline down. Fix the Rung One and Two blockers on that page: text, schema, entity statement, question headings, direct answers. Fix the record: Google Business Profile, licensing profile, directories, named people with credentials. Add the neighboring questions. Set a response standard. Re-run the ten prompts and compare.

Most regulated practices can move one page from Rung One to Rung Three in a couple of weeks of structured work. Once you can see the pattern on one page, the rest is repetition. The full guide below has the tests, the before-and-after rewrites, the compliance notes, and a one-page self-score table you can print.