This week the trade press handed accounting firms a new job title. Firms, wrote Dave Maney in CPA Practice Advisor, must become “authority portfolio managers,” curating the accumulated evidence of expertise that AI systems weigh when they recommend providers. Two things about that piece. Maney runs The Expert Press, a company that sells exactly this kind of authority-building, so read the pitch knowing who benefits. And underneath the self-interest sits a shift that is real, measurable, and mostly being ignored by the firms it will hit.

Start with how clients used to find you: they typed something into Google, a list of blue links appeared, and your website’s job was to be one of them. That mechanism is breaking, not gradually, and there is unusually good data on it.

In July 2025, the Pew Research Center published a study of real browsing behavior from 900 US adults across 68,879 Google searches made in March of that year. When a search returned one of Google’s AI summaries, users clicked a traditional result link on 8 percent of visits. Without the summary, 15 percent. Clicks on the sources cited inside the AI summary happened on 1 percent of visits. And users simply ended their browsing session on 26 percent of pages with an AI summary, against 16 percent without one. At the time, about one in five searches produced a summary; that share has only grown as Google pushes AI deeper into results and standalone assistants take search traffic of their own.

Two honest limits on the data before you act on it. The study measured Google’s AI Overviews specifically, in March 2025; it did not capture the standalone assistants where a growing share of research now starts, and behavior has had a year to move further in the same direction. Neither limit is comforting.

Read those numbers as a firm owner and the conclusion is uncomfortable. The person researching “should I switch accountants” or “does my SaaS company need a 409A valuation” increasingly gets a complete answer on the results page, assembled by a model from other people’s websites, and never visits any of them. Your site is no longer the destination. It is raw material for an answer you do not control. Meanwhile, most firm marketing dashboards still report website sessions and search rankings, which is to say they carefully measure the door while the visitors switch to a window. And because client acquisition is a lagging indicator, a pipeline quietly thinning today shows up in revenue two busy seasons from now, long after the cause has been forgotten.

The old defense, paying Google for placement, does not transfer either. There is no bidding your way into being the recommendation an assistant gives when someone asks it, in a private chat, who should do their books.

So what does the answer get built from? When we mapped the search queries our own readers type, back in June, the pattern across the accounting niche was stark: the top of nearly every important query was occupied by tool vendors answering questions about their own category, and by affiliate listicles paid by the click. Ask the open web whether AI will replace your bookkeeper and the leading answers come from companies selling the AI, whose conclusion, reliably, is no, but buy our tool. Independent, verifiable expertise was the rarest ingredient. AI answers are assembled from this same corpus. A model summarizing “best accounting approach for my restaurant group” is compressing whatever the web already says, and if what the web says about your specialty was mostly written by people selling software, that is what the machine will repeat.

This is the honest core of Maney’s argument, and it deserves the steelman: in a world where models compress the written record, the written record of your expertise is your marketing. Depth beats volume. A firm with one signed, detailed, regularly updated body of work on, say, dental practice accounting in Ontario gives the machine something to cite. A firm with forty generic blog posts titled “5 Tax Tips for Small Business” gives it nothing it does not already have from ten thousand identical pages.

Where we part ways with the emerging industry around this: nobody credibly knows how to optimize for AI recommendations yet. The consultants selling “GEO” and “AEO” packages are selling confidence in a system whose ranking logic is unpublished, unstable, and different across ChatGPT, Gemini, Claude, and Perplexity. Treat anyone promising placement in AI answers the way you would treat someone promising a guaranteed audit outcome.

Strip the hype and a few durable things remain true, none of them new, all of them newly urgent.

Specificity survives. The generic queries were lost to whoever has the biggest content budget years ago, and now they are lost to the summary itself. The winnable ground is the narrow question only a real practitioner can answer well: the industry niche, the cross-border wrinkle, the entity-structure edge case. Ask yourself what question a dream client would type at 11 p.m., and whether anything on the open web with your name on it answers exactly that question.

Verifiable authorship survives. Models and the platforms that feed them are getting better at weighing who said something, not just what was said. Signed work, consistent bylines, credentials that check out, publication in places that vet their contributors: this is the boring infrastructure of being citable. It is also, not coincidentally, everything a pile of anonymous AI-generated content is not. We would know; publishing signed, sourced, human-checked work in a niche is the entire strategy of this newsletter, and the search traffic that compounds comes from exactly the pages built to answer one literal question well.

Original evidence survives. A model can paraphrase opinions endlessly, but a number, a dataset, a documented case that exists nowhere else has to be cited from somewhere, and the somewhere wins the citation. A firm that publishes its own benchmarks, its own survey of a niche, its own worked examples, is manufacturing the one input the answer engines cannot synthesize from the general sludge.

And referrals survive. The Pew data describes strangers researching in a browser. A meaningful share of accounting work has never arrived that way; it arrives because a lawyer, a banker, or another client said a name out loud. AI search raises the value of that channel by making the anonymous one noisier. The firms most exposed to the shift are the ones that quietly replaced relationship-building with content volume, because content volume is the input AI just commoditized to zero.

If you run a small or specialized firm, this shift is, strangely, in your favor. You will never out-publish the platforms, but you can be the single best-documented answer to a question only a few hundred perfect clients ever ask. Pick the two or three questions you most want to own, answer them publicly under your own name, keep the answers current, and let the generic queries go.

If you run marketing at a larger firm, audit what exists rather than what you plan to produce. Search your own practice areas in the major assistants and see what they actually say, and whom they cite. That exercise costs nothing, takes an afternoon, and is the closest thing to mystery-shopping your own visibility that the AI era offers. Expect to find that your firm’s deepest expertise has the thinnest public footprint; that inversion is the norm, and fixing it is the actual work behind Maney’s portfolio language.

If someone is selling you help with this, apply the same discipline you would apply to any vendor. Ask what, specifically, they will publish under whose name, what evidence of expertise it will contain that a competitor cannot generate, and how they will measure whether assistants actually cite it. A pitch that answers with impressions and content volume is describing the old game with new vocabulary.

And whoever you are, do not respond to this by pointing an AI at the problem and generating five hundred articles. The entire dynamic described above exists because models compress undifferentiated text into nothing. Adding more undifferentiated text is not a strategy; it is donating server costs to Google.

The search click is collapsing, the answer is eating the results page, and the corpus that answer draws from is dominated by whoever bothered to write things down with proof attached. Which lands the profession somewhere familiar. AI can do the work, including the work of answering your prospective client’s first question. It cannot sign the work, and signature, in this arena, is precisely what gets cited: the named practitioner, the verifiable credential, the documented expertise a model can point to when a human asks whom to trust. The firms that put their real knowledge on the record, under their real names, are building the only marketing asset the machines cannot generate for your competitors. Everyone else is raw material.

— Footnote

Footnote is an independent publication. It is not professional accounting, tax, or legal advice, and it is not marketing consulting either; how AI systems select sources is not fully public and changes without notice. Study figures are from the Pew Research Center report linked above; commentary on it and on industry practices reflects our reading. We have no relationship with any company named in this article, including The Expert Press. Details are current as of August 11, 2026.