<!-- Trust and Safety Tools for Dating Sites: The Full Stack | https://whitelabeldating.com/guides/trust-safety-tooling-stack/ | Run by High Intent Media, which also runs Dating Partners and DatingAPI. -->

# Trust and safety tools for dating sites: the full stack

> A dating trust and safety stack has six layers: sign-up checks, photo and message screening, behaviour monitoring, member reporting and blocking, a case queue for human moderators, and records for regulators and police. Most brands buy or inherit these tools. Building them only makes sense at scale, and no tool replaces trained people making the final call.

## What is a trust and safety tooling stack for a dating site?

It is the set of tools that stop harmful accounts getting in, spot harm when it happens, and help people act on it quickly. Think of it as six layers, each with a different job.

| Layer | Job | Typical tools | Build, buy or inherit? |
|---|---|---|---|
| 1. Sign-up checks | Keep out bots, banned people and under-18s | Email and phone confirmation, device and IP signals, age checks, optional ID checks | Buy or inherit |
| 2. Photo and message screening | Catch illegal or unwanted content before members see it | Hash matching, image classifiers, text filters for scam language | Buy or inherit |
| 3. Behaviour monitoring | Spot scam patterns over time | Rules and scoring on message volume, copied text, requests to move off the platform | Buy, then tune |
| 4. Member controls | Let members protect themselves | Report, block, mute, safety prompts | Build into the product |
| 5. Case handling | Get each report to a trained person and record the decision | Moderation queue, case notes, appeals, evidence store | Buy or inherit |
| 6. Records and reporting | Show regulators and police what you did | Audit log, risk assessment, law enforcement process | Your own responsibility |

No layer works alone. Screening without a case queue produces flags nobody reads. A report button without moderators produces complaints nobody answers.

## Which trust and safety tools does the law expect?

UK law expects a set of measures that grows with your size and risk, not a fixed shopping list. Ofcom's illegal content codes of practice group the measures for services like dating sites under headings that include governance and accountability, content moderation, reporting and complaints, terms of service, and what members can access and control.

Some measures apply to everyone. Ofcom gives two examples: naming an individual accountable for online safety compliance, and having clear and accessible terms. Others depend on what your risk assessment finds. Ofcom's example is automated tools to detect and remove child sexual abuse material, which may apply depending on a service's size or risk.

Ofcom defines a large service as one used by an average of 7 million or more people in the UK each month. Most dating brands are far below that, but a smaller service that is medium or high risk for a kind of harm can still pick up measures for that harm.

The newest example is intimate image abuse. A measure numbered ICU C14 in the draft laid before Parliament asks certain services to use hash matching to detect it. Our guide to [image-based abuse on dating sites](/guides/image-abuse-dating/) sets out who is in scope. This page is general information, not legal advice.

## What tools check who is signing up?

Sign-up tools answer three questions: is this a real person, are they an adult, and have we banned them before. Each needs a different check.

- **Real person.** Email and phone confirmation, a liveness selfie, and signals from the device and network.
- **Adult.** An age check. Our guide to [identity verification for dating sites](/guides/identity-verification-dating-sites/) explains the levels of check and what each one proves.
- **Banned before.** Matching new sign-ups against the devices, payment cards, phone numbers and photos of accounts you have closed.

The third check is the one new operators forget. A ban that the same person can undo with a new email address is not a ban.

## What tools screen photos and messages?

Two kinds: tools that recognise content already known to be harmful, and tools that judge content they have not seen before.

**Hash matching** recognises known images. Microsoft's PhotoDNA creates a digital signature of an image, called a hash, and compares it with a database of hashes of previously identified illegal images. Microsoft says the hash is not reversible, that PhotoDNA is not facial recognition, and that its cloud service is free for qualified customers. StopNCII.org applies the same idea to intimate images shared without consent: the person creates the hash on their own device, and participating platforms look for matches.

**Classifiers** judge new content. Bumble's Private Detector is a public example on a dating app. Bumble says it uses AI to blur a potential nude image sent in chat, tells the recipient, and lets them choose to view it, block it or report it.

**Text screening** looks for the language of scams and pressure: requests for money, investment talk, attempts to move the conversation to another app.

