Ask ChatGPT or Google’s AI Overview to recommend a CPA firm in your city and see what comes back. For most accounting firms the answer is a competitor, a directory listing, or a generic explanation of how to choose an accountant with no firm named at all. That’s the new front door, and a growing share of business owners are walking through it instead of scrolling ten blue links. Showing up there works differently than ranking on page one, though the two are related more than most people assume.
Key Takeaways
- AI assistants cite sources they can parse cleanly, so structure and clarity beat keyword density.
- Being mentioned on third-party sites matters as much as your own content, because models weight corroboration across sources.
- Specific, verifiable detail (credentials, niches, locations, pricing ranges) gets pulled into answers. Vague marketing copy doesn’t.
- Question-and-answer formatting maps directly onto how these systems retrieve and summarize.
- Track AI referral traffic separately in analytics, since it behaves nothing like organic search traffic.
What’s actually changed
Traditional search gave you a list and made the user choose. AI search gives an answer and cites a few sources underneath. The click-through economics are worse, since plenty of users never click anything, but the conversion quality is better for the firms that do get named. Someone who arrives after an AI assistant recommended you by name shows up warmer than someone who clicked the fourth result.
The mechanics underneath are retrieval plus synthesis. The model pulls candidate passages from indexed content, decides which ones are relevant and trustworthy, and writes a summary. Your job is to be the passage that’s easy to retrieve and hard to argue with. That’s less about gaming an algorithm than it used to be, which is either good news or bad news depending on whether you’ve been publishing anything worth citing.
Write in answer units
A 2,000-word page that meanders through a topic gives a retrieval system nothing clean to grab. A page organized as a series of clear questions with direct answers gives it a dozen options. Structure your content so each heading is a question a client would actually ask and the paragraph directly beneath answers it in two or three sentences before elaborating.
Some question formats that pull well for accounting firms:
- What does a CPA charge for an S-corp tax return in [your metro]?
- When should a business switch from a bookkeeper to a CPA firm?
- How much does outsourced CFO work cost per month?
- What’s the difference between a review and an audit, and which does my lender want?
- Do I need a CPA licensed in my state?
Notice how many of those involve numbers. Pricing questions are among the most common things people ask AI assistants about professional services, and firms that publish real ranges get cited while firms that say “contact us for a custom quote” get skipped entirely. You don’t have to publish a full fee schedule. A range with the variables that move it is enough.
Put the answer first
Old SEO habits taught writers to build up to the point with context and keyword-rich preamble. Retrieval systems punish that. If the first sentence under the heading is setup, the extractable passage starts with setup. Lead with the answer, then explain, then add nuance. It reads better for humans too, which is a rare case where the incentives line up.
Get specific about who you serve
When someone asks an AI assistant for “a CPA firm that handles dental practice accounting in Charlotte,” the model is looking for a firm whose content says exactly that. Generalist positioning is the enemy here. A page titled “Accounting for Dental Practices” that names the specific issues (associate compensation structures, equipment depreciation timing, practice acquisition financing) is retrievable. A homepage that says you serve “businesses of all sizes” is not.
This is the same argument for niche specialization that’s been true for years, now with sharper teeth. The firms already built around two or three verticals have a real head start. The ones that serve everyone have to decide what they want to be findable for.
Your site isn’t the only input
Models corroborate. If your site claims expertise in construction accounting and nothing else on the internet mentions it, that claim carries less weight than one backed by a trade association profile, a podcast appearance, a quote in an industry publication, and a directory listing that matches.
Practical version of that: make sure your firm name, location, and specialties are consistent everywhere. The AICPA directory, your state society listing, Google Business Profile, LinkedIn company page, Clutch or similar review platforms. Inconsistent information across those sources doesn’t just hurt local SEO. It gives a model conflicting signals, and the safe move for a model facing conflicting signals is to cite someone else.
Guest content and podcast appearances have gotten more valuable than their referral traffic would suggest. A partner interview on a construction industry podcast with a transcript published on the host’s site creates an independently indexed page that names your firm alongside the expertise. That’s exactly the corroboration these systems reward.
Technical basics that still matter
Schema markup isn’t magic, but it removes ambiguity. Organization schema with your legal name, address, and phone. Person schema on partner bios with credentials listed. FAQPage schema on your question-and-answer content. Service schema on service pages. Each one tells a parser what it’s looking at instead of making it guess.
Check whether your robots.txt blocks AI crawlers. Some firms blocked GPTBot, ClaudeBot, and similar agents reflexively a couple of years back, then forgot. If you want to appear in AI answers, those crawlers need access. Blocking them and then wondering why you’re invisible is a mistake we still find in audits.
Content in PDFs is another quiet problem. Firms love publishing tax guides as downloadable PDFs. Those get parsed inconsistently and rarely cited. Publish the same content as an HTML page and offer the PDF as a bonus.
Measuring something that resists measurement
There’s no rank tracker for AI answers in the way there is for Google positions, and the answers vary by user, phrasing, and day. So do it manually and do it on a schedule. Build a list of 20 questions a prospect might ask, run them through the major assistants monthly, and log whether your firm appears. Crude, but it’s the only honest baseline available right now.
On the analytics side, set up a segment for referrals from chatgpt.com, perplexity.ai, claude.ai, and similar sources. Volume will look small next to organic search. Watch the conversion rate instead, which typically runs several times higher, and ask new clients during intake how they found you. A surprising number now say some version of “I asked AI.”
None of this replaces the rest of your marketing. It sits on top of the same foundation: clear content, a specific market, and other people vouching for you. The difference is that the machine reading it now summarizes on your behalf, and it can only summarize what you’ve made legible.