AI Visibility for Accountants: What You Can Measure Without §7216 Risk

A client upload can create §7216 risk. Reading what ChatGPT says about your firm discloses nothing at all

A CPA firm can measure how often ChatGPT names it without sending the model a single byte of client data, because public AI answers contain no tax return information. Start with what the statute actually says. IRC §7216 is the federal criminal provision governing what a return preparer may do with tax return information: a knowing or reckless disclosure or use of that information, other than as authorized, carries a fine of up to $1,000 and up to a year in prison for each violation (IRS Section 7216 Information Center). "Tax return information" is broad enough to cover the plain fact that a person is your client.

Note the shape of the rule, because it is not a flat ban. §7216 turns on *unauthorized* disclosure or use. The regulations under §301.7216 contemplate valid client consent, and they carve out several exceptions and permitted uses. So the accurate framing is a risk framing: uploading client tax return information to a general-purpose AI tool can create §7216 disclosure risk unless a valid consent or an applicable exception covers it. Whether one does is a facts-and-circumstances question for your own counsel, your professional liability carrier, and your engagement letters, and nothing in this article is legal advice on it.

The reason the risk is treated as live rather than theoretical is that the IRS has issued no guidance addressing general-purpose AI tools, and practitioner commentators have filled the silence with warnings. Writers at Tom Talks Taxes and Compass Tax Educators argue that pasting client data into ChatGPT, Claude or a similar model should be analyzed as a disclosure under §7216, and they are skeptical that the auxiliary-service exception, drafted for a very different kind of vendor relationship, is a comfortable fit (Tom Talks Taxes, Compass Tax Educators). That is their reading, not an IRS position and not ours. What is not in dispute is the practical consequence: prudent firms are keeping client data out of consumer AI tools until the picture is clearer, which makes accounting one of the few local trades where declining to paste is a defensible compliance posture rather than mere caution.

Now notice how little of that touches the thing this article is about. Typing "which CPA firm in Sacramento handles restaurant clients?" into ChatGPT and writing down the names it returns involves no client, no return, and no return information. It is a question about your firm's public reputation, asked of a public tool. RecoSignal reads public AI answers about your firm and never touches client files, which is why the measurement described below should not implicate §7216 at all, because it sends no client or return information anywhere.

57% of businesses found their accountant through a referral and 3% through an ad. What happens when the referral is a chatbot?

Referrals still bring accountants most of their clients, but referral volume cannot be planned, and a growing share of first questions now goes to an assistant instead of a colleague. In an August 2025 survey of 350 US businesses with $1M to $100M in revenue, TaxDome found 57% located their accountant through a referral and 3% through advertising (TaxDome). Weigh the source: TaxDome sells practice-management software to accounting firms, so a survey about how clients choose accountants is research produced by a party with something to sell to the accountants reading it. The referral-heavy picture is consistent with what most firm owners describe, but it is vendor research, not independent research.

The same survey found 98% of businesses that left a niche specialist moved to another niche specialist rather than back to a generalist, and companies above $1M in revenue are twice as likely to hire a specialist. Retention is not where accounting firms leak. Discovery is.

In a 2026 survey of 1,002 US consumers by BrightLocal, which sells local SEO and reputation tools, 45% said they use AI to get local business recommendations, with ChatGPT at 31% and Google AI Mode at 23% (BrightLocal, LCRS AI trust 2026). That study covers local business as a whole with no breakout for professional services, so read it as the size of the channel, not a figure measured for accounting.

Self-prepared e-filings grew 1.7% last season while pro-prepared grew 0.4%: the IRS numbers behind the DIY squeeze

IRS filing data shows accountants losing share to self-preparation slowly but steadily, and it shows a year of demand opening in late January and closing on April 15. For the week ending April 17, 2026, the IRS counted 72,821,000 individual returns e-filed by tax professionals, up 0.4% year over year, against 64,796,000 self-prepared e-filings, up 1.7% (IRS Filing Season Statistics). Those add to 137,617,000 e-filings, roughly 53% professional against 47% self-prepared, and the smaller number is growing faster.

Firms are absorbed by their own operations while that happens. Thomson Reuters' State of the Tax Professionals 2025 report put enterprise-level generative-AI use at 21% of firms, up from 8% in 2024, with 79% expecting a high or transformational impact and only 14% holding a comprehensive AI strategy (Thomson Reuters). Read the adoption numbers with the publisher in view: Thomson Reuters sells tax and accounting software, including AI-assisted products, so a report showing rapid AI adoption and a strategy gap describes a market it also serves.

The scheduling consequence is specific: do the visibility work in October, November and December, while engines and directories still have months to recrawl before the first W-2 lands. Fixing listings in February fixes them after the season that needed them.

Are taxpayers warming to AI or backing away from it? Adobe and Invoice Home measured 2026 and reached opposite conclusions

Two 2026 studies of American taxpayers point in opposite directions, and accountants should know both figures. Adobe reported in March 2026 that roughly 26% of American workers planned to use AI when filing their taxes, against 11% a year earlier (Adobe, via PYMNTS).

