AI visibility for tax advisors: being named for the client's situation

What counts as being named, for a tax firm?

An answer can do three things with a firm's name. It can include the name anywhere, even in passing or with a warning attached. It can offer the firm as a fit for the client's request, which is an AI recommendation proper. Or it can give the firm first, ahead of every other name. Our methodology builds its visibility score from the first of those alone, the share of the twelve questions in which the firm was named, and reports first place beside it rather than weighting it in.

A tax firm needs a fourth reading: what the answer says the firm is for. A recommendation "for small business bookkeeping" does nothing for the client who asked about a crypto year. An answer that gives a partner the wrong designation, or a fee basis the firm dropped two years ago, puts a wrong fact in front of someone who may check it. So each answer gets read twice, once for whether you are in it and once for what it says about you.

Why a tax client's question carries the situation

Typed to an AI assistant, a request for help can hold everything a first phone call would: the first year of a business, a rental bought last spring, stock sold at a loss, a move across a border, a pile of letters from the tax authority. The situation is where firms differ, so it is where a measurement has to look. A panel that asks only for "the best accountant in [town]" tests the one question on which every firm looks alike.

Which situations clients raise, and the facts on a firm's site that answer each, are set out in what clients ask an assistant before choosing a tax advisor. This article is about finding out whether those questions name you. The method underneath, the same for every trade, is in how to measure AI visibility, and this page does not repeat it.

A panel built from situations and places

Start the panel from a grid: the situations the firm wants clients for down one side, and across the other the ways a client names a place, whether a town, a neighborhood, "near me", "online" or a country they are leaving. Each question fills one cell, typed the way a client types, and a few more kinds of question cover what the grid misses. For a paying client we run twelve; six will do for a first pass of your own.

Leave out questions that ask for a tax answer rather than a firm. "How is crypto taxed" and "can I deduct my home office" ask for information, and an answer to either tells you nothing about who a client would be sent to. Keep the questions in a dated file, one line each, and never reword a line that already has history.

The same question in March and in August

A panel is read month against month, and for a tax firm the season gets in the way. The client who asks in March wants someone who can still take this year's return. The client who asks in August may be starting a business and want the books kept from the first month. Neither question is the better one, and the panel should hold both.

What the panel should not do is change with the season. Reword a question in March and it loses its February history; swap the season questions out in May and June has nothing to compare against. Put the season questions in from the first month, as lines of their own with their own start dates, keep them all year, and note the month beside every reading.

Then read the season questions twice: against last month, and against the same month a year before. Whether the firm is taking clients is a fact that changes, and answers can trail the page. A capacity line you took down in March that still turns up in answers in June is a finding in itself, because some page or listing is still carrying it.

What to write down about a tax firm in each answer

Every answer gets the five fields listed in setting up prompt monitoring: named, recommended, first, who else, and the answer itself, word for word. A tax firm adds four, each one a fact the answer attaches to you:

• The situation it named you for, in the answer's own words. If it says "for small businesses" and the client asked about a move abroad, the gap between the two is the finding.
• The designation it gave you or your people, checked against what each person holds and against the issuing body's register.
• The fee basis it stated, if it stated one, checked against your fees page.
• Whether it said you are taking new clients, and whether that was true on the day of the reading.

Mark each stated fact right, wrong or out of date. A wrong one is traced to where it came from, one of your own pages or somebody's listing of you, and corrected there; the assistant is not where the fix goes. When AI gets facts about your business wrong covers the tracing.

Reading the first month

The first month is a starting point for comparison and nothing more. For scale: as part of our audit of 9,483 small business websites (40 trades, 90 metros in the US, Canada, Europe and the UAE, late August to early September 2026), we asked Claude one buyer question per business, with web retrieval on, for 9,240 of them, each question taken from that business's own trade and market. 14.9% of the businesses were named. Where the answer was a clearly differentiated list and not a hedge, which was true of 5,928 answers, the rate was 17.7%.

Both are Claude's figures, taken on one day with a single question per business and every trade pooled; the audit publishes no naming rate for tax advisors. Read them as direction. If a first panel names the firm on two questions of twelve, or never, the probe gives no reason to call that unusual. What a baseline is for is the comparison the second reading makes possible.

What the reading leaves out

A panel counts answers, not engagements. The client who heard your name from an assistant and then rang the office appears nowhere in it; your intake form can ask where they heard of the firm, in their own words, and that is where a count of engagements has to come from. The quality of the returns is outside it too. A rising score says the firm's pages became easier to use and nothing about the work, which is why it stays out of anything you tell clients.

It does not see the client's own conversation either, where the town, the income and the deadline may all have come up before the question the panel asks. And the reading changes nothing on its own. What an answer says is decided by the company that runs the assistant; a firm's part is the material answers are drawn from, starting with the facts on the tax advisors overview.