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

Why consultancies stay invisible in AI search

When a managing director looks for a sparring partner for restructuring today, they no longer necessarily type "management consultancy Munich" into Google. More and more often they ask ChatGPT, Perplexity or Gemini directly: "Which consultancy helps a mid-sized company with succession planning?" The AI answers with three or four specific names. If your firm isn't among them, you simply don't exist for the customer in that moment.

The tricky part: you don't notice it. There's no bounce rate, no empty shopping cart, no inquiry that never came. The prospect never saw you and will never tell you. That's exactly why many consulting firms grope in the dark while their competitors show up in the language models' answers.

Generative Engine Optimization, GEO for short, is the attempt to measure this black box. Instead of watching Google rankings, you systematically measure whether and how you get recommended in AI answers. For management consulting this is especially relevant, because trust and reputation decide the engagement here, and AI systems aggregate exactly these reputation signals.

What a real analysis of AI answers measures

A robust GEO analysis doesn't work with a single question. You define a catalog of realistic prompts, the kind your target clients would actually ask. For a consultancy those could be: "Who advises on the introduction of SAP S/4HANA in mid-sized companies?", "Which strategy consultancy specializes in family businesses?" or "Recommend me a consultancy for post-merger integration in mechanical engineering."

You pose each of these prompts repeatedly and across several models. Language models don't answer deterministically, the same question can contain your name today and not tomorrow. That's why you measure frequencies: in what percentage of runs are you mentioned? At which position? In what tone? Only this repetition turns a chance hit into a robust metric.

At the end there's no gut feeling, but a table: visibility rate per prompt, per model, compared to three to five competitors. That's the difference between "I think we do get mentioned" and "For succession questions we show up in 40 percent of answers, for digitalization only in 8 percent."

The metrics that really matter

The most important number is the mention rate: how often does your name even come up? After that follows the position within the answer. Are you named as the first recommendation, or only in a subordinate clause after the big names like McKinsey, Roland Berger or BCG? For a mid-sized specialist consultancy the second tier is often realistic, but being mentioned at all beats any invisibility.

Just as revealing is the context of the mention. Is your consultancy linked with the right competencies? It's of little use if the AI recommends you for "payroll accounting" when your core business is restructuring advice. Such misassignments show that the data situation about you on the web is unclear.

Finally, consistency across the models counts. If you appear only in Perplexity, which searches live on the web, but never in ChatGPT without web access, that tells you something about the source of your visibility: you live off current web content, but haven't yet made it into the "memory" of the training data.

A concrete example from everyday consulting

Take a fictional but typical boutique consultancy for turnaround management with 15 consultants. For the prompt "consultancy for restructuring mid-sized industrial companies" it gets named in 6 of 20 runs, mostly in third or fourth place. A direct competitor of the same size shows up in 14 of 20 runs, often in first position.

What causes the difference? On closer research it turns out: the more visible competitor has published three detailed expert articles on restructuring procedures, is cited as an expert in a Handelsblatt article and has structured case studies on its own website. The less visible consultancy has done excellent work, but has publicly documented almost none of it.

That's exactly the central insight of such analyses: AI visibility doesn't correlate with consulting quality, but with the density and clarity of publicly available traces. Those who stay silent don't get recommended by the machine, no matter how well the engagements are going.

Why consultancies have it especially hard

Hardly any industry is as discreet as management consulting. Engagements run under NDA, references are named only on request, and many firms deliberately hold back their methodology. This discretion makes sense in a trust business, but for AI systems it's a problem. They can only recommend what they know something about.

Add to that the interchangeability of the presentation. Open ten consultancy websites and you read ten times "holistic," "tailored," "at eye level" and "sustainable value creation." For a language model such platitudes are empty of content. It can't derive any specific competence from them and can't assign you to any concrete occasion.

The third hurdle is the dominance of the big names. The training data is full of articles about the top strategy houses. You don't beat this superiority with volume, but with specificity: the narrower and clearer your niche is described on the web, the more likely you become the obvious answer for exactly that niche.

