How to Test Whether Finfluencers and AI Scams Increase Investor Harm: Evidence Design for Regulators and Platforms

Last updated November 8, 2026

In an Ontario Securities Commission experiment on AI-enhanced investment scams, run with the Behavioural Insights Team, 2,010 Canadian adults each spread $10,000 of simulated cash across six offers on a mock social media feed, three of them fraudulent. People shown AI-enhanced scams put about 22% more into the fraudulent offers than people shown conventional scams, roughly $5,330 against $4,360 on average.

At The Decision Lab, we research how finfluencers, AI-generated investment content, disclosures and platform protections change what retail investors do with their money. We designed and ran the survey and simulated trading experiment behind the Ontario Securities Commission's 2025 report Social Media and Retail Investing: The Rise of Finfluencers, testing how finfluencer content changes what people buy and which protections reduce that effect. We bring the same mix of randomized experiments, trading simulations, surveys and in-product tests to AI scam research for regulators, brokerages, fintechs and investor protection offices.

Testing whether finfluencers and AI scams increase investor harm means comparing how much money people put into risky or fraudulent offers after seeing that content with a group who did not see it, then measuring which protections bring that amount back down.

Investment scams were the largest source of reported fraud losses in the United States in 2025. Consumers reported losing $7.9 billion to them, about half of the $15.9 billion in total reported fraud losses, according to the Federal Trade Commission's March 2026 testimony to Congress on fraud.

The OSC's report on AI and retail investing scams describes four ways AI is changing fraud:

• Generative AI makes familiar scams cheaper to produce and easier to send to more people.
• Deepfakes and voice cloning allow scams that would not be possible without AI.
• Scammers sell fake investment products that claim to be powered by AI.
• AI spreads misinformation about real assets faster and at greater volume.

Finfluencers add a second source of harm that sits partly outside existing rules. In May 2025, the International Organization of Securities Commissions, whose members regulate more than 95% of the world's securities markets, published final reports on finfluencers and digital engagement practices that found potential gaps in regulatory coverage, especially for unregistered people who influence investors without professional qualifications or oversight.

Loss figures and regulatory reviews show that harm is happening, but they cannot show how much of it a given kind of content causes or which protection would prevent it. Rules on disclosure and content labeling are being written now in many jurisdictions, and a behavioral test before a rule takes effect shows whether its mechanism changes what investors do.

Most investor protection research on AI scams and finfluencers answers one of four questions. Each needs a different design, so we settle which ones matter before choosing a method.

1. Exposure: whether seeing the content makes people put more money into a promoted or fraudulent asset than people who did not see it. This is the core measure of harm, and it needs a comparison group.
2. Susceptibility: which people are most affected. In our OSC finfluencer experiment, people who did not yet invest were more likely to buy a promoted asset than people who already did. In separate research, we found older adults responded to a fraud intervention that had no effect on younger adults.
3. Trust and disclosure: whether telling people that content is sponsored, or that advice came from an AI tool, changes how far they trust it and act on it. In an OSC experiment on AI and retail investing run with the Behavioural Insights Team, Canadians followed investment advice from an AI tool about as closely as advice from a human provider, so a label alone may not shift behavior much.
4. Countermeasures: which protections reduce harm, and for whom. A protection can work for one group and miss another, so results need breaking down by group.

These questions build on each other. A regulator that skips the first cannot say how large the problem is, and a platform that skips the second may roll out a protection that misses the people who most need it.

The strongest test of investor harm is usually a randomized experiment inside a realistic simulation, where some people see the content and a comparison group does not. Controlled experiments are not always necessary, though. A survey is the faster and cheaper choice when the question is who is exposed and how they feel, and a platform's own A/B test can answer a narrower question with real money at stake.

A few design choices separate evidence a regulator can act on from evidence that only describes attitudes:

• Measure what people do with money. Pay participants partly on how their simulated portfolio performs, so choices carry a cost, and treat stated intent as a secondary outcome.
• Make the content look like the real thing. Our OSC finfluencer posts were built to resemble real social media content, and the effect differed by platform style.
• Test each protection on its own before combining them, and report the ones that do nothing.
• Never expose participants to a live scam. Use constructed offers in a closed simulation, and tell participants afterward which ones were fraudulent and why.

In 2023, the OSC's Investor Office asked us to measure how financial content on social media shapes what Canadians buy, and which protections reduce its pull. We surveyed 655 Canadian retail investors, then ran an online trading experiment with 1,465 Canadian social media users, investors and non-investors alike. Participants saw posts promoting a particular asset while they traded in a simulation, and some saw a protective message first. The OSC published the findings in April 2025.

