Last updated: 24 October 2026
AI visibility KPIs for asset managers come down to three questions: does AI name your funds, does it get them right, and does it recommend them when a selector asks for a pick. Each question needs two numbers, one the board can read and one the team can act on. A single visibility score answers none of them.
94% of business buyers use AI during a purchase, then check the answer against peers and analysts. The shortlist starts in that answer.
Forrester, The State of Business Buying 2026, nearly 18,000 business buyers worldwide.
01 · Named
Does AI name your fund?
- Board
- Share of voice
- Team
- Shortlist rate
02 · Right
Does AI get your facts right?
- Board
- Fact accuracy
- Team
- Message pull-through
03 · Recommended
Does AI pick your fund?
- Board
- Demand signal
- Team
- Recommendation rate
Report each KPI by engine · mode · language · persona
Traffic is the wrong test
AI engines send asset managers very little traffic, and that is the usual case against investing in AI visibility. It measures the wrong event. A wholesale selector who asks ChatGPT for European small-cap funds with a five-year record reads the answer, not your factsheet. If the fund is missing, or present with last year’s fee, the selection has started without you. Referral sessions count the few who click. They miss the many who never needed to.
Fame does not fix it either. In the Index, how well known a firm is barely predicts its AI share of voice. And the last mile of fund distribution was never visible in web analytics: it runs through bank shelves and buy lists, which is why attribution breaks in asset management long before AI arrives. AI visibility needs its own scorecard.
Six KPIs answer three questions: named, right, recommended
The scorecard adapts the seen, believed, chosen model that Seer Interactive presented at MAICON in October 2026 to the way fund buyers ask. Board KPIs show whether the investment works. Team KPIs show what to fix first. For the metric underneath, see what AI share of voice is and why it moves before traffic.
| Question | KPI | How to calculate | Read by |
|---|---|---|---|
| Named | 1. Share of voice | Mentions of your firm or fund ÷ all mentions across a fixed peer set, per engine and language | Board |
| Named | 2. Shortlist rate | Searches the engine runs behind the answer that already name you ÷ all such searches | Team |
| Right | 3. Fact accuracy | Figures quoted about you that match your latest published value ÷ all figures quoted about you | Board |
| Right | 4. Message pull-through | Answers that repeat one of your key messages ÷ answers that mention you | Team |
| Recommended | 5. Recommendation rate | Answers to selection questions that put your fund forward ÷ all answers to selection questions | Team |
| Recommended | 6. Demand signal | Branded searches for fund names and ISINs, factsheet downloads and selector inbound, read against KPIs 1 to 5 over time | Board |
Named: share of voice and shortlist rate
Share of voice is the share of mentions your firm or fund receives in AI answers to buyer questions, against a fixed peer set. Measured that way it sums to 100% within each cut and stays comparable from one quarter to the next. Track flagship funds as well as the firm: a selector asks for a global high-yield fund, not for a parent brand. In the Index, about ten firms carry the category, so for most managers the realistic target is a place on the shortlist, not first place.
Shortlist rate is the share of the engine’s own background searches that already contain your name. Before answering, engines rewrite a long prompt into several web searches, a process called query fan-out. In Seer’s example, most of those searches were site searches on each shortlisted vendor’s own pricing page: the engine had picked its candidates before it read a single page. For a fund, the page it goes to next is the factsheet or the fees page. Not every engine exposes these searches, so report the KPI where it can be observed and say where it cannot.
Right: fact accuracy and message pull-through
Fact accuracy is the share of figures AI quotes about you that match your latest published value. For an asset manager, accuracy is mostly freshness: AUM with its as-of date, ongoing charges, share classes, SFDR classification, launch dates, a fund merged two years ago. An engine that quotes your 2023 AUM makes you look smaller than you are, in front of the person choosing. Report the share of correct figures and the age of the wrong ones. Fund names make it harder: European ETF names keep getting longer, and engines shorten, merge and confuse them. A citation that points to a superseded KID or factsheet is a compliance exposure, not only a visibility problem, and the sources engines cite decide which version of you they repeat.
