<!-- Dating Fraud Detection: Signals, Systems and Metrics | https://whitelabeldating.com/guides/dating-fraud-detection/ | Run by High Intent Media, which also runs Dating Partners and DatingAPI. -->

# Dating fraud detection: signals, systems and metrics

> Dating fraud detection combines checks at sign-up, rules and machine learning models that score behaviour, human review of flagged accounts, and member reports. No single layer is enough. Measure how many members a scammer reaches before a ban, not only how many accounts you remove, and warn the members a banned account has contacted.

## What kinds of fraud does a dating site need to detect?

A dating site needs to detect several different kinds of fake or dishonest account, and each one behaves differently. Treat them as separate problems, because a rule that catches a bot will not catch a patient scammer.

| Type | What they want | Typical signs |
|---|---|---|
| Bots and spam accounts | Clicks to another site, app or paid service | Many accounts from one device or network, identical bios, links in first messages |
| Romance scammers | Money, sent over weeks or months | Quick move to another chat app, reasons never to meet, a crisis or an investment tip |
| Stolen-photo or AI-photo profiles | Attention, or cover for one of the above | Photos that do not match a live selfie, or that appear on other accounts |
| Payment fraud | Free access with a stolen card | Mismatched country, repeated failed cards, then a chargeback |
| Sextortion | Intimate images to use as a threat | Fast pressure to swap images, then demands for payment |

This guide covers detection. For what to tell members and how to support victims, see our guide to [romance scam prevention](/guides/romance-scam-prevention/).

## How big is the problem?

It is large and growing, and regulators now treat it as a core risk of dating services. The US Federal Trade Commission (FTC) said nearly 70,000 people reported a romance scam in 2022, with reported losses of about $1.3 billion and a median loss of $4,400. ICIJ reports that the FTC's figures for 2025 were more than 70,000 reports and $1.47 billion lost.

Not all of that starts on dating services. In the FTC's 2022 data, 40% of people who lost money said the contact began on social media, and 19% said it began on a website or app.

In the UK, Ofcom's guidance for dating services says its research found that 29% of adults who use the internet have experienced romance or dating scams, and 6% have met fraud on a dating site or app. Ofcom lists fraud among the risks a dating service must assess under the Online Safety Act.

## Which signals show a fake or scam account?

No single signal proves fraud, so detection works by adding up weak signals from the profile, the behaviour and the device. These are the common ones.

**Profile signals**

- Photos that fail a comparison with a live selfie, or that match another account.
- A bio or username built to pass contact details, such as a handle for another chat app.
- A stated location far from where the device appears to be.

**Behaviour signals**

- Contacting far more members than a normal person does, or sending the same opening line each time.
- Asking to move to another app within the first few messages.
- Stories the FTC says scammers favour: stationed at a distant military base, working on an offshore oil rig, a sudden illness or arrest, a parcel that needs a fee, or an offer to teach cryptocurrency investing.
- Any request for gift cards, cryptocurrency or a bank transfer. The FTC found gift cards were the most frequently reported way of paying, while cryptocurrency and bank wires together made up more than 60% of reported losses.

**Account and device signals**

- Many sign-ups from one device or network address.
- A network address that keeps changing country.
- A card that fails, or a country mismatch between card, device and profile.

A 2026 report by the Minority staff of the US Congress Joint Economic Committee, based on Match Group documents and interviews, shows why the last group matters. It says internal documents found that Tinder's models missed signals such as device location, changes of network address and swiping behaviour. Match Group told the committee that its rules catch these.

## How do the parts of a detection system fit together?

A working system has four layers: checks at sign-up, automatic scoring, human review and member reports. Each one covers gaps in the others.

1. **Sign-up checks.** Email and phone confirmation, device checks and, on many platforms, a selfie check. See our guide to [identity verification](/guides/identity-verification-dating-sites/).
2. **Rules and models.** Rules are fixed tests ("more than a set number of first messages in an hour"). Models score each account on many signals at once. Both need upkeep.
3. **Human review.** Moderators look at flagged accounts and decide. Our guide to the [content moderation workflow](/guides/content-moderation-workflow/) covers queues and appeals.
4. **Member reports.** A report button on every profile and message, with a category for scams.

