Why AI recommends some professional services firms and skips others

Here's a finding that should bother every managing partner. In one small study of audit firms, the busiest ones got recommended the least.

Accounting Today reported on it in October. A research group called AIWorthy collected 500 AI answers to buyer questions about single-audit firms in North Carolina's Research Triangle. The four firms doing the most of that work were mentioned 24 times. The six doing the least were mentioned 184 times. It's only 10 firms in one region, and the article says so. But the pattern behind it is the most useful thing I can tell a professional services firm about AI search.

AI doesn't recommend the best firm. It recommends the firm it can describe and back up.

Those aren't the same firm. And if you're the expert in your space, that gap is probably costing you introductions you'll never hear about.

Why expertise doesn't translate into AI answers

Professional services firms sell people. Clients hire your partners' judgment, your team's experience, and the trust they've built over years. That's why referrals have always worked so well for firms. A referral carries all of that in one sentence from someone the buyer trusts.

Think about how a good referral happens at a dinner party. Someone mentions they need help with a tough audit. A friend says, "Call the firm we use. They do nonprofit audits for groups your size, right here in Raleigh." Someone else at the table nods. "We used them. They're great."

That referral works because of two things. The friend could describe the firm in one clear sentence. And someone else backed it up.

AI assistants make recommendations the same way. They can't sit across from your partners. They only know what they can read about you, and they're more confident when other sources agree. Your expertise can be real and deep and still be invisible to them.

We're a professional services firm too, so I've tested this on Marketri. I wrote up what I found, including the shortlists we made and the ones we didn't. This post is the field guide I wish I'd had for firms specifically.

Figure out which way you're missing

Not every firm goes missing for the same reason. I've come to believe that there are three, and each one needs a different fix:

1. Unreadable. AI can't get to your content, so it can't use it.
2. Unspecific. AI can read your site but can't tell what you do, for whom, or where.
3. Uncorroborated. AI can describe you, but nobody else vouches for you.

To find out which one you have, you need to look. Write 10 to 15 questions a buyer would really ask. Mix category questions ("best construction accounting firms in Ohio") with problem questions ("how do I prepare for a single audit"). Run each one in the assistants your clients use.

Run each question more than once, too. AI shortlists shift every time you ask. Rand Fishkin of SparkToro studied this with Gumshoe, a company that sells AI visibility tracking tools. They found the same list of brands showed up in fewer than 1 in 100 runs. His takeaway was that how often you appear, across many prompts and runs, is the number worth tracking. Your rank in one answer isn't.

Then look at the answers for clues. Never mentioned, even for your own firm's name? Start with the first gap. Mentioned, but described vaguely or wrongly? That's the second. Described accurately but left off the shortlists? That's the third.

Unreadable: AI can't get to your content

This is the least glamorous gap and the fastest to fix. It's also more common than you'd think at firms, because so much of the best thinking lives in places machines struggle to reach.

Start with your robots.txt file. OpenAI is clear about this in its crawler documentation. Sites that block its search crawler, called OAI-SearchBot, won't be shown in ChatGPT search answers. Blocking GPTBot is a different choice. That one only affects whether your content is used for training. It's easy for a well-meaning IT team to block every AI crawler in one line without telling marketing.

Then look at where your expertise actually sits:

• Insights trapped in PDFs. Your best article on the new lease accounting rules might live in a designed PDF. A plain web page is usually easier for an assistant to read and quote.
• Gated content. An assistant can't fill out your form to read the white paper behind it. Publish a real summary page for anything you gate.
• Bios built for looks. Partner bios that load through a script, or that say "trusted advisor" and little else, tell a machine nothing.

Google keeps its bar simple. Its guidance on AI features says there are no special requirements beyond being indexed and eligible to show with a snippet. So the floor is basic: be crawlable, be indexed, and put your thinking on real pages.

Unspecific: AI can't tell what you do

This is the gap I see most at firms, and it's the one that explains the audit study.

Read your homepage and your service pages as a stranger. Most firm sites say some version of the same thing: client-focused, full-service, deep expertise, trusted relationships. A UK index of more than 5,300 law firms, covered by Legal Futures, described the sector as looking broadly the same to AI. When every firm sounds alike, the assistant has nothing to choose you for.

Look back at the AIWorthy example. The firm websites cited most often named their region and described the buyer's specific type of work in depth. The busiest firm in the study had 1,821 pages on its site and never said where it was located.