How to improve your AI visibility in a targeted way

The first lever is thematic depth. Instead of a general service page "strategy consulting," you write substantial expert articles on clearly defined questions: "How does a post-merger integration in a family business proceed?" or "Which early indicators point to a looming liquidity crisis?" Such content gives the AI exactly the building blocks it uses to construct answers.

The second lever is external confirmation. An expert article on your own site is good, a quote in an industry publication, a talk at a conference with a documented program or an interview in a business media outlet is more valuable. AI systems weight mentions that don't come from you yourself more heavily, because they count as an independent signal.

The third lever is structure. Clear headings, concrete numbers, named industries, named methods. A statement like "We have supported 40 restructuring procedures under StaRUG" is a thousand times more usable for a language model than "many years of experience in crisis situations." Be precise where others stay vague.

What the analysis reveals about your positioning

Often the biggest insight isn't the visibility number itself, but the gap between self-image and AI image. You consider yourself the leading digitalization consultancy in the region, but the AI names you exclusively for process optimization. This contradiction shows you where your public communication misses your strategy.

Just as telling is when the AI doesn't connect you with your target segment at all. If you want to increasingly support family businesses with succession going forward, but don't show up for a single succession prompt, then that's not a GEO problem, it's a positioning problem that GEO only makes visible.

Use these discrepancies as a compass. They tell you which topics you have to occupy before the next prospect types their question into a chat window. The analysis is thus less a marketing report than an honest assessment of your perception in the market.

Realistic expectations and the long haul

GEO isn't a switch you flip. Training data updates at long intervals, and it often takes months before new content seeps into the model's knowledge. It works faster with web-based systems like Perplexity, which retrieve current pages live. There you see an effect sometimes within weeks, if your content is easy to find and clearly structured.

Don't expect miracles against the global market leaders. A realistic goal for a specialized consultancy isn't first place for "best management consultancy," but reliable presence for your five to ten niche prompts. That's exactly where the engagements that fit you get decided.

So measure regularly, not once. A quarterly rhythm is enough to detect movement and prove the effect of your content. Those who document the development turn a diffuse unease about AI into a controllable metric, and treat visibility in language models with the same seriousness as Google ranking used to be treated.

Your 30-day roadmap to the first robust measurement

Before you tinker with content, you need a baseline. In week one, collect ten to fifteen questions that potential clients really ask: "Who helps with succession planning in mid-sized companies?" or "consultancy for restructuring in the manufacturing industry." Pose these questions to several AI systems and record whether your consultancy appears, in what context and alongside which competitors. That's your zero line against which you measure every later progress.

In weeks two and three you sort the gaps. If you're not mentioned at all, substance is missing on the web. If you're classified wrongly, your positioning in the text isn't right. In week four you deliberately publish two to three expert articles on exactly the topics where you were invisible. That way you turn a diffuse feeling into a measurable process with a clear before and after.

Common questions from consultancies about AI analysis

"Isn't a good website enough?" No. AI systems draw their knowledge from many sources: expert articles, interviews, directories, talks, third-party mentions. A pretty homepage alone delivers too little signal. What's decisive is how often and how consistently your name appears in connection with your core topics on the open web.

"How often should I measure?" For consultancies a monthly rhythm is enough. AI answers change more slowly than classic search results, because the underlying models and sources aren't reassessed daily. A quarterly comparison shows you the real trend without letting daily fluctuations unsettle you.

"Does this do anything for a small consulting office?" Especially then. Small, specialized firms often have a sharper positioning than large generalists. If you consistently occupy your niche topic, you get recommended more easily for exactly these questions than a more broadly positioned competitor without a clear profile.

Where the analysis reaches its limits

As useful as the numbers are, they don't replace judgment. An AI answer is always a snapshot of probable phrasings, not an objective ranking of the best consultancies. Two almost identical questions can deliver different names. That's why you should never judge based on a single answer, but on patterns across many runs.

Also true: visibility isn't an engagement. If you get recommended, you've opened the door, no more. The actual close still depends on trust, references and the personal conversation. Use the analysis as an early indicator of your market presence, not as a substitute for sales. It shows you where you stand in the AI's relevance space, the path to the engagement you walk yourself afterward.