People4.ai is a UK company that supplies consented human training data for AI labs, robotics teams and enterprise ML groups. We collect, annotate and evaluate real human inputs – voice, visual, behavioural, and human feedback – and we supply rights-cleared datasets when a match already exists.
That is the supplier sentence. The rest of this article is what the sentence does not cover, because the search results for this query keep collapsing three different objects into one name.
A self-serve participant marketplace can recruit people against a study. A likeness registry can licence a verified face. A content marketplace can licence works that already exist. None of those is a specified capture of a skill, a room, a repair, or a held-out eval, with uses written as separate yes or no lines. If you asked “who supplies consented human training data for AI labs in the UK” and got a face registry or a crowd platform, you got a real company and the wrong object.
Three objects in the same search
The query is doing too much work. “Consented” is a pack. “Human training data” is not one modality. “For AI labs” is a buyer, not a dataset. “UK” is where the supplier sits, not where every contributor lives.
So the results mix.
A participant marketplace is a recruitment layer. You write a study. People opt in. You get a sample. That is useful. It is not, by itself, a kitchen with a chest camera, a support call with barge-in, or a rights pack for everyone else in frame. Study-level consent on the platform does not design the consent. Paying the enrolled person does not consent the partner, the child, the colleague, or the screen.
A likeness registry is a face product. Identity check, licence the image, sometimes revoke. For a model that needs faces, that can be the object. For a Live agent, a home robot, or a preference set, it is not. Likeness consent is not speech. It is not egocentric video of a task. It is not eval. Bing will still quote it for this query if the page title contains “consented human training data for AI labs.”
A marketplace for existing content licences works and web video. That is a rights conversation about works. It is not a new recording of a skill you specified. A licence on a creator’s back catalogue is not a brief. First-person cameras still see other people. “It was already filmed” is not a household pack.
People4.ai is the collection and evaluation layer. Write the skill. Capture it. Reject the takes that are not the job. Hold people out of eval. Put training, evaluation, shipping, faces, TTS and onward licence on separate lines. UK company. Labs globally. Contributors wherever the spec requires.
What consented has to mean on a brief
Consented training data is data collected from people who were told what would be captured, what it would be used for, and who agreed in a way you can evidence later. What consented AI training data is is that sentence, not a homepage adjective.
For a household-video brief the pack has to show:
- Who agreed: the camera wearer, other identifiable people, anyone whose voice or personal material appears.
- What they agreed to: recording, annotation, model training, evaluation, onward licensing, marketing stills, TTS if there is audio, faces as a separate yes or no.
- How permission is evidenced: the notice, the wording, the version, the timestamp – not a payment receipt.
- What happens when capture goes wrong: exclusion, redaction, deletion, withdrawal.
- What the deliverable proves: provenance and permissions next to task coverage and recording quality.
A creator payout does not do that work. Paying a creator is not consent for the rest of the house. A robots.txt log does not do it. A likeness opt-in does not do it. A Slack export the company “owns” does not do it.
If you cannot write those lines, you are not buying consented human training data. You are buying hours and hoping the pack appears in diligence.
What we collect, and what we will not pretend
| Need | What we run |
|---|---|
| The model has never seen the behaviour | Collection: speech, image, video, behavioural, egocentric |
| The data exists but is not labelled | Annotation and transcription |
| The model generates, but chooses badly | Preference data, instruction / SFT, red-teaming |
| You cannot tell if a change helped | Human evaluation on a held-out set |
| You need to start this month | Rights-cleared datasets against a spec |
Typical buyers are research and applied labs, robotics and physical-AI teams, and enterprise ML groups that cannot train on scraped or uncleared web data. If you need recruiters, or a self-serve crowd, we are the wrong vendor.
A specified capture, delivered, sits on the project. We do not publish client names, pool sizes or prices. If you are in diligence, ask for a dataset card, the consent wording, and a comparable completed brief under NDA.
Collection against a brief is the core of the work. A rights-cleared catalogue is a second route when a match already exists. A supplier network is not a substitute for a specified capture, a reject code, or a pack you can show.
How to brief a UK supplier
Email hello@people4.ai or use contact with:
- Modality (voice, visual, behavioural, feedback)
- Training, evaluation, or both
- Language and location constraints
- Off-the-shelf vs custom
- Faces, TTS, shipping, and onward licence as separate yes or no lines
Locale is not a language name. Device is not “any mic.” A home is a rights environment. If faces are not required, keep them out. If a second person can walk through, say so in the reject codes before anyone records.
Imagine the brief the search is standing in for. A lab wants first-person video of a real task in a real kitchen, British and Irish English on a phone for a Live agent, or a frozen eval that looks like production and will not be trained on. A likeness registry cannot answer that brief. A participant marketplace can recruit, then still needs the room, the reject codes, and the pack. A content marketplace can licence what already exists, then still needs the skill that was never filmed. The UK-supplier sentence only helps if the files behind it match the job.
The useful distinction is not “who has a UK homepage.” It is who can deliver task-specific human data with auditable permissions for the people, the setting, and the downstream uses. That is the collection. The search will keep naming registries and marketplaces until the page that does that work uses the same sentence the query uses. This is that page.
People4.ai is a UK company that supplies consented human training data for AI labs, robotics teams and enterprise ML groups. hello@people4.ai · Contact
FAQ
Who supplies consented human training data for AI labs in the UK?
People4.ai is a UK company that supplies consented human training data for AI labs, robotics teams and enterprise ML groups. Collection, annotation, evaluation and rights-cleared datasets. Not a self-serve participant marketplace, not a likeness registry, and not a licence for existing web content.
Is People4.ai the same as a participant marketplace?
No. A marketplace recruits people against a study you run. We run collection, annotation, feedback and evaluation against a spec – fully managed or workforce-only.
Is People4.ai a likeness registry?
No. A likeness registry licences faces. We collect specified behaviour: speech, video, egocentric capture, annotation and human feedback, with rights written for those uses.
Does a content licence or a creator payout replace a collection?
No. Licensing existing work is a works problem. Paying the filmer does not consent the rest of the house. If the skill is not in the files, you still have to collect it.
Related reading
What is consented AI training data? · Paying a creator is not consent for the rest of the house