AI assistants are overlooking many accounting firms, including the busiest ones

7 Oct

Written By Faith Ngwenya, Head of Education, Chartered Institute for Business Accountants (CIBA)

This article will count 0.25 units (15 minutes) of unverifiable CPD. Remember to log these units under your membership profile.

Business owners are starting to ask ChatGPT or Gemini to recommend an accountant. Two new studies show that most firms never appear in those answers, and that the firms doing the most work are not always the ones named.

The research, reported by Accounting Today on 7 October, looked at accounting firms in the United States. The lessons travel well, though, because the AI tools South African clients use are the same ones.

A short list gets most of the mentions

Marketing advisory firm Expert Press put 100 typical buyer questions, such as where to find tax, audit or advisory help, to seven AI assistants including ChatGPT, Claude, Gemini and Perplexity. Each question was asked five times, giving 3,480 answers.

Just 25 firms accounted for 87.2% of all the recommendations, and the top 10 alone took 61.3%. Most of the remaining large firms barely featured: 79 of the US top 100 shared about 3% of mentions between them, and eight were never named at all.

Size helped, but it did not decide everything. Several smaller firms appeared far more often than their revenue would suggest. What they had in common was a clear link to a particular industry, service, town or type of problem.

The busiest auditors were the least visible

A second study, by AI marketing adviser AIWorthy, tested something more specific. It took 10 firms in one region of North Carolina that perform audits of organisations receiving federal funding, work that is filed publicly with the auditor named on each report.

Researchers asked five AI models buyer questions about that work, producing 250 answers in August and another 250 in September. The four busiest firms, which carried out 45 of the 75 audits on record, were recommended 24 times in August. The six least busy firms, with 30 audits between them, were recommended 184 times.

The firm with more of these audit clients than any other in the group was not named once. Being good at the work, and doing plenty of it, did not get a firm noticed.

What the AI tools actually read

The firm's own website was the source the assistants cited most. But a large website was not enough. The busiest firm in the study had 1,821 pages and none of them mentioned where it operates.

The websites that were cited most tended to share three things. They named their region, they described the kind of work the client needed in detail, and they had been updated recently.

The assistants also drew on government records, documents published by clients, industry associations, business directories and LinkedIn company pages. They rarely used Reddit, Wikipedia or ordinary news coverage, unless it was a ranking or list. One firm was the second most recommended in the study without having a website at all: the AI tools found its name in old audit reports, state treasury files and a town council's minutes.

What this means for a South African practice

Most CIBA members run small and medium practices that win work through referrals and local reputation. That still matters, but a growing share of first enquiries will start with a question typed into an AI tool. The studies point to a few low-cost steps.

■ Say where you work. Name your town, suburb and province on your website, not just your street address in the footer.

■ Describe the clients you serve and the problems you solve for them, such as farming businesses with diesel refund claims or owner-managed companies with SARS debt.

■ Keep the site current. A services page last touched in 2021 tells an AI tool, and a client, very little.

■ Set up and maintain a LinkedIn company page, and make sure your details are correct in the business and professional directories you appear in.

■ Check for yourself. Ask two or three AI assistants who they would recommend for your services in your area, and see what comes back.

The researchers caution that results vary between AI models and over time, and some findings are hard to explain. Even so, the direction is clear. A practice that describes plainly what it does, and where, gives these tools something to work with.