---
title: "A2A for Law Firms: Is Your Website Ready for AI Agents?"
description: "I tested Agent2Agent (A2A) on my own site. What law firms need to know about agent-ready intake and the ethics rules that still apply."
url: "https://carlosarias.com/blog/law-firm-marketing/your-next-law-firm-client-may-never-visit-your-website"
---

[Law Firm Marketing](/blog/categories/law-firm-marketing)

# A2A for Law Firms: Your Next Client May Never Visit Your Website

A2A for law firms, tested on my own site: what your Agent Card must declare, and the ethics rules that still apply to agent-ready intake.

  [Carlos Arias](/blog/authors/carlos-arias) · October 18, 2026  · 14 min read

![Ink sketch: a person with a briefcase walks away from an open laptop toward a columned building.](/_astro/cover.DrxFEmQL_Z1LjuHB.webp)

*Ink sketch: a person with a briefcase walks away from an open laptop toward a columned building. AI-generated illustration by Carlos Arias .*

The agent-first web is coming. Is your law firm ready for A2A?

I built an A2A interface on my own site and let an outside AI agent use it. Here is what A2A for law firms actually requires, and where it is still too early.

By Carlos Arias, Marketing Engineer for Law Firms

> Before you read: This article is about marketing and web architecture. It is not legal advice. I’m a marketing engineer, not a lawyer, and the professional-conduct rules that govern intake and advertising vary by state. Run any intake or advertising change past your firm’s ethics counsel or your state bar’s ethics hotline. Full disclosure: I build agent-ready websites and intake systems for law firms, so I have a stake in this topic.

## Key takeaways

- A2A (Agent2Agent) is an open protocol that lets AI agents discover each other and exchange tasks directly, without one of them pretending to be a human clicking through a website. It reached a stable v1.0 in March 2026 and is governed under the Linux Foundation.
- I implemented A2A on CarlosArias.com and had an outside AI agent find it and send me an inquiry. It failed once on a missing field, read the error, fixed its request, and got through. No form. No browser.
- Prospective clients are already handing their research to AI. In a 2026 consumer study, 41.9% said they’d use ChatGPT to research which lawyer to hire, up from 9% in 2023.
- SEO and AEO get your firm found. A2A lets an agent take the next step. They’re different layers, not replacements for each other.
- The ethics rules don’t change because the visitor is software. Prospective-client confidentiality and the advertising rules both apply to an agent-facing intake channel. So do your supervision duties. Design for all of it from day one.
- This is early. Most production A2A traffic today is enterprise-to-enterprise. Start small and measure.

For 30 years, we’ve built websites on one assumption: a human being will visit.

They search Google, click a result, read your practice area pages, and eventually fill out a form or pick up the phone. We’ve optimized every step of that path. Faster pages. Clearer navigation. Better calls to action. More landing pages than anyone wants to admit.

But what happens when the person looking for a lawyer never visits your website?

What happens when they send an AI agent instead?

I think we’re heading into the biggest change in website architecture since search engines. Over the next six to twelve months, I expect AI agents to take on a lot more of the errands people used to handle by browsing. That’s my prediction, not a guarantee. For law firms, it changes how prospective clients find attorneys and how they make first contact.

So I stopped predicting and tested it on my own website. It worked.

## Why are we making AI agents use websites like humans?

Every element on your website exists because humans need a visual way to talk to computers. Menus. Buttons. Forms. Scheduling widgets.

An AI agent doesn’t need any of that. It doesn’t need to see a blue “Schedule a Consultation” button. It doesn’t need to scroll, hunt for the form, guess which field wants a phone number, or simulate a mouse click.

Yet that’s how most agents interact with the web today. They launch a browser, load the page, guess which element is the form, type into fields, and hit submit. Some spin up an entire virtual computer to do it.

As an engineer, that drives me a little crazy. This is the part of the job that made me throw out four job titles and keep one: we’re making software operate an interface built for eyes and fingers. Multiply that by millions of agents running browsers to send what amounts to a few lines of structured data, and you get an enormous amount of wasted compute, plus brittle automations that break the moment someone redesigns a contact page.

