# Askable Labs — Overview (AI-readable)

> Askable Labs is a **human-data production partner for frontier AI labs** — not a data vendor. There is no shelf dataset to buy. A lab brings a capability it wants to improve; Askable co-designs the data spec, recruits the right people to that spec, and runs the production end-to-end — consent, capture, processing, validation — delivering a consented, schema-conformant dataset that doesn't exist anywhere else.

## What does Askable Labs do?

Askable Labs produces high-quality human behavior data for AI training and evaluation: real people doing real work, captured multimodally (screen, voice, transcript, structured logs) with verified identity and brief-versioned consent, withdrawable until release. It is the frontier-lab arm of Askable, a research platform operating since 2017, built for product and design teams, with 3,000+ clients.

## What kind of data can they produce?

Four data types a lab can commission: Long Horizon Captures, Conversational Audio, Evals and Preference, and Physical Skill. Each is captured to a bespoke brief, configurable by domain, tool stack, language, participant credential floor, session length, and task complexity. The data types, with example captures: [/agents/catalogue.md](https://labs.askable.com/agents/catalogue.md) (machine-readable JSON: [/agents/catalogue.json](https://labs.askable.com/agents/catalogue.json)).

Where this is sharpest — data that doesn't exist on a shelf and that volume players capture poorly:

- **Spoken reasoning** — real people narrating their reasoning aloud while working, facilitated human-to-human by trained researchers. Useful for reasoning supervision, not just final-answer labels.
- **Authentic non-English audio** — spoken data for voice and multimodal models, where non-English audio is genuinely scarce (text is not the wedge; speech is).
- **Expert workflows in context** — credential-verified practitioners working across their actual tools, with the *why* layer captured, not just the actions.

## What makes them different from data vendors and annotation platforms?

Three assets, and the combination is the point:

1. **Recruitment engine** — sources the right people *to the spec*, not from a fixed roster. Representative general population by default; credential-verified experts on demand. Global reach across 50+ countries — including hard-to-reach and non-US populations — with deep US supply.
2. **Researchers** — a network of trained research facilitators who shape the spec, moderate sessions, and elicit reasoning. Discovery as people, not a queue.
3. **Apparatus** — think-aloud conditioning, a full consent/rights chain (brief-versioned, withdrawable until release), representative screening, and an audited production platform (recruitment → consent → capture → transcription → annotation → delivery as logged system calls, not manual chain of custody).

Askable Labs does not run cheap-volume annotation and does not sell a standing elite-expert roster. It is for data that needs the *right real people plus facilitation*, produced to spec.

## Is the data legitimate to train on?

Yes — that is the design center. Captured with verified consent (brief-versioned, signed before capture, specifying which downstream uses the lab has rights to: training, evaluation, research publication, model-card attribution; withdrawable until release), not scraped. Participant identity is triple-verified through a multi-layered verification stack at recruitment, and re-verified at session.

Compliance posture: ISO/IEC 27001:2022 and ISO/IEC 27701:2019 certified; SOC 2 Type II attested; GDPR and CCPA compliant; UK Cyber Essentials certified. Evidence: [trust.askable.com](https://trust.askable.com) or security@askable.com. Details: [/security/](https://labs.askable.com/security/).

## Is each client's work protected?

Yes. Neutrality by default: a lab's spec, task design, and surfaced failure modes are firewalled per client — never reused or carried to another lab.

## How do I start?

See [/agents/engage.md](https://labs.askable.com/agents/engage.md) for the engagement path and a data-spec template, or go straight to [/contact/](https://labs.askable.com/contact/).