A software product can be described. It has features, a price, a G2 page, and ten thousand reviews that all say roughly the same thing. A consulting firm, an engineering practice, or an agency cannot be described the same way. What it sells is judgment, and the only evidence AI has of that judgment is what other people have written about it.
That makes professional services the hardest industry for AI to evaluate, and the numbers show it. We analysed the most recent edition of twelve professional services categories tracked by QuadrantX, spanning management consulting, civil engineering, staffing, marketing agencies, M&A advisory, crisis communications, implementation services, and legal AI. Each category is scored independently by five production AI models. Across those twelve categories, only one produced a unanimous leader. In half of them, no firm appeared in all five models' top five.
The One Category Where Everyone Agrees
Management consulting is the exception that explains the rule. All five models named McKinsey & Company first, with Boston Consulting Group, Bain, and Deloitte close behind. Roughly three-quarters of the firms in any one model's top five appear in every other model's top five. McKinsey has held the top spot in 98% of the editions we have generated this year.
This is what a memorised category looks like. The "MBB plus Big Four" narrative has been written so many times, in so many places, that every model has absorbed it identically. There is no information left to disagree about.
Now compare that to every other category we track in the sector.
Where the Recommendations Scatter
Specialist consultancies
Ask the five models which technology and innovation consultancies lead in the Nordics and you get five different worldviews. Two named Accenture. One named McKinsey. Perplexity's Sonar Pro named Consid, a Swedish specialist, and GPT-5 named Reaktor, a Finnish one. Seventeen distinct firms filled the five top-five lists, and more than half of all firms mentioned in the category came from a single model.
The Oracle Eloqua implementation services category shows the same pattern with a twist: two models named Accenture, two named Oracle itself, and Sonar Pro named The Pedowitz Group, a marketing-automation specialist. Nearly two-thirds of the firms mentioned were mentioned by one model only.
Agencies
Among marketing agencies in New York City, four models agreed on Ogilvy as the leader, and that is where the agreement ended. Sixty-seven agencies appeared in the category. The five top-five lists shared almost nothing beyond the leader, producing sixteen distinct names, and the top spot has changed hands in roughly four editions out of ten this year, with R/GA holding it early in the year before Ogilvy took over.
Brand Gravity — The tendency of AI models, when they lack specific evidence about a specialist category, to default to the largest and most frequently described firm in an adjacent category. Accenture leading a Nordic consultancy category and an Eloqua implementation category is brand gravity at work: the model is not evaluating the specialism, it is reaching for the name it knows.
Categories the models define differently
The most revealing category was crisis communications. Two models, Claude and Gemini, treated it as a public relations discipline and named Edelman. The other three treated it as a software category and named Everbridge, a mass-notification platform. Both readings are legitimate. But a PR firm measuring its AI visibility in this category would find itself absent from three of five models, not because those models rate it poorly, but because they are answering a different question.
In professional services, the first thing AI has to decide is what the category even means. Different models decide differently, and firms live or die by that decision without ever knowing it was made.
The Stable Categories Have Something in Common
Not every professional services category is chaotic. Civil engineering, staffing, and bespoke travel all showed strong agreement: AECOM led civil engineering for four models with Jacobs and WSP behind; Adecco led staffing for four with Randstad the alternative; Abercrombie & Kent led bespoke travel for three. In each, the leader has held its position in 95% or more of this year's editions.
What these categories share is a small number of very large, globally reported firms with consistent naming and long public histories. What the chaotic categories share is the opposite: many mid-sized firms, regional scope, inconsistent naming, and thin coverage. The pattern mirrors what we found in financial services, where consumer brands earned consensus and advisory firms did not.
The Retrieval Model Is the Specialist's Friend
One model diverged from the others in six of the twelve categories, and it was the same model every time. Perplexity's Sonar Pro, the only model in our production set that searches the live web before answering, named Randstad over Adecco, Jacobs over AECOM, Consid over Accenture, Pedowitz over Oracle, Virtuoso over Abercrombie & Kent, and Ventum Financial over Canaccord Genuity in small-cap M&A advisory.
These are not random picks. They are firms with recent, specific, well-indexed content about the exact service in question. A retrieval-based model rewards that content immediately. Training-data models reward it only after it has been repeated widely enough to shape the next generation of training data, which can take a year or more.
For a specialist firm, this is the most actionable finding in the data. You cannot out-publish McKinsey across the whole web. You can out-publish McKinsey on the specific question a retrieval model is trying to answer this week.
What the Newest Category Tells Us
The youngest professional services category we track is generative AI for legal professionals, added this spring. It is also one of the most unsettled: Harvey AI led for two models, Thomson Reuters and its CoCounsel product for the other three, and the top spot has changed hands in well over half of consecutive editions. Five different firms have held it. This is what a category looks like before the narrative hardens, and it will not stay this way. In our data, categories settle over time, and once a leader is unanimous it tends to stay that way for months.
If you are a specialist firm, measure your visibility in the categories that matter to you across several AI models, and pay particular attention to which models omit you and why. If the omission is a category-definition problem, you need content that anchors your firm to the right reading of the category. If it is a brand-gravity problem, you need specific, citable content that gives models a reason to name you instead of the default. Either way, the retrieval-based models will respond first, and their divergence from the others is your earliest signal that it is working.
Reputation Was Always the Product. Now It Is Also the Data.
Professional services firms have always known that reputation is what they sell. What has changed is that reputation is now being read by machines that summarise it into a shortlist before a prospective client ever picks up the phone. In categories where the reputation story is settled, the shortlist is settled too. In the many categories where it is not, the firms that give AI a clear, consistent, specific story about what they do will be the ones it recommends.
To see how your firm's category looks across models, explore the QuadrantX category data, or read our guide to how AI engines decide which sources are authoritative.