Invoice Home's U.S. Tax Filing Report 2026, covered by the AICPA's The Tax Adviser in February 2026, measured the opposite motion: 37% of taxpayers would consider trusting AI instead of hiring a professional, down from 43% in 2025, with millennials falling from 54% to 50% (The Tax Adviser). Use of AI in filing is rising while willingness to substitute AI for a human is falling. These two research groups have not reached a consensus, and this article will not manufacture one for them.

A 2026 New York Times test ran Gemini, ChatGPT, Claude and Grok through eight tax scenarios and found the chatbots' refund and liability figures off by an average of more than $2,000 (reported via PYMNTS). Meanwhile 63% of active AI users say they trust AI recommendations, though 88% check the sources behind an answer (BrightLocal, 2026). The assistant is becoming the shortlist, not the substitute.

The sources AI actually cites when someone asks for an accountant nearby, ranked by measured citation volume

No published study has measured which sources AI cites for accountants specifically, so this ranking rests on the two largest measurements of local-business citations. Everything below about accounting directories is a hypothesis, labelled as one.

The strongest available signal comes from an adjacent licensed profession. BrightLocal's July 2025 study of 20 queries across 10 verticals found niche industry directories are a regular source, with ChatGPT and Perplexity leaning heavily on Superlawyers.com and Findlaw.com for legal queries (BrightLocal; see AI visibility for law firms). Whether the accounting equivalents behave that way has never been tested.

• 1. Your own firm website. It is the only source where a fact about the firm is asserted rather than repeated: the entity types you file, the states you are licensed in, the industries you serve, your fee ranges. The engine reads that text; it does not read CPAverify's back end and it does not read your practice-management system. Two measurements agree. In BrightLocal's July 2025 test across four engines, the business's own site was the overwhelming majority of sources for every engine and vertical; in December 2024 ChatGPT used it as a source 58% of the time (BrightLocal). In the three markets we measured ourselves, dental, med spa and plastic surgery in Dallas, July 2026, 73% to 79% of the citations behind AI recommendations pointed at the businesses' own websites rather than at directories (our benchmark); accounting was not in that sample, so read it as direction, not as your number. Dallas, July 2026 — one market, one day, and a primary measurement: the raw answers are stored and the method is published, so it can be checked and rerun. Do this: put services, industries, city, credentials and fees in plain text, never in a PDF.
• 2. Yelp, by a distance. Foundation Marketing, an agency, and AirOps, which sells AI-search software, analysed 28 million small-business queries across four engines in Q4 2025: Yelp 512,680 citations, BBB 149,700, Angi 145,600, Thumbtack 56,000, HomeAdvisor 33,600, Nextdoor 10,300, with Yelp supplying 72.5% of directory citations in Google AI Mode and 62.1% in Perplexity (Foundation Marketing + AirOps). Yelp alone beats the other five combined, which come to 395,200. Do this: claim the profile, set the category to Accountants or Tax Services, and match the address to your license record and to your own site. A profile that agrees with your site is the second independent source an engine needs; one that disagrees with it subtracts confidence rather than adding any.
• 3. Google Business Profile. BrightLocal's July 2025 work found Google Business Profile is strongly preferred by Google's own models (BrightLocal). Do this: a practice with a suite number and no walk-in traffic still needs a complete profile, because Gemini reads it.
• 4. The accounting-specific directories: free, verifiable, unproven. The IRS Directory of Federal Tax Return Preparers lists credentialed PTIN holders only: CPAs, attorneys, enrolled agents and actuaries (IRS RPO). CPAverify is the only free national database of licensed CPAs and firms, fed directly by state boards across 53 jurisdictions including disciplinary records (CPAverify). Do this: complete both plus a state society listing such as CalCPA Find a CPA; treat citation by an engine as unproven.

What a CPA may not write on a website: the AICPA rule, the Texas verifiable-facts standard, and the GEO research that agrees with both

Accountants are barred from writing the claims most marketing agencies produce, and the restriction points at exactly the content generative engines reward. The AICPA Code of Professional Conduct puts it plainly in ET §1.600.001: "A member in public practice shall not seek to obtain clients by advertising or other forms of solicitation in a manner that is false, misleading, or deceptive." The interpretation also bars creating false or unjustified expectations of favorable results (AICPA Code of Professional Conduct).

State boards go further than the Code does. Texas rule 22 TAC §501.82 bars "self-laudatory statements not based on verifiable facts", testimonials not based on verifiable facts, and false comparisons with other accountants, and requires firms to keep direct-mail ads and recipient lists for 36 months (22 Tex. Admin. Code §501.82). "Best CPA in Austin", "#1 rated" and "we guarantee a bigger refund" are not bold copy for an accounting firm. They are rule violations.