Here's what specific looks like:

• One page per real specialty. Name the industry, the problem, the size of client, and the place. "Single audits for nonprofits receiving federal grants in the Research Triangle" beats "audit and assurance."
• Bios with bylines and proof. Google encourages accurate authorship information, such as bylines. Connect each partner to the articles they wrote, the industries they serve, and the outside profiles that confirm it.
• Your facts, in plain text. Who you serve, where you are, who leads the firm, and how engagements usually work. If you don't publish a fact, AI may fill the gap anyway. Hinge Marketing described chatbots producing fee estimates for firms that don't publish pricing, built from old directory listings and other firms' prices.
• The right schema. Schema.org has deprecated the general ProfessionalService type for local businesses. Use a specific type such as LegalService or AccountingService, or Organization and Person markup with links to your official profiles.

We publish an AI-facing company information page at Marketri for exactly this reason. It's the plainest page on our site, and that's the point.

Uncorroborated: nobody else is vouching for you

Here's where firms that do everything above still lose. AI can describe you accurately and still recommend three other firms.

The reason is the trusted nod at the dinner table. Assistants lean on outside sources when they build a shortlist. In professional services, those sources are directories, rankings, reviews, trade press, and the named people who write about your field.

The legal data makes this vivid. PR firm 5W ran 24 legal questions three times each through ChatGPT, Claude, Gemini, and Perplexity. In its 2026 report, Best Lawyers was cited in 20.4% of answers, Chambers in 17.0%, and Martindale in 15.8%. Some of the biggest firms' own websites were rarely cited at all.

The engines also differ. Yext analyzed 6.8 million AI citations and found that OpenAI's models lean heavily on third-party listings, while Gemini favors first-party websites. So a firm can look fine in one assistant and absent in another.

I saw this in my own test. The one query where Marketri got recommended first was "fractional CMO firm, Philadelphia, B2B." Most of the sources behind that answer weren't our pages. They were other firms' roundups that included us.

The concentration is real. In the Expert Press test Accounting Today covered, seven AI assistants answered 100 accounting buyer questions five times each. Across 3,480 responses, 25 firms took 87.2% of the mentions. Eight firms from the Top 100 list never came up at all.

To close this gap:

• Fix your directory and ranking profiles first. Find the ones that show up in your category's answers, then make sure every profile is accurate, consistent, and complete.
• Earn coverage where your buyers read. Trade press, association publications, and podcasts in your clients' industries carry more weight than a press release.
• Get your partners writing under their own names. A partner with a real point of view on LinkedIn becomes a source in their own right.
• Ask happy clients for reviews on the platforms that show up in your sources.

The hard part is doing this every month

By now you can probably see what's wrong at your firm. You might even fix the unreadable gap this week. The other two are harder, for three reasons.

First, measuring it properly is real work. A meaningful read means a fixed set of questions, several assistants, several runs of each, and careful logging every month. Most firms try it once, get a confusing answer, and stop.

Second, you have to trace the sources. The most useful output of any AI visibility check is the list of directories, roundups, and publications behind the shortlists you're missing from. Building that list takes time, and acting on it takes relationships.

Third, authority compounds slowly. Content can change in a week. Outside coverage takes months of steady outreach before AI notices the pattern.

What most firms need at this point is a team that runs all three together. Someone needs to keep the measurement honest while fixing the content and earning the outside proof.

Where an AI visibility program fits

That's what Marketri's AI Visibility service does. It starts with a baseline audit: the questions that matter commercially, where you and your competitors appear, and the sources shaping those answers. From there, three workstreams run together. Optimize rebuilds your priority pages around direct answers, clear specialties, and evidence. Earn builds the outside authority AI looks for, through media outreach, bylined content, and expert commentary. Track measures your visibility, citations, and the leads that follow.

In one recent client program, ChatGPT-attributed website form submissions rose from 3 in July to 8 in August 2026. We're careful about how we report numbers like that, because AI influences buyers in ways no analytics tool fully captures.

Start with the dinner-party sentence

Here's your challenge for this week. Write the one sentence a friend would use to refer your firm. Make it specific: who you serve, what problem you solve, and where.

Then check three things. Does your website say that sentence, plainly, on a real page? Do at least three outside sources say something close to it? And when you ask an AI assistant your buyer's question five times, does your name come up?

If the answer to any of those is no, you know which gap to work on first. If you'd like help closing all three, our AI Visibility team can show you where you stand today.