Why launch a browser to fill out a form when two systems could just talk?

AI agents shouldn’t have to browse your website. They should be able to communicate with it.

That’s the problem A2A solves.

## What is A2A for law firms?

A2A (Agent2Agent) is an open protocol that lets AI agents built by different companies, on different frameworks, discover what another agent can do and send it work directly. Google announced it in April 2025 and donated it to the Linux Foundation that June. The project shipped v1.0, its first stable, production-ready release, in March 2026 (Google Open Source Blog). At its one-year mark, more than 150 organizations supported it, with deep integration across the major cloud platforms (Linux Foundation). In August 2026 it joined the Agentic AI Foundation, alongside MCP (A2A Project).

Here’s what law firms need to understand: A2A doesn’t replace your website. It adds a second door.

You publish a small machine-readable file called an Agent Card at a standard location on your domain: /.well-known/agent-card.json. It tells any compatible agent who you are, what your agent can do (its “skills”), where to send requests, and what authentication it requires. An agent that wants to work with you reads the card, then sends a structured message to your endpoint.

Picture a personal injury firm with an A2A-enabled intake agent. A prospective client tells their AI assistant:

> “I was hurt in a car accident in Miami. Find me a personal injury attorney who handles these cases and ask whether they offer free consultations.”

Today, that assistant searches, opens websites, reads practice area pages, and tries to fill out contact forms. With an agent-ready firm, it reads the firm’s Agent Card and confirms the firm handles auto accidents in Miami. The inquiry goes straight to intake. The firm’s agent answers with how consultations work and what happens next. The person reviews it and decides whether to go ahead.

No browser. No form. No guessing which button submits.

Both sides need compatible systems for this to work, and anything sensitive needs permissions and safeguards. I’ll get to those. But the protocol isn’t hypothetical. It’s a published specification with official SDKs in Python, JavaScript, Java, Go, .NET, and Rust (A2A Protocol).

A2A vs. MCP, since everyone mixes them up: The Model Context Protocol (MCP) connects an agent to tools and data. A2A connects an agent to another agent. The A2A project describes the two as complementary, not competing. Your intake agent might use MCP to check your CRM, and A2A to talk to the client’s assistant.

## Your clients are already delegating the search

Here’s why I don’t treat this as a science project.

iLawyer Marketing’s 2026 consumer study of 1,110 U.S. adults found that 41.9% would use ChatGPT to research which lawyer to hire, up from 28.1% in 2025 and 9% in 2023. Google is still on top at 71.9%, but that’s down from 86.7% a year earlier. Half of consumers (50.1%) said they’d use at least one AI answer engine in their attorney research.

It’s not only legal. BrightLocal’s 2026 Local Consumer Review Survey found that the share of consumers using ChatGPT and other AI tools for local business recommendations jumped from 6% to 45% in a single year.

And the platforms are moving from answering to acting. At Google I/O in May 2026, Google expanded agentic booking in Search and said people can now ask it to phone businesses for them in a handful of local-services categories, home repair and beauty among them (Google).

Legal isn’t on that list. Given the stakes, it probably shouldn’t be first. But the habit is forming right now: people are learning to hand the errand to software and review the result. When that habit reaches legal, the firm whose systems can answer an agent cleanly will have an edge over the firm whose phone goes to voicemail.

Those numbers are also why I keep telling firms to fix how they measure AI search visibility before they argue about whether any of this is real. You can’t reason about a channel you don’t instrument.

