AI visibility for management consultancies: named for which problem

What do named, recommended and first mean for a consulting firm?

They are the three verdicts a firm can record for every answer, and our measurement keeps two of them, named and first; for a consultancy each one only makes sense next to a problem. Named means the firm appears anywhere in the answer, under any spelling, even as a caution. Recommended means the answer puts the firm forward as a fit for what the operator described. First means it is the first firm the reader meets, at the top of a list or first in the prose.

Take an operator who asks for help with pricing. An answer that recommends your firm for pricing work is an AI recommendation for the problem your pricing page was written about. An answer that lists you among "firms offering strategy services" is a mention with nothing attached. Both count as named. Only the first ties the firm to the problem the operator described.

The visibility score is set out on the methodology page and is the same for every trade: a number from 0 to 100 that is the share of the twelve questions in which the firm was named, with no weighting. What differs for a consultancy is what the panel asks, and what each record keeps beyond the three verdicts.

A panel built from problem, industry and company size

An operator looking for a consultant describes the problem, the kind of company and, in many questions, its size, as the page for consulting firms sets out. The panel is built from the same three. Each row pairs one problem the firm solves with one industry and one size band from its own engagements, written the way an operator might type it, and two rows name a partner.

Choose the problems the firm has done before and wants more of, then pair each with the industries and size bands its engagements came from. Which of those problems deserve a page of their own is a separate decision, covered in the article on stating a specialty. The panel should hold the problems operators already bring to the firm as well as the ones it is building.

The partner rows are there because an operator who heard a partner speak, or was handed a name by a board member, may ask about the person before the firm. Those answers show what an assistant states about that person: what they do, and whether they still work at the firm.

Write every row the way an operator types, with the size as a number and the industry in the operator's words ("family-owned food distributor" rather than "consumer goods"). Then freeze it. Setting up prompt monitoring covers the panel file, the three runs per assistant, and why a reworded question starts a new series instead of continuing the old one.

Four columns a consultancy adds to each record

Every answer already gets the usual fields: named, recommended, first, who else was named, and the answer itself kept verbatim. A consulting firm needs four more. Each is filled in from the verbatim text by the person who read it, and together they say what the firm was named for, through whom, on what terms and on what evidence.

Read the problem column first. Being named for the wrong problem and not being named at all are different findings, and they send you to different work. The first means an answer had something to say about the firm, only not the thing the operator asked, so check what your pages and your profiles elsewhere say about each of the two problems. The second means no answer connected the firm to the question, and the page for that problem is where to start.

The person column is there because an operator hiring a consultancy is hiring the people who turn up. When an answer names a partner who left in [year], look for pages that still list them with the firm: its own people page, a conference program, an old directory entry. Correct what you control, and ask the hosts of the rest.

Why a month's reading will not match a month's new engagements

The question and the engagement can sit far apart. An operator can ask an assistant in one quarter, talk to two firms in the next, wait for the board, and sign in the quarter after that. By then, the reading for the month the contract was signed describes answers that operator never saw.

So compare the panel with itself, quarter against quarter, and keep a second record of how each engagement began. On every first call, ask how the operator found the firm and what they asked, and write the answer down with the date. After a year, lay the two records side by side: the problems the panel shows the firm being named for, and the problems the first calls came in for.

Neither record proves the other. A reading that rises for [problem A] beside more first calls about [problem A] is a direction worth acting on. It is not a causal finding, and it should not be reported as one.

What no panel sees: referrals, RFPs and the former client who calls

Plenty of consulting work never passes through an assistant. A former client moves to a new company and calls. A board member hands the owner a name. A procurement team sends an RFP to firms it already knows and scores the replies against its own criteria. None of that is a question typed into ChatGPT, Claude, Gemini or Perplexity, and none of it shows in a reading.

That is why a reading is a direction and not a pipeline forecast. The methodology page says as much about the score itself: it measures visibility, not bookings, and it cannot predict revenue. For a consultancy the panel covers one channel among several, and it says nothing about the others.

For Google's AI features there is a second instrument. Since August 31, 2026, when it reached all websites, Search Console's generative AI performance report has shown impressions of links to a site in AI Overviews and AI Mode, by page, country, date and device, with no query dimension. It can tell you which case pages and articles were linked. It cannot tell you which firm an answer named.

A low first reading is not a verdict either. In our audit's probe, one buyer question per business, put to Claude once with web retrieval on, named 14.9% of 9,240 businesses across 40 trades. That is one assistant, one question and one day, with no figure for consulting firms alone, and it is why a baseline is for comparing against rather than for grading.

Reading the result problem by problem

One score for the whole panel hides what a consultancy can act on. Split the reading by problem: rows 1 to 3 for [problem A], rows 4 to 6 for [problem B], rows 7 and 8 for [problem C]. Where one problem's rows move while the others hold, look first at that problem's page and at whatever changed on it. A problem whose rows never move is the next page to work on, starting with what the answers say about the firms named instead.

The partner rows read differently. They are not about a stranger finding the firm; they show what answers state about a named person, what that person does now and for whom. A wrong answer there sends you to the pages about that person, yours and other people's, to look for one that is out of date, and an outdated page is quicker to fix than one that was never written.

What each problem page has to state is on the page for consulting firms. How to show the work behind it when a client's name cannot appear is in the article on confidential cases.