A quarter now take AI financial advice . Nobody approved it.

The AI financial advice number the trade press walked past#

A fox does not chase the clamour. It waits, watches the wind, works out which rustle repays a run, then moves once. The trade press took the loud line from TSB's research, published on 7 August 2026: 59% of respondents who acted on financial advice found on social media have regretted it, according to TSB. Regret reads well. Regret does not redraw a media plan, and it says nothing at all about AI financial advice.

The quieter question in the same release is about AI financial advice, and it is the one worth working. TSB reports that 25% have used AI for financial advice, rising to 43% of 25-34 year olds, and that 51% are not confident they can identify AI-generated financial content. That is the AI financial advice gap in two lines: a quarter of social media users are asking the machine, and half the sample say they are not confident they could tell a machine's answer from a human's.

The cost is measurable. Of the 32% who have acted on financial advice on social media in the past twelve months, 56% lost money as a result, at an average loss of almost £700, per TSB's news release. TSB sets that beside investment fraud losses averaging £3,000 per case. Those are two different measures of the same slow leak, and neither of them is the point of this piece.

What the loud number hides#

Carys Barnes, Head of Current Accounts and Savings at TSB, is the named spokesperson on the release. That detail matters more than it sounds. A named source with a stated method is the raw material of a citation, and an unattributed statistic is the raw material of a correction. Every generative engine reading this story is making exactly that distinction on your behalf.

Set the two channels side by side and the asymmetry appears. Social media financial advice is the channel every regulator, every platform and every compliance officer is already watching. AI financial advice is the channel almost nobody has instrumented. Trust in the older channel is already high: of those who had seen financial advice on social media, TSB reports 56% trusted it, rising to 72% of 25-34 year olds. No equivalent trust figure has been published for the machine, which is precisely the measurement gap.

That is a trust gap with a trail leading straight to your product pages. If a quarter of social media users are taking AI financial advice about what to do with its money, the assistant's answer is a distribution channel whether you planned for one or not.

Read the methodology note before you quote the number#

Here is the part of the release that will not appear in anyone else's write-up, and flagging it is more useful than laundering it. TSB states its sample size two different ways. The summary says almost 2,000 UK adults. The methodology note says 3,457 consumers who use social media, with data collected between 20.06.2025 and 15.07.2026. Those are not the same study description.

It also matters that this is a distinct study. TSB published separate social media research in July 2025, and conflating the two produces a number that belongs to neither. The fieldwork window quoted above spans mid-2025 to mid-2026, which is longer than most consumer surveys run and is itself a fact a careful, candid citation should carry.

Naming the wobble is not pedantry, it is the point. A fintech marketing agency that quotes cleanly builds a citation record a machine can rely on, and a record like that compounds quietly. The alternative is a brittle brief: numbers borrowed from a summary line, unchecked, and reproduced until somebody senior asks where they came from.

Fieldwork provenance is worth the paragraph because a generative engine repeating your page inherits your sourcing habits. Cite "almost 2,000 UK adults" simply because that sentence sits highest on the page, and you have taken somebody else's imprecision and put your own domain behind it. The correction, when it comes, lands on your brand rather than theirs. Every AI financial advice figure quoted in this piece rests on the methodology note.

The second version is longer, duller and considerably harder to argue with. It is also the one an assistant can lift without inventing anything, because every clause in it survives on its own. That is the whole of generative visibility in a sentence, and it is what content marketing is actually for once the market has an AI financial advisor in it.

Sources like Censuswide and standards bodies like the Market Research Society exist precisely so that a claim can be checked. Use them by name, and the check gets easier for everybody downstream of you, including the model.

The compliance perimeter has already moved, just not where you think#

Financial promotions compliance is not a new discipline and the rules are not obscure. A financial promotion is an invitation or inducement to engage in investment activity, it must be fair, clear and not misleading, and it must be issued or approved by an authorised firm, as the FCA's financial promotions guidance sets out. The regime followed the audience onto social platforms, and the FCA's finalised guidance on financial promotions on social media is explicit that a post, a story or a creator partnership can be a promotion like any other.

Approver liability is the sharp edge of that regime, and it is where firms discover the perimeter the expensive way. We wrote about who carries the consequence in who carries the can when a promotion goes wrong, and the mechanics of the risk in promotion risk in financial services marketing.

The platforms police the same perimeter privately. Advertisers running financial services creative face a certification and verification layer before a single impression is served, set out in Google's financial products and services policy and in Meta's advertising standards. Certification, approval, register: three separate gates, all of them on the channel that is already being watched.