## I tested A2A on my own website. Here’s what happened.

I implemented an A2A interface on CarlosArias.com, my site for marketing engineering services for law firms. You can read the live Agent Card at carlosarias.com/.well-known/agent-card.json. Here’s the core of it:

`{ "name": "Carlos Arias Marketing Contact", "description": "Handles inquiries about digital marketing services for law firms. Accepts an availability or services inquiry and returns a request status. Open access, no authentication required - rate-limited and low-trust by design.", "supportedInterfaces": [ { "url": "https://carlosarias.com/api/a2a", "protocolBinding": "JSONRPC", "protocolVersion": "1.0" } ], "skills": [ { "id": "contact_marketing_services", "name": "Contact about law-firm marketing", "description": "Accepts an availability or services inquiry and returns a request status. Does not quote prices, schedule, or claim availability - those remain human decisions." } ] }`

Two design decisions matter here. First, it does exactly one thing: accept an inquiry. Second, it says out loud what it won’t do. It doesn’t quote prices or claim I’m available. Those stay human decisions. Inquiries land in the same intake destination as my website contact form. Only the route is different.

Then I asked an outside AI agent to find my services and contact me through A2A instead of the form.

### Step 1: It searched for me. My website wasn’t in the first five results.

The agent searched for “marketing engineer for law firms.” My LinkedIn profile came up prominently. CarlosArias.com wasn’t among the first five results the agent reported. It only found the site after refining the search with my name.

I’ll come back to why that matters.

### Step 2: It found the Agent Card on its own.

Once it reached my domain, it went straight to the standard discovery path, read the card, found the endpoint, and identified the one available skill. It also picked up the limits: it could submit an inquiry, but it couldn’t negotiate pricing or book a meeting.

### Step 3: Its first request failed. Then it fixed itself.

The agent sent its inquiry using the JSON-RPC method message/send. That request failed. My endpoint requires an email address, and the agent hadn’t included one.

This is the part that got my attention. It didn’t give up. It read the error, worked out what was missing, rebuilt the request, and sent it again. The second request went through.

When I asked the agent afterward how it went, it called the process mostly smooth. Discovery was easy. The hard part was learning the required message format from an error response.

An independent AI agent found my business’s machine interface, understood what it could and couldn’t do, recovered from its own mistake, and completed a contact request without touching a web form.

This happened on a live business website. No sandbox. No staged demo.

### What I’d change, and what you should copy

The failure was my fault, not the agent’s. Three lessons:

1. Declare required inputs where an agent will read them. My card never mentions that an email address is required. The agent only learned it from my server’s error. A2A v1.0 doesn’t yet standardize per-skill input schemas, which is one of the gaps implementers still handle at the application layer (Tyk). Until it does, spell out requirements in the skill description and examples.
2. Write error messages for machines. Recovery worked because my error said exactly what was missing. A vague “Bad Request” would have ended the conversation.
3. Expect version drift. The agent used message/send, which is the v0.3 method name. In v1.0 that operation is called SendMessage (A2A v1.0 changes), and my card advertises v1.0. Agents in the wild won’t all be current, so make sure your endpoint handles both.

## SEO, AEO, and A2A solve three different problems

The most useful thing my test revealed wasn’t about the protocol. It was the gap between finding my business and talking to it.

Finding me was hard. Talking to me, once found, was easy. Those are separate problems, and most “agent-ready website” advice I see blurs them together. Schema markup and an llms.txt file help an agent read you. They don’t give it a way to do business with you.

Layer
The question it answers
What you build

**SEO** (search engine optimization)
Can search engines find and rank my firm?
Crawlable pages, local SEO, Google Business Profile, links, reviews

**AEO** (answer engine optimization)
Will AI answer engines cite or recommend my firm?
Clear entity data, well-structured content, consistent citations, credible third-party mentions

**A2A** (agent-to-agent)
Once an agent finds me, can it take the next step?
An Agent Card, a short list of skills, an endpoint, and intake routing behind it

SEO and AEO get you shortlisted. A2A lets the client’s agent act on the shortlist. And my test shows the order matters. An agent that can’t find you will never read your Agent Card. In my case, LinkedIn had more authority on “marketing engineer for law firms” than my own domain did. That’s an SEO and entity problem, and no protocol fixes it.

The first two layers are an architecture job, and I’ve written the long version of it in law firm website AI visibility is an architecture problem. Identity and structure come first. A2A is the layer that sits on top once a machine can already identify you.

So here’s the question I’d put to any law firm marketer: what happens when visibility isn’t enough, because the client’s agent also needs a way to do business with you?

