Who gets recommended? What analyses of AI answers reveal about the visibility of management consultancies

Why consultancies go missing from AI answers

When a managing director wants a sparring partner for a restructuring, they no longer necessarily type "management consultancy Munich" into Google. Increasingly the question goes straight to ChatGPT, Perplexity or Gemini: "Which consultancy helps a mid-sized manufacturer with succession planning?" What comes back is three or four names, and the audience for those answers is enormous. ChatGPT reported 900 million weekly users in late February 2026; Google's Gemini app passed a billion monthly users in August 2026. If your firm is not in the answer, it is not in the running.

The awkward part is that nothing tells you. There is no bounce rate, no abandoned cart, no half-filled inquiry form. The prospect never saw you and will never mention it. Pew Research, watching the real browsing of 900 US adults across nearly 69,000 Google searches, found that when an AI summary appeared people clicked through to a website in 8 percent of visits instead of 15 percent, and clicked a link inside the summary itself in 1 percent. The demand is real; the visit never arrives.

Generative Engine Optimization, GEO for short, is the attempt to measure that black box. Instead of watching Google positions, you sample AI answers and count: how often you are named, where in the answer, and for what. The two are not the same scoreboard. An Ahrefs analysis found only about 6 to 8 percent of the URLs ChatGPT cites also sit in Google's top ten for the same query, and roughly 80 percent do not rank in the top hundred at all. In consulting, where reputation closes the deal, that gap is worth measuring on its own terms.

What a serious AI-answer audit actually measures

One question proves nothing. You build a catalogue of prompts your buyers would really type: "Who advises on an SAP S/4HANA rollout in a mid-sized manufacturer?", "Which strategy firm specializes in family businesses?", "Recommend a consultancy for post-merger integration in mechanical engineering." Occasion, industry and company size, phrased the way a managing director writes rather than the way a marketer writes.

Then you run each prompt repeatedly, across several engines. The answers are not deterministic, so the same question can name you today and skip you tomorrow, and none of the vendors publishes the formula that decides which retrieved page becomes a cited one. That is why you measure frequencies instead of anecdotes: what share of runs name you, in what position, in what tone. Repetition is what turns a lucky hit into a number you can defend in a partner meeting.

What you end up with is a table, not a hunch: mention rate per prompt, per engine, next to three to five named competitors. That is the difference between "I think we come up sometimes" and "we appear in most succession runs and almost none of the digitalization ones."

Mention rate, position, context, consistency

Start with the crudest measure, the mention rate: does your name come up at all? Then position. Are you the first recommendation, or a trailing clause after McKinsey, BCG and Roland Berger? For a specialist firm the second tier is the realistic target, and second tier beats absent every single time.

Context matters as much as placement. Is the answer attaching you to the right competence? Being recommended for payroll work when you do turnaround advisory is worse than silence, because it means the public record about you is muddled, and these systems state wrong things with complete confidence. Columbia Journalism Review's Tow Center ran 1,600 source-identification queries across eight AI search tools and found more than 60 percent of the answers wrong; ChatGPT misidentified 134 of 200 articles while expressing uncertainty just 15 times and never declining to answer.

Last, consistency across engines. If you show up in Perplexity, which retrieves live pages and respects robots.txt, but never in ChatGPT when it answers from memory, that tells you where your visibility actually lives. Profound's look at roughly 730,000 US ChatGPT conversations found only about 18 percent trigger a web search at all; the rest answer out of the model's own training data.

A worked example: two boutiques, one gap

Here is an invented but unremarkable case. A fifteen-person turnaround boutique runs the prompt "consultancy for restructuring mid-sized industrial companies" twenty times and is named in six of those runs, usually third or fourth. A competitor of the same headcount lands in fourteen, often first. Same size, same market, same fee bracket.

The difference is not talent. Dig in and the visible competitor has three detailed pieces on restructuring procedure on its own site, gets quoted as an expert in trade press an editor actually vets, and publishes structured case write-ups with industries, constraints and outcomes named. The quieter firm has done work of the same quality and documented almost none of it in public.

That is the finding underneath most of these audits: AI visibility tracks the density and clarity of your public record, not the quality of your work. Nothing can recommend what it has never read.

Why consulting is a hard case for AI systems

Few industries are as deliberately quiet as consulting. Engagements sit under NDA, references go out on request, and plenty of firms treat the method itself as the asset. In a trust business that discretion is rational. To a retrieval system it reads as absence: all three major engines assemble business answers out of documents they can fetch, so a firm with no fetchable documents has no case in the room.

Then there is the sameness of the language. Open ten consultancy sites and you get ten rounds of "holistic," "tailored," "at eye level" and "sustainable value creation." To a language model those words carry no distinguishing information. None of them names a competence, an industry or an occasion, so none of them can be matched to a question.

