For 20 years, business development at accounting firms followed a playbook every partner knows by heart: Referrals came first, the website existed to back them up, and Google search filled whatever was left of the funnel. That playbook is quietly breaking. A growing share of prospects now ask ChatGPT, Perplexity, Gemini or Google's AI Overviews a question like, "Who is the best CPA firm for a small business near me," and act on the three or four names the model gives them. If your firm is not one of those names, you were never in the running, and nothing in your analytics will ever tell you the opportunity existed.
The mechanics of this shift matter more than the acronyms attached to it, though the acronyms are worth knowing: Answer engine optimization, or AEO, is the practice of becoming the answer an AI assistant returns, and generative engine optimization, or GEO, is the broader discipline of showing up inside generated results and the sources they cite. What makes both different from classic search engine optimization is that the answer is a shortlist, not a results page. A page of search results gives you 10 chances to be seen. An AI answer names a handful of firms and moves on, so visibility is close to binary. Worse, the click often never happens. A prospect reads the synthesized answer, picks a name, and types that firm's name directly into a browser. Google Analytics dutifully records a direct visit or a branded search, and the AI referral vanishes from view.
That measurement gap is where I would start, because you cannot manage what you cannot see. The fix costs nothing: Ask every lead how they found you and record the answer in the prospect's own words. "I asked ChatGPT which CPA firm I should use" is a very different data point from "I Googled you," and the difference only survives if whoever answers the phone writes down the actual words rather than ticking a category box. Put the question on the intake form, train the front desk to ask it out loud, and review the log at your monthly partner meeting. Within a quarter you will know whether AI referrals are real for your firm and roughly how they compare to referrals and search.
The second discipline is to run the prompts your prospects run. Sit down for an hour each quarter with 15 or 20 questions phrased the way a business owner, practice owner or CFO would actually phrase them, and run each one through the major assistants. Record two things: Which firms get named, and which sources the answers cite. The citations are the strategy. If the models keep citing a directory you are not listed in, get listed. If they cite a Reddit thread, that community matters. If they cite a competitor's service page, you now know exactly what kind of page you need to build. There are tools emerging that automate this kind of citation tracking, but the manual audit stays valuable because it shows you the actual answers your prospects are reading.
What you will likely discover is that the sources feeding these answers are mundane and largely within your control. For any question with local intent, your Google Business Profile has arguably become more important than your homepage because assistants lean on Google's business data directly and through the web results they browse. Most firm profiles are skeletons: a name, an address, the category "accountant" and a few reviews. A model cannot recommend you for tax, audit and accounting services if your profile never mentions them, so list every service explicitly, complete every field, and ask satisfied clients to name the specific service when they leave a review. A review that says the firm made a tax-filing deadline painless teaches both Google and the models precisely what you are good at.
Your website deserves the same rethinking, because generative engines do not match firms to questions; they match pages to questions. When someone asks which CPA firm specializes in construction companies, the model looks for a page that answers exactly that, and a generic services page loses to a competitor's specific one every time. The test I would apply to every page on your site is simple: Could an AI model quote this page as a complete answer to one specific question a prospect asks? A page that passes is an asset. A page that fails is a brochure.
Timing helps too, since this profession's demand is seasonal and so are its questions. The same anxieties surface every year in the same order, from filing season through Form 5500 deadlines to year-end planning, and the engines reward whoever has already published the clearest answer by the time the questions peak. Publishing a plain-English explainer four to six weeks ahead of each season, with the question as the headline and the answer in the first two sentences, is unglamorous work that compounds.
Finally, pay attention to where the models look when the question is a recommendation, which is very often Reddit and similar communities. Threads asking how someone found their 401(k) auditor get scraped, indexed and quoted in generated answers for years. The right response is genuine presence: Pick two or three communities, answer questions with real substance, and never astroturf. Fake recommendations get detected and removed, and the reputational damage lands on your name. Being the visibly knowledgeable professional in the thread is the entire play; the models learn the association between your name and your specialty on their own.
None of this requires a marketing budget. It requires attribution discipline at the front desk, an hour of prompt auditing each quarter, an honest Google profile, pages that answer real questions, content timed to the calendar your clients already live by, and authentic participation in the places where they ask for advice. What makes the moment unusual is how empty the field still is. Most firms have done nothing, which means the partners who start now are competing against almost no one. In a channel where the answer is a shortlist of three or four names, being early is most of the advantage, and the window in which early is still possible will not stay open long.