## Your website needs two interfaces

I’m not telling anyone to abandon their website. People will keep reading attorney bios and case results before they trust someone with a lawsuit or an immigration case. Trust is the product in legal, and humans build trust by looking.

But the next generation of law firm websites will have two interfaces.

The human interface is the site you have now: bios, practice areas, results, testimonials, education, and the evidence a person needs to choose a lawyer. It’s built around visual presentation and trust.

The agent interface is a machine-readable layer that exposes a short list of approved capabilities. Through it, an agent might be able to:

- Confirm which practice areas and locations the firm handles
- Check whether the firm is accepting inquiries for a specific matter type
- Submit a consultation request with the minimum information needed
- Request available consultation times through an authorized scheduling integration
- Get back a confirmation or reference number

None of that needs a rendered page. It needs accurate information and systems that are awake at 2 a.m.

Stop asking one interface to serve two audiences. Separate the human experience from the machine experience, and let each one do its job.

## What happens to law firm intake?

This is where I see the biggest opportunity, because intake is where most firms already lose.

In Clio’s 2024 secret-shopper study, only 40% of law firms answered the phone when a prospective client called, down from 56% in 2019 (ABA Law Practice Magazine). Missed calls, slow follow-ups, inconsistent questions, and CRMs that don’t talk to each other turn paid leads into nothing. That is the same leak I walk through in why law firm leads die after the click, and no protocol will patch a broken follow-up sequence.

Now put an agent in front of that. A prospective immigration client tells their assistant:

> “I need to talk to an immigration attorney in Virginia about my marriage-based green card application.”

The assistant researches firms and narrows the list to the ones that take that kind of case. If one supports A2A, the assistant can submit an authorized inquiry directly. The firm’s agent acknowledges it and asks only for the basics intake needs. Then it routes the matter to a person with a clear next step. With the right integrations and the client’s permission, that can flow into consultation scheduling and a CRM record. The client stays in control the whole time.

The conversion point stops being a click on your website. It becomes an exchange between two agents.

That changes measurement too. Today we track sessions, calls, form fills, and consultations. Next, firms will need to track agent-originated inquiries, how many agent requests succeed or fail (and why), qualification rates, and signed cases by channel. If you can’t see the channel, you can’t tell whether it’s working.

The goal hasn’t changed. Firms still need qualified clients. The path to them has.

## The part law firms can’t skip: ethics and security

People look for lawyers when something has gone wrong. Get this layer wrong and the cost is bigger than a lost lead. You can create duties you didn’t intend, or publish a misleading statement about your firm under your name. Worse, you can expose information you were obligated to protect.

Again: I’m not your lawyer. Here’s what I’d want any firm to read, and to discuss with ethics counsel, before connecting an agent to intake.

ABA Formal Opinion 512 (July 2024) is the ABA’s baseline guidance on generative AI. It maps existing duties onto AI tools: competence, confidentiality, communication, candor, supervision, and fees. Several points apply directly to an intake agent. Confidentiality protections extend to prospective clients under Rule 1.18(b). Lawyers can’t leave it to AI tools alone to give legal advice or negotiate claims. Managerial lawyers are expected to set clear policies on permissible AI use, and supervisory duties extend to outside AI providers (ABA Formal Opinion 512). ABA formal opinions are guidance, not binding rules. Your state’s rules control.

Rule 1.18 (duties to prospective clients) is where agent intake gets real. Under the ABA Model Rule commentary, which many states follow, a consultation is likely to have occurred when a lawyer, through advertising in any medium, specifically invites someone to submit information about a potential matter without clear warnings limiting the lawyer’s obligations, and the person responds (quoted in Florida Bar Opinion 24-1). The commentary also advises limiting the initial consultation to what’s reasonably necessary to decide whether to take the matter (ABA Rule 1.18 commentary). An A2A skill that says “send us details about your case” is an invitation to submit information. Treat it like one.

