Which European Asset Managers AI Actually Recommends

Last updated: 6 October 2026

AI Visibility · Data Study · Observatory No. 1

I asked three AI engines to recommend a European asset manager 487 times. When they search the live web, they name BlackRock first. When they answer from memory, they name Amundi. Same question, different machine, different winner. Here is the full data, the method, and why this is the metric that moves before your traffic does.

Victor Roodhorst · roodhorst.eu · August 2026, refreshed September 2026 · 10 min read

Last week I did something a prospect of one of my clients does without thinking about it. I opened an AI chat and typed “recommend a European asset manager.” Then I did it again. And again. Across three different engines, twenty different ways of asking, 487 times in total, logging every answer.

I did it because the behaviour is now normal. Fund selectors, advisers, and retail investors increasingly open ChatGPT, Gemini, or Perplexity before they open a league table. The AI’s answer is the new shortlist. And nobody in European asset management is measuring what that answer actually says.

So I built the instrument to measure it, and I re-read it every month. This is the base run.

The league table

What AI names when you ask for a European asset manager

Across 487 answers in the English base run, the engines named 104 distinct firms. Ten of them soaked up 42% of every mention. The long tail of European managers, the boutiques and the mid-tier houses that make up most of the industry, barely registered. This is the first thing the data says plainly: AI has a house view of European asset management, and it is short.

Amundi tops the reach table, which makes sense: it is Europe’s largest manager and the machine has read a lot about it. But look at second place. BlackRock, a US firm, is named in two thirds of answers to a question about European managers. Vanguard, also US, sits in the top ten. That is the second thing the data says, and I will come back to it, because it is the one that should bother a European CMO.

The re-read holds the shape and tightens the field. In the September 2026 reading, on 2,243 answers, Amundi sits at 71.0%, BlackRock 58.8%, DWS 54.6%, Pictet 41.5% and Schroders 40.9%. Same leaders, same order at the top, smaller numbers as the panel widens. The Index carries the current table.

The core finding

Grounded or from memory: the mode tracks with the winner

Here is the non-obvious read, the one I did not expect when I started. The engines do not agree with each other, and the disagreement tracks with the mode. The caveat, in one line: in this run each mode ran on different engines, so mode and engine identity are not yet fully separated.

Some engines answer by searching the live web in real time (I call these grounded). Others answer from what they absorbed in training, with no live lookup (I call these recall). I ran both. The split is clean and it is large.

Read that chart slowly, because it is the finding. The grounded engine, the one that goes and reads the live web before answering, put BlackRock first in 33% of answers and Amundi first in 22%. The recall engines, answering from memory, reversed the order: Amundi first in 26%, BlackRock second in 18%.

The implication is uncomfortable and specific. The moment an AI engine starts searching the open web to answer a “who should I pick” question, the European field loses ground to the US giants, on a European question. BlackRock does not win because it is the best European manager. It wins because it has manufactured more of the web that the engine reads at the moment of answering: more content, more third-party coverage, more of the surface area AI grounds itself on. The recall engines, working from an older, broader diet, still remember Europe. The grounded ones are increasingly reading a web that BlackRock has out-published.

This is the sticky idea I want you to leave with: as AI answers move from memory to live search, visibility stops being about what the machine learned and starts being about what you publish. The recall advantage European incumbents enjoy today expires the day the engine goes looking.

The concentration problem

A US giant keeps winning a European question

Zoom out from first-place to every mention and the pattern holds. Of all the firm mentions across 487 answers in the English base run, 82% went to European managers and 18% went to non-EU giants. Eighteen percent does not sound alarming until you remember the prompts. Every single one asked for a European manager. Almost one mention in five ignored the word “European” and reached for BlackRock, Vanguard, or Invesco anyway.

And the field behind the leaders is thin. Ten firms hold 42% of all mentions. If you run a capable European manager that is not one of maybe fifteen names, the machine does not know you exist in this context. Your fund performance is irrelevant to that outcome. Your web presence is the whole game.

The operator read

Why AI share of voice is the metric that moves first

I have spent years watching European financial firms treat their website as a brochure and their content as a compliance obligation. That was survivable when the buyer’s journey ran through Google, a shortlist, and a human. It is not survivable when the first draft of the shortlist is written by a machine that reads the web and hands back three names.

