AI agents for accounting firms

An accounting firm's website gets two very different visitors in the same week: a business owner comparing firms a month before a deadline, and an existing client looking for the upload link. Both ask questions your team already answers by phone and email every day. This page covers where AI agents help a CPA, tax or bookkeeping practice, and where they have to stop.

Who visits an accounting firm's website

• Business owners choosing a firm. They want to know whether you handle their entity type (S corporation, partnership, nonprofit), their industry and their state, what a typical engagement costs, and whether you are taking new clients before the next deadline.
• Individuals with a tax problem. Unfiled returns, a letter from the IRS, a move between states. They are often anxious and want to know two things: can you help, and how soon.
• Existing clients. They look for the portal, the organizer, your busy-season hours or the status of an extension.
• Referral partners and candidates. Attorneys, financial advisors and bankers checking who to send clients to, and accountants reading your careers page.

Most arrive with one specific question. If the answer takes three clicks and a contact form, many leave.

Which ASDK agents fit

The three agents below share one knowledge base built from your website, so a prospect gets the same answer whichever one they talk to.

Chat: answer service and process questions (early access)

Chat answers visitors from your own pages (services, industries, published fees, onboarding steps, portal instructions) and links the page each answer came from. When your content does not cover a question, it says so and offers to take contact details instead of guessing.

Example flow

1. A visitor asks: "Do you do returns for multi-member LLCs in Ohio, and roughly what does it cost?"
2. Chat answers from your business tax page: you prepare partnership returns for clients in Ohio and nearby states, starting at the fee you publish, and links that page.
3. The visitor asks whether you can also fix last year's return, prepared by someone else. That needs a practitioner, so Chat says so and offers a 15-minute call or a callback.

Sales: qualify prospects and book consultations (early access)

Sales asks the questions your intake person asks (entity type, rough revenue, services needed, current software, deadline) as a short conversation rather than a long form. You write the qualification rules in plain language. A good fit gets a consultation booked through Cal.com, Calendly or Google Calendar, and the lead goes to your team with a summary, by email or Slack, and to your CRM through webhooks and Zapier or Make.

Example flow

1. A visitor on your outsourced bookkeeping page asks whether you work with restaurants.
2. Sales confirms from your industries page, then asks three things: how many locations, which point-of-sale and accounting software they use, and whether they need payroll too.
3. The answers match your ideal client, so Sales offers two slots with the partner who leads restaurant clients and books one. The partner reads the summary before the call.

When the visitor is not a fit, for example a one-off personal return in a season when you are closed to new individual clients, Sales says so politely and points to what you do offer.

Agent-Ready: be easy for AI assistants to read (early access)

Agent-Ready hosts an llms.txt file that summarizes your firm for AI systems, shows which AI agents visit your site (on Pro and above), and offers simple actions such as "ask a question" or "book a consultation" to browser agents through WebMCP. WebMCP is still a draft from a W3C community group as of October 2026, not a finished standard. The section on AI research below explains why this matters for a firm.

Guardrails for accounting firms

An agent on your website speaks for your firm. In 2024 a British Columbia tribunal held Air Canada responsible for wrong information its website chatbot gave a customer, and called the idea that the chatbot was responsible for its own words "a remarkable submission" (Moffatt v. Air Canada). Configure the agent as if every answer carried your letterhead:

• General information, never individual advice. The agent explains what you do and how engagements work. It does not tell a visitor what to deduct, how to structure their business or what they owe. Circular 230 sets standards for written tax advice, including advice given "by means of electronic communication" (31 CFR 10.37), one more reason to leave specific answers to a practitioner.
• A clear disclaimer in the greeting. Answers are general information, not tax, accounting or legal advice, and no client relationship exists until an engagement letter is signed. ASDK's widget also tells every visitor they are talking to an AI.
• No sensitive data in chat. Instruct the agent never to ask for Social Security numbers, bank details or tax documents, and to send clients to your secure portal. The FTC names tax preparation firms among the businesses covered by its Safeguards Rule, which requires a written information security program (FTC), and the IRS reminds tax professionals that federal law requires a written information security plan (IRS, IR-2026-92). Keep the chat out of the path of client files.
• Hand off urgent matters early. IRS letters, audits, back taxes, anything due in the next few days: the agent's job is to capture the details and get a person on it, not to reassure.
• Fees only as published. If your site says business returns start at a certain price, the agent can repeat it. It never estimates a fee you have not written down.
• Deadlines from a page you keep current. Visitors ask about filing and extension dates every season. Keep one dated page with the dates you want quoted, and the agent answers from it.

How AI assistants research accounting firms

A business owner can now start the search with an assistant: "find a CPA firm near Denver that handles multi-state payroll for a 40-person company and can start this month." Assistants such as ChatGPT, Gemini and Claude search the web, read a handful of firm websites and summarize what they find. Browser agents can go a step further and fill in your contact form or book a call. Either way, they can only repeat what your site states plainly; see how AI assistants decide what to say about your company.

Make these easy to find, as text on ordinary pages:

• Services and entity types you handle, and the ones you do not.
• Industries you specialize in, with a sentence on what you know about each.
• Service area: the states where your CPAs are licensed, and whether you work remotely with clients elsewhere.
• Fee ranges or starting prices, and how you bill (fixed fee, monthly retainer, hourly).
• Capacity: whether you are taking new clients, and your cutoff date for the season.
• Credentials: CPAs and enrolled agents on staff, and the software certifications you hold.
• The next step: how to book a consultation and what to bring.

Mark up the firm and each office with structured data, and put the states you serve and your new-client cutoff in llms.txt as well as on the page, so an assistant does not recommend you to someone you cannot take on. The agent-ready website checklist walks through each step, and the free agent readiness checker shows what an AI agent sees on your site today. Agent-Ready keeps llms.txt and the agent actions in step with your pages.

Where to start

Start with Chat on your services and fees pages, with the guardrails above written into its instructions, and turn on Agent-Ready. Read a week of conversations: the questions your site cannot answer yet are your next pages. Add Sales once those questions start turning into consultation requests. ASDK is in early access, and pricing shows which plan includes each agent.