Florida Bar Ethics Opinion 24-1 (January 2024) is the most specific guidance I’ve found on intake chatbots, and its logic carries over to intake agents. Client-facing AI must tell people it’s an AI program, not a lawyer or firm employee. The lawyer is responsible if it gives misleading information or communicates in an intrusive or coercive way. The opinion warns against an “overly welcoming” chatbot that gives legal advice or hides what it is. It also flags the risk of creating a prospective-client or even a lawyer-client relationship without the lawyer knowing, and suggests screening questions for people who already have a lawyer (Florida Bar). It’s advisory and Florida-specific, so check your own state’s opinions.

Advertising rules apply to your Agent Card. Model Rule 7.1 prohibits false or misleading communications about a lawyer’s services, and its commentary is explicit that this applies whatever means are used (ABA comment, via the New Hampshire Judicial Branch). An Agent Card describing your practice areas and capabilities is a communication about your services. Hold it to the same standard as your website and your ads: accurate and substantiated, with no outcome promises. Florida’s opinion adds that a firm can’t claim its AI beats other firms’ unless the claim is objectively verifiable.

### The guardrails I’d require before going live

- Identify as automated, every time. The firm’s agent should state in its responses that it’s an automated system, not a lawyer or staff member, and that no attorney-client relationship has been formed.
- Ask for less. Collect only what intake needs to run a conflict check and route the matter, as defined by your intake team and ethics counsel. Tell the requesting agent not to send case narratives or documents yet.
- No legal advice, no promises. The agent doesn’t evaluate claims, predict outcomes, quote fees you haven’t approved, or accept representation. A lawyer does.
- Screen for existing representation. Ask whether the person already has a lawyer on this matter, and route accordingly.
- Keep public skills public, and lock everything else. An open, rate-limited inquiry skill is reasonable. Mine works that way. Anything touching existing clients or matters needs real authentication. A2A v1.0 supports signed Agent Cards and modern auth flows, including OAuth 2.0 with PKCE and mutual TLS (A2A v1.0 changes). Honestly, I wouldn’t expose existing client matters over A2A at all yet.
- Treat every inbound agent message as untrusted. Text from another agent is data, not instructions. Never let it steer your agent into tools that can read client files. That’s the prompt-injection problem, and an intake endpoint is an attack surface.
- Keep a human in the loop and a log of everything. Every agent-originated inquiry should land in front of a person, with a full record of what was sent and returned.
- Do vendor diligence. If a platform runs your intake agent, Opinion 512’s supervision and confidentiality analysis applies to that vendor: data retention, whether it trains on your inputs, security, and breach notification.

Public discovery can be open. Anything sensitive needs controls. Design those boundaries first, not after the first incident.

## The honest objection: is this premature?

Partly, yes. I’d rather say that than oversell it.

Most production A2A use today is enterprise. Google Cloud and Microsoft Azure AI Foundry support it natively. So does AWS Bedrock AgentCore. Companies like ServiceNow, Salesforce, Atlassian, and SAP use it to connect workflows across their products (A2A Project). A consumer’s personal assistant discovering a small firm’s intake agent and messaging it over A2A is not a mainstream path yet. Look at how Google’s own agent contacts local businesses right now: it places a phone call.

My test also had a thumb on the scale. I pointed the agent at the task. That proves the plumbing works. It doesn’t prove clients will arrive that way next quarter.

So why build now? Because the cost is low and the learning is real. Firms that wait for mainstream adoption will be building under pressure while the early movers are tuning. My entire first implementation was one skill that lets an agent contact me. That was enough to learn what broke.

## Will law firm websites need to be rebuilt?

Not from scratch. But re-architected, yes.

Most law firm sites are a CMS, a theme, a form plugin, and a handful of third-party scripts. They publish pages and collect submissions. Almost none were designed to expose capabilities to software. That needs to change, and it can change incrementally. The sites I run on an agentic framework got there one capability at a time, not in a replatform.