Supervisory scrutiny is widening, so read it precisely#

On 7 August 2026 the FCA said it had sent an information request to 900 Annex 1 firms, having contacted 300 firms in late 2025, per its statement on increased scrutiny of Annex 1 firms. Be precise about the population. Annex 1 covers lending and custody businesses rather than payments or crypto, so this is evidence of supervisory appetite widening, not a rule that reaches every fintech balance sheet.

Across the Channel the same widening is visible with a different accent. The EBA, EIOPA and ESMA issued a joint statement on 31 July 2026, reference JC 2026 25, on frontier AI models. Three authorities, one document, proportionality anchored to Article 4 of DORA. The European Securities and Markets Authority is a signatory rather than the sole author.

Nor is every regulatory movement additive. The FCA finalised rules on 3 August 2026 that it expects to save firms £108 million a year, cutting industry transaction reporting costs from £493 million to £385 million and reducing reportable fields from 65 to 52, with the rules effective from 3 April 2028, per its announcement of PS26/15. Regulators do remove cost. They just rarely remove it from the channel you were worrying about.

In the EU the transparency burden on generated content itself sits with the model provider and deployer under Article 50 of the AI Act, not with the brand an assistant happens to paraphrase. Read that carefully and the gap in the middle becomes obvious. Nobody in the chain is responsible for whether a piece of AI financial advice about your product is right.

So the practical financial promotions compliance question is narrower than the noise suggests. Where your own page is the source an assistant paraphrases, the promotion rules reach it already, because the promotion is yours. Where somebody else's page is the source, you carry the consequence without the control. Both roads run to the same simple remedy: publish precisely, publish plainly, and keep the definitive version current.

The AI financial advice channel has no gatekeeper#

Here is the asymmetry stated plainly. Social media financial advice runs through advertiser certification, approver liability and a promotion register. AI financial advice runs through none of them. There is no certification gate on an assistant's answer, no authorised firm approving the paraphrase, and no artefact filed anywhere that a compliance team could review the following morning.

The consumer-side numbers explain why that vacuum matters. TSB reports 46% said they do not know how to check the credentials of whoever is giving the advice, 49% say financial content has made them feel pressured, and 33% have considered changing their financial goals as a result. Checking credentials is precisely the task a person delegates to an assistant, and precisely the task the AI financial advice channel performs least visibly.

This is not only a young-person story#

The age split is instructive, and it cuts against the lazy reading. TSB reports 27% of 25-34 year olds sought savings advice and 18% sought investment advice, which is the cohort everyone expects. It also reports that 22% of 45-54 year olds and 18% of over-55s had acted on financial advice found on social media. Older money moves too, and it moves in larger amounts.

Being the source an assistant repeats is now a compliance control#

This is the practical turn, and it is a reframe rather than a new budget line. For years, being cited by a machine was a growth tactic that sat with search. With a quarter of social media users taking AI financial advice, it is a compliance control, because the alternative to your sentence is somebody else's guess about your product, delivered with your brand name attached and no approver anywhere near it.

A fox does not fight the hedgerow, it finds the gap and takes the trail through it. The AI financial advice gap here is specific: assistants prefer sources that are self-contained, dated, attributed and dull. That is an unusually comfortable brief for a regulated firm, because fair, clear and not misleading is already the house style your compliance team enforces.

None of that requires a replatform and none of it depends on a prediction about regulation. It is the same discipline we run through SEO and GEO services, applied to a category where a wrong answer is a conduct issue rather than a bounced session.

Start with the 51%. Take the ten questions your best customers actually ask, put them to two or three assistants, and read what comes back as though a compliance officer were reading it. Most firms find the same thing: the answer is roughly right, dated wrongly, and sourced from somebody who has never seen the product. That test takes an afternoon, and it turns an abstract AI financial advice worry into a list with owners against it.

Two adjacent habits help. First, keep the paid and organic stories consistent, because a certified financial ad and an uncertified assistant answer describing the same product should not contradict each other; that alignment is a paid social job as much as a content one. Second, say the boring, verifiable thing about your product in the same words everywhere, which is what brand strategy buys you when a machine is doing the paraphrasing.

There is precedent for trust arriving as an infrastructure feature rather than a campaign. We looked at exactly that dynamic in Confirmation of Payee and the trust it buys, where a dull back-office control turned into the most persuasive marketing asset the sector had.

The quiet quarry, then, is not the finfluencer. It is AI financial advice answering a savings question at eleven at night for somebody who cannot check the credentials and, on TSB's numbers, probably will not try. A fintech marketing agency that treats that answer as owned inventory is doing compliance work and growth work with one motion, and a firm that leaves it unowned has an approver-free promotion running at scale on its behalf.

That is the whole argument. The trail is visible, the ground is soft, and the tracks are fresh. If you want the AI financial advice surface mapped, measured and owned rather than described, that is what folkfox FinTech marketing does.

Frequently asked questions#