Ask any supplier how its tool performs on dating content in particular. Dating photos include beachwear, gym shots and close crops that a general-purpose filter may flag wrongly. Our guide to [AI moderation for dating sites](/guides/ai-moderation-dating/) goes deeper.

## What tools handle reports and cases?

A case tool takes every report and flag, puts them in priority order, and records who decided what. This is the layer regulators and police will ask about.

A usable set-up has:

1. A report button on every profile, photo and message, with clear reasons to pick from.
2. A way for non-members to report, for example someone whose photos have been stolen.
3. One queue that mixes member reports and automated flags, sorted by severity.
4. A view that shows the moderator the account's history, linked accounts and earlier reports.
5. Set actions: remove content, warn, suspend, ban, escalate to the police.
6. A message back to the person who reported.
7. An appeal route, and a log of every decision.
8. Secure storage for evidence, with a written process for police requests.

Our guide to a [content moderation workflow](/guides/content-moderation-workflow/) covers the process, and [building a trust and safety team](/guides/trust-safety-team-building/) covers the people.

## Should dating platforms share safety signals with each other?

Yes where a lawful scheme exists, because banned people move between platforms. The clearest example is in child safety.

Lantern, run by the Tech Coalition, lets technology companies share signals about online child sexual exploitation and abuse. Signals can be content-based, such as hashes and URLs, or incident-based, such as attempts to move a conversation off the platform. The Tech Coalition says 41 companies take part, that 2,047,982 signals were shared between 2023 and 2025, and that these led to 353,806 enforcement actions. Each company assesses a signal against its own policies before acting. The Tech Coalition is explicit that Lantern does not automate enforcement.

Lantern is open to qualifying technology companies and financial institutions. Match Group, which owns Tinder and Hinge, announced in May 2023 that it had joined the Tech Coalition.

For an independent dating brand, the practical version of signal sharing is a shared network. When one platform runs many brands, a person banned on one brand is known on the others.

## Should you build, buy or inherit the stack?

Buy or inherit unless safety tooling is what your company is for. The detection layers need training data, constant updating and specialist staff, and a new dating brand has none of these.

| Route | Suits | Watch for |
|---|---|---|
| Inherit from a white label or app platform | New brands and media owners | You depend on the platform's standards, so ask to see them |
| Buy specialist tools and join them up | Operators with their own software and a moderation team | Integration work, per-check fees, several contracts |
| Build your own | Large groups with data and engineers | Cost, and the time before it works well |

Questions for any supplier:

- What does the tool detect, and how was it tested on dating content?
- What happens to a flagged item: is it blocked, blurred or queued for a person?
- How often are flags wrong, and how do you measure that?
- Where is member data processed and how long is it kept?
- How is it priced: per check, per member or flat?

## What does a white label platform provide?

It provides most of the stack as part of the service, with the same tools and moderators working across every brand on the network. You keep layer six: knowing what is done in your brand's name.

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 handles moderation and support for partner sites, and is for dating websites only.

DatingAPI, also part of our group, runs safety and moderation, and ID and age checks, as part of its platform for dating apps.

Our [safety and moderation](/platform/safety-and-moderation/) page explains what our platform does. If you run your own software, a white label is not a way to add tools to it; you would buy specialist tools instead.

## Common questions

### Can a small dating site use hash matching?

Yes. Microsoft says its PhotoDNA Cloud Service is free for qualified customers, including technology companies and developers, and organisations apply through a form. Ask your moderation supplier whether it is already built in.

### Is hash matching the same as facial recognition?

No. Microsoft says a PhotoDNA hash cannot be reversed to recreate an image, and that PhotoDNA is not facial recognition software and cannot identify a person or object in an image.

### Do automated tools remove the need for moderators?

No. Tools sort and flag; people decide. Ofcom's hash matching measure, for example, expects human moderators to review an appropriate proportion of what is detected.

### What is the first tool a new dating site should have?

A report and block button on every profile and in every chat, feeding a queue that someone checks every day. Ofcom's codes also expect every service to name a person accountable for online safety.

### How should we measure whether the stack is working?

Track time from report to decision, the share of harmful accounts caught before a member reports them, how often automated flags are overturned on review, and how many banned people get back in.