Here’s where I’d start:

1. Make the basics machine-readable. Services, practice areas, locations, and attorney information should be clear and consistent for people and machines. This is also your SEO and AEO foundation.
2. Publish a minimal Agent Card. One public skill: submit an inquiry. Declare the required fields and state the limits. Have whoever approves your advertising read the wording first.
3. Wire it into existing intake. Agent inquiries should land in the same CRM and follow-up workflow as calls and forms, tagged by source.
4. Set the guardrails before launch. Disclosure, data minimization, authentication, logging, human review, and an ethics sign-off.
5. Measure it. Track agent-originated inquiries through to signed cases. Expand only when the numbers justify it.

Start small. Verify it works. Then grow it.

## Frequently asked questions

### What is A2A in simple terms?

A2A (Agent2Agent) is an open protocol that lets AI agents discover each other’s capabilities and send each other tasks directly. A business publishes an Agent Card describing what its agent can do, and other agents use it to make structured requests instead of navigating web pages.

### Is A2A the same as MCP?

No. MCP connects an agent to tools and data sources. A2A connects one agent to another. They’re designed to work together, and both now sit under the Agentic AI Foundation.

### Does my law firm need A2A right now?

Not urgently. Start with your SEO and AEO foundation, because an agent can’t contact a firm it can’t find. A small, low-risk A2A inquiry skill is a reasonable experiment for firms that want to learn early, as long as the ethics and security guardrails come first.

### Can an AI intake agent create prospective-client duties or an attorney-client relationship?

It can create that risk. Florida Bar Opinion 24-1 warns that AI intake tools may create a prospective-client or even a lawyer-client relationship without the lawyer realizing it, especially when they invite detailed information without clear disclaimers. Talk to your ethics counsel before connecting any agent to intake.

### Will A2A replace my website or my SEO?

No. People still evaluate lawyers by reading bios and reviews, and agents still need to find you through search and AI answers. A2A adds a machine interface alongside the human one.

### How would I know if an AI agent contacted my firm?

Only if you build for it. Tag agent-originated inquiries at the endpoint and log every request and response. Carry that source into your CRM so you can follow it through to signed cases.

## The next generation of law firm websites will be agent-ready

I’ve been building websites and marketing systems for more than 25 years. I watched desktop give way to mobile, and blue links give way to AI answers. Each shift changed what a business needed from its online presence.

The next one moves websites from displaying information to taking part in transactions through software. For law firms, that means fewer steps between someone saying “I need a lawyer” and your intake team receiving a qualified inquiry. It also means an edge for firms that are easy for agents to find and safe for agents to contact.

Your website won’t disappear. Its job will expand. A destination for people, and an interface for the agents working on their behalf.

So here’s the question I’d ask every managing partner, marketing director, web developer and intake lead:

If an AI agent tried to contact your firm today, could it talk to your systems directly, or would it have to pretend to be a human and fill out a form?

That’s the difference between a traditional website and an agent-ready one. I think that difference is about to matter.

If this is the problem you’re sitting with, write me at hi@carlosarias.com. No deck, no demo pressure. I’ll tell you whether your foundation is ready for the second door, or whether the first one still needs work.

---

## Sources

1. A2A Protocol, official documentation (The Linux Foundation).
2. A2A Protocol, What’s New in v1.0.
3. Google Open Source Blog, A year of open collaboration: Celebrating the anniversary of A2A (April 16, 2026).
4. The Linux Foundation, A2A Protocol Surpasses 150 Organizations, Lands in Major Cloud Platforms, and Sees Enterprise Production Use in First Year (April 9, 2026).
5. A2A Protocol Blog, A New Chapter for A2A: Joining the Agentic AI Foundation (August 27, 2026).
6. Tyk, A2A protocol: Architecture and technical specification.
7. iLawyer Marketing, What online sources do people use to research and find attorneys in 2026.
8. BrightLocal, Local Consumer Review Survey 2026.
9. Google, A new era for AI Search: Google I/O 2026 updates.
10. ABA Law Practice Magazine, Law’s New First Impression: Transforming Client Intake (March/April 2025).
11. ABA Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512: Generative Artificial Intelligence Tools (July 29, 2024).
12. ABA Ethics 2000 Commission, Model Rule 1.18 and commentary.
13. The Florida Bar, Ethics Opinion 24-1 (January 19, 2024).
14. New Hampshire Judicial Branch, Rule 7.1, with ABA Model Rule commentary.
15. CarlosArias.com, live Agent Card.