The third obstacle is the weight of the big names. Everything ever written about strategy consulting is thick with the top houses, and you will not outpublish that. Specificity is the move you actually have, and there is at least suggestive evidence it pays off asymmetrically: the 2024 SIGKDD paper that put "GEO" on the academic map tested its tactics against a simulated answer engine and found the gains landed on lower-ranked sources, while the source already sitting at rank one lost ground. Narrow beats loud.

Three levers that actually move the number

Lever one is depth on a named question. Instead of a service page called "strategy consulting," publish the piece that answers what someone actually asks: how a post-merger integration runs inside a family business, which early indicators precede a liquidity squeeze, what the first ninety days of a carve-out look like. Those are the blocks an answer gets built out of.

Lever two is somebody else saying it. Your own article helps; a quote in a trade publication, a conference talk with a published program, an interview you did not commission helps more. This is the best-evidenced lever there is: across roughly 75,000 brands, Ahrefs found how often a brand is mentioned around the web correlates with its AI citation rate at about 0.66, roughly three times the correlation it measured for backlinks.

Lever three is structure and specifics. Real headings, named industries, named methods, actual figures. "We have led forty court-supervised restructurings of industrial suppliers with 50 to 500 employees" is usable; "many years of experience in crisis situations" is not. In the same SIGKDD experiment, adding statistics, quotations and source citations moved the authors' visibility metrics most, while keyword stuffing did nothing at all. Be precise where your competitors are vague.

The gap between your self-image and your AI image

Often the useful finding is not the score, it is the distance between how you describe yourself and how the machine describes you. You think of the firm as the region's digitalization advisor; the answers name you only for process optimization. That gap is where your public communication has drifted away from your strategy.

Equally telling is being absent from your growth segment altogether. If succession work in family businesses is the plan for the next two years and you appear in none of the succession prompts, that is not a GEO problem. It is a positioning problem that GEO happened to expose.

Treat the discrepancies as a to-do list. They name the topics you have to own before the next buyer types the question into a chat box. Read that way, the analysis is less a marketing report than an outside opinion on what the market thinks you do.

How long this takes, honestly

GEO is not a switch you flip. Models absorb new material slowly: in Ahrefs' analysis of 1.4 million real ChatGPT prompts, the pages it cited had a median age of roughly 500 days. Retrieval moves faster than memory, so a clearly structured piece can start appearing in a live-fetching engine like Perplexity within weeks rather than quarters.

Do not set the bar at beating the global brands. The realistic goal for a specialist is dependable presence on your five to ten niche prompts, which is where the engagements that actually suit you get decided anyway.

And measure on a schedule rather than once. Quarterly is enough to see movement and to show that a piece of content did something. Firms that keep the series turn a vague unease about AI into a number they manage, the way Google position used to be managed.

A 30-day plan to get your first baseline

Before you write anything, get a baseline. In week one, write down ten to fifteen questions clients really ask, such as "who helps with succession planning in a mid-sized company?" or "consultancy for restructuring in manufacturing." Put each one to several engines more than once and record whether you appear, in what context, and beside whom. That table is the zero line every later claim of progress gets measured against.

Weeks two and three are for sorting the failures. Absent entirely means there is nothing on the web to retrieve. Present but mislabeled means the words on your own pages point the wrong way. In week four, publish two or three substantial pieces aimed exactly at the prompts where you were invisible, then re-run the identical prompt list next quarter. That is how a hunch becomes a process with a before and an after.

What consultancies ask us first

"Is a good website not enough?" No. These systems pull from expert articles, interviews, directories, conference programs and third-party coverage, and Google's own guidance is explicit that no special markup, schema or llms.txt file is required to be eligible: a page has to be indexed and worth showing with a snippet. What moves the needle is how often and how consistently your name appears next to your core topics on sites you do not own.

"How often should we measure?" Monthly runs, quarterly conclusions. AI answers drift more slowly than search positions, because the underlying models and source sets are not re-scored daily. A month of data keeps you honest; a quarter of data shows the trend without every random fluctuation reading as a crisis.

"Does this matter for a three-person office?" More, if anything. A small firm usually has a sharper answer to "what are you for" than a large generalist does. Own one narrow question convincingly and you become the easy recommendation for it, ahead of a broader competitor with no distinct profile.

Where the numbers stop being useful

Useful as the table is, it is not a verdict. An AI answer is a probable phrasing, not a ranking of the best firms: two near-identical questions can return different names, and no vendor publishes what actually decides a citation. So judge patterns across many runs, never a single answer. Profound also found citations thin out as a conversation goes on, from about 12.6 percent of first exchanges to 3 percent by the twentieth, which means where in a chat someone asks changes what they get.

And visibility is not an engagement. Getting named opens a door, nothing more. The rest still runs on trust, references and a conversation with a human being. Use the measurement as an early indicator of market presence, not as a substitute for business development. It tells you where you sit in the machine's idea of relevance; the walk from there to a signed engagement is still yours.