The Joint Economic Committee report is a useful warning about upkeep. It says internal documents described many Tinder rules as outdated, redundant and inconsistent, and said most models had not been updated for years. Match Group says it retrains models every year or two, while outside experts told the committee that monthly retraining is the minimum. After the investigation began, Match reported adding 169 new rules, improving 108 and building new models for contact details, usernames and bios.

Two design choices matter as much as accuracy. First, hide a suspect account from recommendations while it is under review, so it stops reaching new members before a decision. Second, give banned members a way to appeal, because every system flags some real people. Our guide to [AI moderation](/guides/ai-moderation-dating/) covers how to keep a person in the loop.

## Do selfie and ID checks stop scammers?

They reduce fake accounts sharply, but they do not end fraud. Tinder's Face Check asks new members for a short video selfie, compares it with their profile photos and checks whether the same face appears on other accounts. TechCrunch reported in October 2025 that Tinder claimed a 60% fall in exposure to potential bad actors and a 40% fall in reports of bad actors since launch.

The Joint Economic Committee report records a similar claim from Match Group: a sustained drop of more than 70% in interactions with bad actors within a month of the US rollout. It adds a caution. The figure combines scammers with spam, so the effect on scams alone is unclear. These are company figures, and no independent audit is published.

The lesson for a smaller brand is that a selfie check is worth having and is not a reason to switch off the other layers.

## What should you measure?

Measure exposure, speed and accuracy, not only the number of accounts removed. A large removal count can hide the fact that each scammer reached hundreds of members first.

| Measure | What it tells you |
|---|---|
| Profile views before a ban | How many members saw the account |
| Share of banned accounts that matched or exchanged contact details | How many members were put at real risk |
| Median time from sign-up to ban | How fast detection is |
| Share of bans overturned on appeal | How often you get it wrong |
| Reports per 1,000 active members | Whether members are seeing more or less fraud |

The committee's report shows why the first measure matters. Using Match Group's data, it says the median scammer on Tinder had fewer than 70 profile views before a ban in May 2024 and more than 500 by December 2025. Match's wider data says at least half of spam and inauthentic accounts were banned within 12 hours, which the report says probably overstates the speed for scammers. Track scammers as their own category.

## What must you do after banning a scammer?

Warn the members they contacted, keep the evidence and check your legal duties. In the US, the Romance Scam Prevention Act was signed in October 2026 and takes effect a year later. The bill text requires a notice to every member who received a message from an account banned for fraud, normally within 24 hours, naming the account and warning against sending money.

That means your system must be able to list everyone a banned account messaged, and send the notice by email or text. Our guide to [US dating regulations](/guides/us-dating-regulations/) has the detail. This is general information, not legal advice.

## What does a white label platform handle?

On a white label network the platform runs fraud detection for every brand, so you do not build any of this. Dating Partners, our sister company run by the same group as this site, lists four layers of member checks: identity checks, AI scam detection, human review of every profile and ongoing behaviour monitoring. Members can also take an optional photo ID check for a badge. It also handles payment fraud screening and chargebacks. Dating Partners is for websites only, and it does not accept adult or sugar dating brands. Our [safety and moderation](/platform/safety-and-moderation/) page explains more.

Whichever provider you consider, ask:

- Which checks run at sign-up, and which are optional?
- How often are detection models and rules updated?
- Is a suspect account hidden from other members while it is reviewed?
- What is the median time to remove a scam account?
- Will members who were contacted by a banned account be told?

## Common questions

### Can AI detect romance scammers on its own?

No. Models catch patterns at scale, but they age quickly as scammers change tactics, and they make mistakes. Keep human review and member reports alongside them.

### Should I tell a member why their account was banned for fraud?

Tell them it was banned and how to appeal, but do not reveal the signals that caught them. Detail helps the next fake account avoid the same checks.

### Do scammers pay for subscriptions?

Some do, often with stolen cards, which then come back as chargebacks. That is one reason payment fraud screening and profile fraud detection should share signals.

### Are AI-generated profile photos a real problem?

They make stolen-photo checks less useful, because the face has never appeared anywhere else. A live video selfie compared with the profile photos is the main defence platforms now use.

### How quickly should a scam account be removed?

As fast as you can, and before it reaches many members. No regulator sets a number, so track your own median time to removal and work to bring it down.