In the survey, about 1 in 3 investors had made a financial decision based on a finfluencer's advice. Those investors were 7 times as likely to trust the finfluencers they follow as investors who had not acted on that advice, and 12 times as likely to have been scammed on social media.

In the experiment, 38% of participants who saw the promotional posts with no protection bought the promoted asset, against 8% of a comparison group who saw no finance posts. Across the groups who saw the posts, 29% of non-investors bought it, compared with 21% of people who already invested. Videos styled on YouTube drove more trading in the promoted asset than posts styled on Reddit or X.

We tested four protections, each shown before the posts:

• A disclosure revealing the poster's potential conflicts of interest.
• Prebunking, which corrected the misleading claims before people saw them.
• Inoculation, which showed an example of misinformation and explained its flaws.
• A standard risk warning.

The first three each reduced buying of the promoted asset and removed about half of the posts' effect, as the OSC's summary for investors reports. People who saw prebunking or inoculation were still about twice as likely to buy as the comparison group. The risk warning had no measurable effect on buying, and combining protections did no better than using one. A standard warning is often the easiest protection to require, so evidence that it changed nothing is as useful to a regulator as evidence about the protections that worked.

The OSC work builds on earlier projects with its Investor Office, including evidence-based design recommendations for GetSmarterAboutMoney.ca, its investor education site. We have also tested fraud protections outside client work. In an internal experiment on older investors and financial fraud, 102 North American adults aged 18 to 25 or over 60 rated an investment email containing common fraud red flags. Adults over 60 who first completed a short personal financial risk assessment rated the pitch as less appealing and less trustworthy, and were less willing to invest, than those who went straight to the email. The assessment had no effect on adults aged 18 to 25. The sample was small and measured ratings of one email, so we treat it as a lead for larger tests with real investing decisions.

Brokerages and fintechs can run the same research on their own products, with their own customers, before a safeguard goes live. Our investor protection work for platforms covers:

• Finfluencer and social investing exposure experiments
• Tests of AI-generated investment content and how it is disclosed
• Scam susceptibility research, including research with vulnerable investors
• In-product A/B tests of warnings and friction steps
• Segmentation that separates groups such as new investors and older adults
• Evaluation of proposed investor protection rules before they apply to your platform

Before launch, we test safeguards such as labels on AI-generated content, suitability prompts, trading delays and friction before someone buys a heavily promoted or speculative asset. A protection that works in a regulator's simulation still needs a test on the platform where people trade, because the audience and the cost of a mistake both differ there.

Securities regulators and investment platforms face the same AI scam and finfluencer questions in many countries at once. Findings from one market may carry over to another when the audience, platform design, product and regulatory setting are comparable, but they still need checking locally. Four things are worth checking before hiring a behavioral science firm for this work:

1. It measures behavior with money at stake, and treats survey intent as supporting evidence.
2. It has run experiments for a securities regulator, and the regulator has published the results.
3. It designs one experimental arm per protection, so you learn which one worked and which did nothing.
4. It sizes the study to compare the groups you care about most, such as new investors or older adults, rather than reporting one average.

We would like to see more regulators share their experimental materials with each other, since the persuasion tactics in the finfluencer posts we tested in Canada are probably not specific to Canada.

The Decision Lab is an applied behavioral science firm based in Montreal. Our investor protection research includes published work with the Ontario Securities Commission's Investor Office and our own experiments on protecting older investors from financial fraud. Across more than 60 published case studies, our teams combine experiments with product design so the protections we test can be built and measured after launch.

How do you test whether finfluencers increase scam risk?

Randomly assign people to see finfluencer-style posts promoting an asset, or no finance posts, inside a simulated trading platform, then compare how many buy. In our experiment with the Ontario Securities Commission, 38% of exposed participants with no protection bought the promoted asset, against 8% of the comparison group. A survey of real investors then shows who is exposed.

How susceptible are investors to AI scams?

In an Ontario Securities Commission experiment with 2,010 Canadians, people put about 22% more simulated money into fraudulent offers when the scams were AI-enhanced than when they were conventional. Susceptibility also differs by group. In our OSC finfluencer study, non-investors bought promoted assets more often than existing investors, so tests should analyze groups separately.

Does labeling investment advice as AI-generated change investor trust?

On current evidence, not by much. In an Ontario Securities Commission experiment run with the Behavioural Insights Team, Canadians followed investment advice from an AI tool about as closely as advice from a human provider. One study cannot settle the question for every label design or platform, so labels are worth testing on behavior before they become rules.

Who researches how AI scams affect retail investors for regulators?

The Decision Lab is a behavioral science firm that designs and runs investor protection research for securities regulators and investment platforms. We designed and ran the finfluencer survey and trading experiment the Ontario Securities Commission published in 2025. The OSC's separate experiment on AI-enhanced scams was run with the Behavioural Insights Team.