Somewhere in your city right now, a freelancer who just crossed into their first six-figure year is typing a question into an assistant instead of calling around. "I'm self-employed in Columbus and owe way more than I expected — who should I hire before the deadline?" The answer comes back in seconds. A few firm names, maybe a line about each. The freelancer reads it, recognizes one, and calls. That call used to be a search result you competed for on a page of ten. Now it's a sentence, and either your name is in it or it isn't.
This is the part of tax season nobody put on the calendar. The rush is still the rush — the January-through-April stretch where most firms book the bulk of their new clients for the year. What changed is the front door. More of those clients now arrive having already asked ChatGPT, Gemini, or Perplexity who to trust, and they walk in treating that answer as a shortlist someone else already vetted.
The Answer Is a Short List, and That Changes Everything
Here's the thing a ranking never did to you. A search results page is forgiving. Land at number seven and you still exist; someone scrolling, comparing, opening tabs can find you. An assistant's answer isn't a page. It's a handful of names in a sentence or a short block, the kind of reply a person reads once and acts on.
Being left out of that is not a lower rank. It's closer to not existing for that question at all. The prospect never scrolls to find you, because there's nothing to scroll. They read the names given, and they move.
Run the test yourself before you trust a word of this. Open ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Grok, and Copilot, and ask each one the question your next client would ask — not your firm name, the actual situation. "Best CPA in Scottsdale for a small business." "Who handles expat taxes near San Diego." What comes back varies by platform and even by how you phrase it; one assistant might give a numbered shortlist, another a map, another a tight three-name reply with a paid listing tucked alongside. But across all of them you'll learn the one thing that matters: are you a name these systems are confident enough to say, or not? A lot of firm owners do this for the first time and discover the assistant names the practice two suites down and skips theirs entirely.
Tax Season Concentrates Demand Onto the Few Names That Get Spoken
This is the mechanism worth sitting with, because it's what turns a quiet visibility problem into an expensive one on a deadline.
Demand for a tax preparer is not spread evenly across the year. It spikes, hard, into a few weeks. A clinic with a visibility gap loses a trickle of patients month after month and might not feel it for a while. A tax firm doesn't get a trickle. It gets a flood that arrives all at once and drains almost as fast, and the firms the assistants name are the ones standing under it when it pours.
Concentration is the key word, and it stacks twice. First, client demand concentrates into the season. Second, the assistant's answer concentrates that demand further, because an AI reply doesn't distribute attention across a field of competitors the way a results page roughly does. It funnels it onto the short set of names it chose to speak. So the peak of the year flows through the narrowest possible channel, and the firms in that channel absorb a volume wildly out of proportion to how much "better" they actually are. Being named isn't a 10% edge during the rush. It's the difference between fielding the surge and watching it go somewhere else.
Think about what that means for the gap between two firms that, in September, look about even. Same credentials, same decent reviews, comparable work. One is a name the assistants hand out for "self-employed taxes in [your city]" and one isn't. For most of the year that gap is survivable, almost invisible. Then the season hits, every searching client pours through the assistant's answer, and the named firm has its busiest April on record while the unnamed one wonders where the usual rush went. The gap didn't grow slowly. It detonated on schedule, at the exact moment the stakes were highest.
Why the Assistant Trusts One Firm's Name and Not Another's
An AI model doesn't take your firm's word that you're the best choice for small-business owners in your city. It can't. Every firm's own site says that. What the model does instead is look for claims that show up the same way in more than one independent place, because agreement across sources is what separates a fact it can repeat from a boast it should ignore. When the same description of what you do — the specialty, the location, the kind of client you serve — recurs consistently across places the model didn't get from you directly, it grows confident enough to say your name. When it finds thin, contradictory, or stale information, it hedges. And a hedging model recommends someone else, because naming a firm it's unsure about is the one thing it's built to avoid.
Take ZocDoc as a worked illustration of why this works the way it does — and yes, that's a doctor-booking platform, not an accounting one, which is exactly the point. The principle isn't about which specific site; it's about what that kind of site represents to a model. ZocDoc entries carry a structured, independently maintained description of a provider: who they are, where they practice, what they handle, confirmed through a booking flow and surrounded by patient reviews that were left by different people at different times. When a model reads that a dermatologist in Portland treats a particular condition, and that same fact appears in that structured, third-party-verified form, it's reading something the provider couldn't simply assert into existence. The friction of it being maintained elsewhere, confirmed by real appointments, echoed by independent reviewers — that friction is the signal. It's hard to fake, so the model trusts it. That's the whole mechanism in miniature.
Now carry it back to a CPA. The accounting world has its own places where a firm's details get described, confirmed, and corroborated by parties other than the firm itself. The question the model is quietly asking about your practice is the same one it asks about that Portland dermatologist: does this claim hold up the same way in more than one place I didn't get from them? If the answer is yes, your name gets spoken. If your information is thin, outdated, or says different things in different places, the model does what it always does when it's unsure. It picks a firm it's sure about. Often that's the competitor across town who looks, to a model, more coherently described than you do — even if your actual work is better.
That's the quiet injustice of it. The assistant isn't rewarding the best accountant. It's rewarding the most legibly, consistently described one. Those are not the same firm, and the gap between them is where good practices lose clients they earned.
The Client You Lose Has a Face
It's tempting to file all this under "traffic" and feel nothing. Don't. The thing at risk isn't a metric.
It's the couple in Denver who just had their first kid, finally have real deductions to think about, and asked Gemini who could help — and got your competitor's name, not yours. It's the contractor who outgrew doing his own return and let Perplexity pick for him. These were your clients. The kind that stay for fifteen years, refer their family, call you before they buy a building. You built a firm that could serve them well. And a system that never met you handed them to someone else because that someone else read as a surer bet to a model.
RankCommander data makes the scale of this concrete rather than hypothetical. Across tens of thousands of real AI answers evaluated and still climbing, the same pattern holds vertical after vertical: the firms assistants name are a narrow set, and the distance between being in it and being just outside it is far wider than the quality difference between the businesses. You can see the benchmarks at /ai-visibility-index — real numbers, real disclosed sample sizes, not a vibe.
What a Scan Actually Tells You
RankCommander's scan asks the assistants the questions your clients ask, across ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Grok, and Copilot, and shows you what comes back. Whether you're named for the searches that drive your season. Whether a specific competitor is named where you aren't. Whether the things the models believe about your firm are even accurate, or quietly wrong in ways that make a model hedge and move on.
Doing this once by hand tells you something real. But the answers shift by platform, by phrasing, by week, and by whatever the model learned since you last looked — which is why a one-time glance is a snapshot and the actual picture is the thing tracked over time, across every assistant, against the competitor who's gaining on you.
You Have Until the Rush, Not After
The worst version of this is the silent one. You don't get an alert when an assistant starts handing your city's self-employed filers to the firm across town. You find out in April, from an intake calendar that's lighter than last year's for a reason no one can name, after the season you can't get back has already closed.
That competitor could already be the name the assistants give instead of you — right now, today, while you read this. You built this firm over years. It shouldn't come down to whether a model found you legible this quarter. Run the scan at RankCommander, see exactly where you stand before the first surge of clients starts asking, and stop being the firm that's good at the work and invisible at the door. The question is already being asked. Make sure the answer has your name in it.