What gets contributors removed

Platforms publish the rules that end contributors’ work, and the rules differ from one platform to the next. This lesson sets out the rules, platform by platform, from each platform’s own policy pages (Mercor, Account Security and Conduct) (Outlier Community Guidelines) (Handshake AI, LLM prohibition policy). Read the linked page for any platform you work on: it is the version that binds you, and it can change.

Using AI tools to do the work

Mercor allows AI tools for checking grammar and word choice, and for subtle changes to phrasing, provided the ideas and structure stay your own. It prohibits using them to evaluate the quality of model responses, to write justifications or explanations, or to predict how code will behave, and says submissions evidently produced by AI will be flagged (Mercor, LLM Usage Policy).

Handshake AI is stricter. It prohibits AI tools for developing or refining prompt ideas, generating or editing content and solving tasks, and also for grammar correction, stylistic improvement or formatting, unless Handshake has authorized a tool in writing (Handshake AI, LLM prohibition policy).

So a grammar checker that is allowed on Mercor is prohibited on Handshake AI (Mercor, LLM Usage Policy) (Handshake AI, LLM prohibition policy).

Handshake AI also says prompts taken from Khan Academy, Chegg, textbooks or your own annotation work on other platforms do not count as original, and requires you to keep your drafts, notes and other preparatory materials for at least 90 days after delivery (Handshake AI, LLM prohibition policy).

Outlier’s guidelines ask contributors to complete every task themselves, without help from other people, bots or scripts, or third-party materials or tools, unless an administrator requires otherwise (Outlier Community Guidelines).

Prolific asks participants to complete studies themselves, without automated tools or scripts, and to follow each study’s instructions on when external sources are allowed (Prolific, authenticity checks).

Assessments you get one try at

DataAnnotation’s Starter Assessment can be taken only once, with no retakes (DataAnnotation FAQ).

Most Mercor assessments allow a single submission. Unless an assessment says otherwise, Mercor prohibits using AI to generate answers, copying from outside sources, sharing answers or working with others, and posting assessment content publicly. Some assessments disable copy and paste (Mercor, Assessments).

Outlier requires contributors to complete the qualification tests a project calls for, and to work only on tasks whose required informational materials they have read (Outlier Community Guidelines).

VPNs and where you say you are

Since July 17, 2026, Mercor blocks sign-ins through VPNs, proxies and other anonymizing networks, grants no individual exceptions, and names iCloud Private Relay among the services to switch off (Mercor, Account Security and Conduct).

Outlier prohibits using a VPN or other technology to mislead it about your location or identity unless an administrator requests it, and prohibits misrepresenting where you are located (Outlier Community Guidelines).

DataAnnotation asks you to turn off any VPN on your computer and phone before identity verification, and warns that a work or university computer or network may route traffic through one without your knowing (DataAnnotation FAQ).

One person, one account

Mercor expects each contributor to run one account, in their own name, and to do the work themselves. It reviews for impersonation, account sharing and one person running several accounts (Mercor, Account Security and Conduct).

Outlier prohibits sharing account credentials, creating multiple accounts, and buying or selling accounts (Outlier Community Guidelines).

Confidentiality

Outlier prohibits disclosing who its customers are, sharing project information such as task instructions or screenshots with third parties or other contributors, discussing projects outside approved channels, and storing project materials on personal devices or personal cloud storage (Outlier Community Guidelines).

Mercor treats project materials, client information and prompts as confidential, and reviews for sharing them outside approved channels or posting them publicly or in third-party tools (Mercor, Account Security and Conduct).

Time you bill

Mercor requires logged hours to reflect active work on assigned tasks, and reviews for long idle stretches with the timer running, time logged on unrelated activity, and hours inflated by low or simulated activity (Mercor, Account Security and Conduct).

Quality

DataAnnotation says that work availability depends on demand for your skills and on your past performance, and that subpar or problematic work may cost you access to future projects (DataAnnotation FAQ).

Handshake AI says each project sets its own quality standards and that fellows may be removed from a project if expectations are not met (Handshake AI Fellowship expectations).

Mercor lists late submissions and substandard work needing several revisions, together with poor communication, as signals for removal on performance grounds. A project lead may first run a trial sprint with set expectations, and the contract can be ended if work does not improve (Mercor, Project Offboarding).

What happens next

Mercor describes escalating steps depending on severity and frequency: reduced or reversed payment for the affected work, removal from projects, and in the most serious cases suspension of platform access (Mercor, Account Security and Conduct).

If Handshake AI concludes that an AI tool was used, the work is rejected and not paid, including the time spent on it, and the contributor may be offboarded from the project or removed from the program (Handshake AI, LLM prohibition policy).

Outlier says breaking its guidelines may lead to removal from tasks, disabled accounts or other restrictions (Outlier Community Guidelines).

Prolific starts with a warning when a submission is flagged, may restrict a participant from studies for a while after repeated flags, and says a researcher’s rejection for a failed authenticity check can be disputed (Prolific, authenticity checks).