Aggarwal et al., "GEO: Generative Engine Optimization" (ACM SIGKDD 2024), benchmarked nine methods and found adding statistics, adding quotations from credible sources and citing sources were the top performers, the best lifting visibility by up to 41% on the Position-Adjusted Word Count metric (arXiv). Keyword stuffing produced nothing. Factual, sourced, verifiable writing is what a board demands of a CPA and what an engine rewards, so publish your fee ranges, the entity types you file, the states you are licensed in, and the IRS publication behind each claim (AI visibility vs local SEO).

The fifteen-minute check: six questions a real tax client types, and what the $49-a-month tools sell instead

Any accountant can measure their own standing in AI answers in fifteen minutes, with no tool, no payment and nothing that engages §7216, because not one of these questions carries a client's information into the model. Open ChatGPT, Gemini and Perplexity, paste each question below with your city in the brackets, and count how many of the 18 answers name your firm. Do it in October, not in March; you already know which of those months you will actually have.

One pass is one moment in time, and an accountant of all people should treat a single observation accordingly. The same question asked twice can return different firms, answers shift by city and by phrasing, and one run tells you nothing about whether the firm across town is gaining on you between now and the extension deadline. It is a trial balance, not the ledger. Run it anyway, because it costs nothing but the quarter-hour, and a bad result in the fall is a finding you still have time to act on.

Tools selling this monitoring to accounting firms already exist. Scope charges $49 per month per business, tracks queries such as "best CPA near me" across six assistants, and claims 5,000+ businesses (Scope); CountingWorks PRO sells GEO-ready websites and AI content to accounting firms (CountingWorks PRO). Both are vendors whose market statistics carry no sources.

• "I got a CP2000 notice from the IRS. Which CPA in [your city] handles audit representation?"
• "My S-corp missed the March 15 deadline. Who in [your city] can file a late 1120-S and request penalty abatement?"
• "I sold a rental property and want a 1031 exchange. Which accountant in [your city] has done this?"
• "I'm a US citizen living abroad. Which [your city] firm handles expat returns and FBAR filings?"
• "I got a 1099-K from Stripe for my online store. Which CPA in [your city] works with e-commerce sellers?"
• "I trade crypto. Which CPA in [your city] handles staking income, gains and losses, and reconciling transactions across exchanges and wallets?"

Five questions accountants ask about AI answers, answered with sources

These are the questions CPA firm owners raise first, and each answer stands on its own.

• "Is it legal for me to ask ChatGPT what it says about my own firm?" Yes. Asking a public model which accounting firms it recommends in your city involves no client, no return and no tax return information, so there is nothing for §7216 to reach. The exposure runs the other way: it arises when client information goes into the tool, which is where consent, exceptions and your own counsel come into the picture.
• "Will listing on CPAverify or the IRS preparer directory make ChatGPT cite me?" Unknown, and anyone who says otherwise is guessing. No published measurement shows any engine citing either one. Both are free and correctable: complete them as hygiene, not as a channel.
• "Yelp is for restaurants. Why would a CPA firm be on it?" Because engines read it. BrightLocal's July 2025 study found Yelp was a source in 33% of all searches tested, and Perplexity cited Yelp in every vertical examined (BrightLocal). Accounting was not broken out separately, so treat that as the platform's overall weight (how an AI visibility audit works).
• "Should I niche down to get named more often?" TaxDome, which sells practice-management software to accounting firms, found in its August 2025 survey that 83% of clients paying $10,000 or more a year name tech-savviness as a selection factor, and specialisation commands up to a 25% premium (TaxDome). Nobody has measured citation frequency for CPA sub-niches, so "crypto CPA in Denver" is a reasoned bet, not a proven one.
• "Can I write 'best CPA in Austin' on my website?" Not if your board applies the verifiable-facts standard, and most do. Superlatives, self-awarded rankings and refund promises are what get accountants into trouble. Write the checkable things: license number, years in practice, industries served.

140,222,000 returns and 653,408 licensed CPAs: does the assistant name your firm in that market?

A CPA firm cannot see what an assistant tells prospects about it, and no filing statistic will reveal it, which is what measurement is for. The IRS received 140,222,000 returns in the season ending April 17, 2026, 0.3% fewer than the year before (IRS Filing Season Statistics), and NASBA counted 653,408 actively licensed CPAs as of August 28, 2025 (NASBA). That is a flat market with a fixed population of licensed competitors: share moves between firms rather than arriving from growth. When an assistant answers "which CPA in [city] handles this," it typically returns a handful of firm names, and whether yours is among them is a fact about your firm that you currently do not measure.

You have Google Search Console for search traffic and nothing at all for AI answers. RecoSignal's free AI Visibility Snapshot runs client-style questions across ChatGPT, Gemini and Perplexity and reports how often your firm is named, which firms are named instead, and which sources those answers cite. It sends no client data and reads only public answers. That is the whole of it: a measurement, repeated. It does not promise a ranking, a citation, or a client, and a vendor who promises you any of the three is writing the kind of claim your state board would strike out of your own website.