Message pull-through is the share of answers that describe you in your own key messages. It needs input from you: three to five messages per flagship fund. Then count how often AI repeats them rather than the wording of a fund database or a review site. In Seer’s example, the engine described the vendor almost entirely through third-party review sites. Low pull-through means someone else owns your description.
Recommended: recommendation rate and demand signal
Recommendation rate is the share of answers to selection questions that put your fund forward as the pick. Being one of ten names in a list is not being picked. Test whether the pick holds: rerun the prompt with a named competitor in it. A recommendation that flips the moment a selector mentions an alternative is weak.
Demand signal reads branded searches for fund names and ISINs, factsheet downloads and selector inbound against KPIs 1 to 5 over time. There is no clean line from an AI answer to a subscription, and in a market where banks hold the shelf there never was one. Read demand as direction over several quarters, alongside the four layers an asset manager can actually measure.
Every KPI needs four cuts: engine, mode, language, persona
A blended number hides the cut where you are losing. Report each KPI four ways.
- Engine. ChatGPT, Gemini, Copilot, Perplexity, Claude and Le Chat cite different sources and recommend different funds.
- Mode. An engine answering from memory and the same engine searching the live web can put a different firm first. In the Q3 reading they did. A recall lead expires the moment the engine searches.
- Language. The same question in Dutch and in Spanish returns a different shortlist. A pan-European average hides the market you sell in. See how the shortlist changes by language.
- Persona. An institutional allocator, a wholesale selector and a retail investor phrase the same need differently and get different answers.
Generic AI visibility tools leave these cuts out. They are where the actions are.
One run is an anecdote
AI answers change between runs. Brands are fairly stable; the sources cited are not. In OtterlyAI’s public experiments, 12 to 19% of brands appeared on a single day only, against up to 67% of cited sources. Seer’s repeated-run analysis shows the same pattern from the other side.
Rules that keep the number honest:
- At least five repetitions per prompt, spread across different days.
- A monthly reading for the team KPIs and a frozen quarterly panel for the board, as in the European measurement layer.
- Direction against named peers, not an absolute score.
- Each engine reported separately. A blended number is never the only headline.
Three numbers to drop from the report
- A single blended visibility score. It averages away the language, the persona and the engine where you are losing.
- AI referral traffic as the headline. It counts the clicks, not the decisions made without one.
- Raw citation counts. Citations churn faster than anything else in the answer. Track their stability instead.
If the board asks whether AI search is working, the honest answer is three numbers, cut four ways, read against the same peers every quarter.
Method
The AI Share of Voice Index applies this scorecard to European asset management: 2,243 answers from 9 engines in 5 languages, across 104 firms, Q3 2026 reading. Fixed prompts, fixed peer set, repeated runs. Visibility only, no investment view.
Frequently asked
What are AI visibility KPIs for an asset manager?
Six numbers across three questions: share of voice and shortlist rate (does AI name you), fact accuracy and message pull-through (does it get you right), recommendation rate and demand signal (does it pick you). Each is reported by engine, mode, language and persona.
How is AI visibility different from SEO?
SEO measures where your pages rank and how many people click. AI visibility measures what the answer says about you, including when nobody clicks. The answer is often built from pages you do not own, such as fund databases, press and review sites.
Which AI engines should an asset manager track?
The ones your buyers use in each market. At minimum ChatGPT, Gemini, Copilot, Perplexity and Claude, plus Le Chat for European coverage. Report each separately: they cite different sources and recommend different funds.
What is a good AI share of voice for an asset manager?
There is no absolute benchmark. The category is concentrated: in the Index, a short list of about ten firms takes a large share of all mentions. Read your number against the peers you pitch against, per language, and track the direction quarter to quarter.
Book a 30-minute readout See the full reading
I work for Allianz Global Investors, one of the measured firms, scored on the same rule.
Sources: Forrester, The State of Business Buying 2026. Alisa Scharf, Measuring AI Search, Seer Interactive at MAICON, October 2026. Wil Reynolds, Re-imagining AI Visibility KPIs in a Query Fan Out world, Seer Interactive. OtterlyAI public GEO experiments, #32. AI Share of Voice Index, Q3 2026 reading.