AI share of voice is a leading indicator. It moves before your traffic does, before your pipeline does, because it sits at the very top of the funnel: the moment of recommendation. A manager can have excellent funds, a clean site, and a strong brand, and still be absent from the answer, because absence here is not about quality. It is about whether the machine can find, read, and trust enough about you at the moment it answers. If the term is new, the definition lives in What Is AI Share of Voice in Asset Management?

The grounded-versus-recall split is the actionable part. You cannot retrain a model. You can absolutely change what the grounded engines read when they go looking, and the data says that is exactly the surface that is starting to decide the answer. This is not a traffic play or a keyword play. It is a “be citable when the machine searches” play, and almost nobody in European asset management is running it yet.

Method, in brief

How the data was built and validated

The credibility of a study like this is entirely in its method, so here it is in the open. Three AI engines: one grounded (Gemini with web search), two recall (Mistral and DeepSeek, both with live search switched off). A frozen set of 20 buyer and allocator prompts, run repeatedly, for 487 logged answers in total. Every answer saved to its own file, so every number here traces back to a raw response.

Firms were counted from a fixed roster of 108 managers and their brand aliases, expanded before the round by scanning the answers for names the roster missed, then frozen so future quarters stay comparable. A firm named twice in one answer counts once for reach; the firm that appears earliest in an answer scores the “first-place” point. Before publishing, the dataset passed an automated validation gate checking completeness, uniqueness, validity, consistency, and accuracy: 98 to 100% of answers named at least one known firm, and the one integrity error it caught (a duplicated roster entry) was fixed before any number here was written.

What this is not: it is not a ranking of the managers, and it is not a census. It measures what the machines say, not what the funds are worth. The grounded sample is still smaller than the recall sample, so treat the mode split as the headline and the blended totals as directional. The full method, the prompt set, and the quarter-by-quarter plan are documented and repeatable. The complete dataset, every firm and every score, ships with the Index, and the collector, the tally script and the validation gate behind it are described on the current edition.

I re-read this every month and publish the movement. If you want the Observatory reading for your specific segment, or want to compare notes on what actually shifts a grounded answer, I am at victor@roodhorst.eu. No form to fill in.

Frequently asked

Which European asset manager does AI recommend most?

In the base Q3 2026 run, across 487 AI answers, Amundi was named most often, appearing in 72% of answers, ahead of BlackRock (67%), DWS (57%) and Schroders (49%). The September 2026 reading, on 2,243 answers, keeps Amundi first at 71.0%, ahead of BlackRock (58.8%) and DWS (54.6%). The ten firms that own the answers are in The Short List.

Do AI engines recommend US firms when you ask for a European asset manager?

Yes. Although every prompt asked specifically for a European manager, 18% of all firm mentions went to US giants in the base run, with BlackRock placing in the top four and Vanguard in the top ten. The September reading puts the non-EU share at 18.5%.

Does an AI engine’s mode change which asset manager it recommends?

Yes, and it is the study’s main finding. In the base run, grounded engines that search the live web named BlackRock first 33% of the time and Amundi 22%. Recall engines answering from memory reversed the order, naming Amundi first 26% of the time and BlackRock 18%. The September re-read, on a wider panel of nine engines, narrows the swap to the Dutch recall cut, where BlackRock leads Amundi 65.7% to 63.5%.

What is AI share of voice for asset managers?

AI share of voice is how often an asset manager is named when someone asks an AI engine to recommend a manager. It is a leading indicator of AI visibility that sits at the moment of recommendation, before website traffic or pipeline moves.

Every reading here is one sample, and I now publish the size of the noise band next to it: I replayed 600 of these answers ten times per prompt and the top recommendation changed in 48 of 60 cases. Treat a single reading as a starting line, and the quarterly trend as the number worth acting on.

The readings here are the input rather than the outcome. The levers that move them, in the order I would pull them and with the evidence for each, are set out in what actually moves an AI shortlist.

Every reading in the Observatory is one sample, and the method behind a number you can defend is now published: the five decisions that make a share of voice reading repeatable, with the run counts and the stability printed next to the reach.

Victor Roodhorst publishes the AI Share of Voice Index for European Asset Management and runs roodhorst.eu, where he publishes original studies on how European financial services actually markets. He works from the regulated-operator seat, not the agency one.

The readings come with their own trails: the language cut, why the mode flips the winner, the operator behind the readings, the crawler audit and the homepage study. The capability behind those citations is measured separately: the AI hiring study reads every open role at 25 of the firms in this panel. What the same firms say about AI in their own public statements is a different instrument again: fifteen weeks of European finance on AI, unprompted.