UK wealth management: The answer changes after you sign.

Somewhere in England right now, someone who has just sold a company, or inherited, or watched retirement arrive on the calendar, is asking an AI model who should manage the money. The model answers in seconds: a shortlist, the risks, a favourite. No firm hears about that conversation. We measure it.

What we measured

From 17 August to 13 September we ran this market's buying questions through the models across the decision journey, measured with three models: Gemini, Claude, and ChatGPT. The early stages are measured daily; the late-journey stages (retention, advocacy) are sampled weekly, because loyalty moves slower than choice.

The questions are the ones buyers ask, in their own words:

• "What is the total annual cost, including all management and performance fees?"
• "Are you a fully independent wealth manager, or do you have ties to specific product providers?"
• "How do wealth managers help with inheritance tax planning?"

Read that second one again. Before a single firm is compared, the buyer is already probing the category's structure: who is advising me, and who pays them to say it.

The judging sheet

The verdict: the default converts upward

Coutts leads 81.1% of evaluation answers and 90.3% of decision answers. That direction matters: most brands lose share when the question moves from "who should I compare" to "who should I choose". Coutts gains nine points. The models do not merely include it; they conclude with it. And the conclusion holds through the whole journey: 92.2% of retention answers, 81.5% of advocacy answers. Compared, chosen, kept, and recommended: the full house that edition 5 measured for Zendesk in Danish customer service has a British counterpart.

Below the default, the familiar cliff: Barclays Private Bank carries 34.9% of evaluation answers and 15.1% of decision answers. HSBC Private Banking falls from 18.9% to 1.4%. Investec Wealth & Investment from 23.5% to 4.6%. Being in the comparison is not being in the answer, and outside the top three the average decision share in this category is 2.4%.

Chosen, then questioned

Then the journey continues past the signature, and the market turns inside out.

How to read the table: the percentages show how frequently each firm appears in the answers at each stage of the decision journey, measured across the full window, 17 August to 13 September.

A note on the table: the first two columns are measured daily and rest on 1,139 and 1,067 answers in this window; the last two are measured weekly and rest on 141 and 119. Shares in the weekly columns move in coarser steps.

St. James's Place is the second-most recommended firm at the moment of choice: 57% of decision answers, ahead of everyone but Coutts. Then the buyer becomes a client, the question becomes "should I stay?" or "would you recommend them?", and the same models nearly stop saying the name: 11% of retention answers, 10% of advocacy.

Edition 1's US CRM decode measured the opposite pattern: brands cut at the decision that dominate the loyalty answers. This market completes the square. A firm can win the recommendation at the door and lose it in the hallway, because the loyalty conversation is judged on different evidence: fee experiences, service accounts, the things existing clients write. Fifty-seven percent of the choosing answers, ten percent of the staying answers. Both are true at once, about the same firm, in the same window.

The firms clients are told to keep

Brewin Dolphin holds a third of decision answers and then owns the late journey: 99.3% of retention answers, 90.8% of advocacy. Investec converts almost nothing at decision (4.6%) and stands at 83.7% of retention. Rathbones runs strong at every stage. Arbuthnot Latham appears in under 10% of evaluation answers and 53.8% of advocacy answers.

These are stage specialists, the pattern edition 3 first measured in UK architecture and edition 5 measured end to end: every stage of the decision journey is its own market, with its own evidence and its own winners. What this market adds is the scale of the divergence. The gap between St. James's Place's decision share and its retention share is 46 points. The gap between Investec's, in the other direction, is 79.

The category's loudest disqualifier

The most severe elimination trigger in this market is conflict of interest, and the buyer language is unambiguous:

"Any perceived conflict of interest, like pushing proprietary products, is an immediate disqualifier."

The most severe eliminator in UK wealth management, applied by every model measured.

And it sits oddly against the market it judges, because proprietary product structures are not an edge case in UK wealth management; restricted advice built on in-house funds is the business model of some of its largest firms, St. James's Place most famously among them. The measurement cannot say whether that is what the loyalty answers weigh; we measure shares, not reasons. What it can say is that the category's number-one disqualifier describes a way of operating that the category's number-two decision brand is best known for, and that the same firm's presence collapses precisely at the stages where existing clients' accounts carry the evidence.

For every firm in the category, the eliminator is the same brief: the independence question is being asked before you enter the room, and silence is scored as an answer.

Nobody owns a criterion

Ten criteria decide this market. Not one has an owner: no firm is named by the models as the answer on fees, on performance transparency, on expertise, on independence. Six categories decoded in this series, sixty criteria measured, zero owned. The pattern that started as a curiosity in edition 1 is now the series' most consistent finding: the criteria that decide AI's answers are standing unclaimed, in market after market.

In a category where the buyers are asking "what is the total annual cost, including all management and performance fees?" and no firm owns cost transparency, the first mover claims it with a page. The mechanics of taking over a criterion the incumbents ignore are the subject of the criteria-flip playbook.

The judge reads the regulator

Where edition 5's model citations were led by review platforms, this market's are led by institutions. The models cite the FCA register, the Financial Ombudsman, MoneyHelper, The Pensions Regulator, Companies House; then the verification layer: Which?, the FT, Morningstar, Trustnet.

That is what a regulated category looks like from inside the answers. The models do not ask the market what it thinks of a wealth manager; they ask the state and the auditors of record. For a firm, the consequence is concrete: your FCA register entry, your ombudsman record, and your coverage in the verification press are load-bearing surfaces of your AI reputation. The review-platform playbook that decides Danish customer service barely applies here. Who the judge listens to is itself category-specific, and it is measurable.

The fragmented ledger

One more pattern from the answers: the ledger is keyed to name strings, not firms. Rathbones carries 36.4% of evaluation answers; "Rathbone Investment Management" carries 6.6% separately. "Quilter" appears in 9.7% of evaluation answers while "Quilter Cheviot" holds 73% of retention answers, two entries for what a buyer would call one place. Edition 3 measured the same mechanism splitting an architecture practice 83/15 across two name forms.

The models' recommendation ledger has no merge function. Every name form a firm tolerates in the wild is a separate account, and the shares it earns are split across them.

What claiming it looks like

The measured brief for this category is the independence question, and any firm could run it. In our voice, the play is one page:

Where the evidence must live: who listens where

The criteria in this market are weighed by all three models, and the surfaces that reach them differ:

ChatGPT and Claude lean on text surfaces: the verification press, domain-authoritative articles under named authors, structured pages on your own domain.

Gemini reads the same institutional layer and rewards fresh, dated material tied to the questions buyers actually ask.

For all of them, in this category, the institutional record comes first: the FCA entry, the ombudsman file, the audited numbers. Own-domain evidence pages are the surface a firm controls; the institutional layer is the surface it must keep clean. The retention and advocacy questions are standing unclaimed either way: the models are answering "should I stay with my wealth manager?" every week, from a sample of the market's smallest, most decisive evidence pool.

The lesson

Sources

The measurement behind this edition: three models (Gemini, Claude, ChatGPT), window 17 August to 13 September 2026, evaluation and decision measured daily (1,139 and 1,067 answers), retention and advocacy weekly (141 and 119 answers). Criteria, eliminators, and buyer language extracted per model per run.

The Sunday Shortlist

All editions