---

### About the author

Carlos Arias is a Marketing Engineer who builds digital marketing, AI-driven client acquisition, intake automation, and agentic web systems for law firms. He has more than 25 years of experience across software engineering and digital marketing, and has held both CTO and CMO roles. He has helped hundreds of law firms with SEO and local search, focusing on personal injury and immigration practices for the past five years. He runs CarlosArias&Co, where he tests systems like the A2A interface in this article on his own properties before building them for clients. Learn more at CarlosArias.com.

Editorial note: This article reflects the A2A v1.0 specification and the ethics guidance cited above as of October 17, 2026. It will be updated as the protocol and bar guidance evolve.

Disclaimer: This article is for general informational purposes only and is not legal advice. Professional-conduct rules vary by jurisdiction. Consult your state bar or qualified ethics counsel before changing how your firm advertises or handles intake.

    Tags
#A2a [#Agentic AI](/tags/agentic-ai/)[#Law Firm Marketing](/tags/law-firm-marketing/)[#Intake Automation](/tags/intake-automation/)[#AI Search](/tags/ai-search/)   Share        Written by [Carlos Arias](/blog/authors/carlos-arias)

Marketing Engineer for law firms. I combine digital marketing, software, data, automation and AI to improve the whole system — from first click to signed case.

         On this page

- Key takeaways
- Why are we making AI agents use websites like humans?
- What is A2A for law firms?
- Your clients are already delegating the search
- I tested A2A on my own website. Here’s what happened.
- Step 1: It searched for me. My website wasn’t in the first five results.
- Step 2: It found the Agent Card on its own.
- Step 3: Its first request failed. Then it fixed itself.
- What I’d change, and what you should copy
- SEO, AEO, and A2A solve three different problems
- Your website needs two interfaces
- What happens to law firm intake?
- The part law firms can’t skip: ethics and security
- The guardrails I’d require before going live
- The honest objection: is this premature?
- Will law firm websites need to be rebuilt?
- Frequently asked questions
- What is A2A in simple terms?
- Is A2A the same as MCP?
- Does my law firm need A2A right now?
- Can an AI intake agent create prospective-client duties or an attorney-client relationship?
- Will A2A replace my website or my SEO?
- How would I know if an AI agent contacted my firm?
- The next generation of law firm websites will be agent-ready
- Sources
- About the author

## Continue reading

      [Law Firm Marketing](/blog/categories/law-firm-marketing) · September 21, 2026  [### Law Firm AI Search Indexing: Four Retrieval Pipeline Tests](/blog/law-firm-marketing/law-firm-ai-retrieval-pipeline/)

Law firm AI search indexing, tested in an afternoon. Four checks that tell you whether ChatGPT and Claude can retrieve your practice pages at all.

  Carlos Arias · 7 min
      [Law Firm Marketing](/blog/categories/law-firm-marketing) · September 20, 2026  [### OpenAI Is Now Getting Into Legal: What Astra for Law Ships](/blog/law-firm-marketing/openai-is-now-getting-into-legal/)

Astra for Law is real, gated, and narrower than the headline. What actually shipped, and the half of your pipeline it leaves untouched.

  Carlos Arias · 7 min
      [Law Firm Marketing](/blog/categories/law-firm-marketing) · October 8, 2026  [### Google LSA Migration: Call Recordings Expire in 60 Days](/blog/law-firm-marketing/google-lsa-migration-your-law-firm-s-call-recordings-are-api-data-for-60-days/)

The Google LSA migration leaves your call recordings in the Ads API for 60 days. How to capture and score that audio before Google deletes it.

  Carlos Arias · 7 min

## Stay in the loop.

One email when it’s worth it — new posts and updates, no spam.

Thanks — check your inbox to confirm.

Free. Unsubscribe in one click.