AI tools and original work

Every answer you submit must be your own work, written by you. Don't use ChatGPT, Claude, Gemini, Copilot or any other AI tool to write, rewrite, check or translate your answers, unless a project's instructions explicitly allow a specific tool for a specific step. A confirmed case ends your access on the first occurrence, and the work isn't paid.

The rule

• Everything you submit is your own work, written by you.
• No AI tool (chatbots, code assistants, autocomplete, AI writing, grammar or translation tools) writes, rewrites, checks or translates any part of it, unless a project's instructions allow a specific tool for a specific step.
• Reference material is fine (documentation, textbooks, standards) when you use your own words.
• Don't copy from other sources, other experts or earlier answers unless the task allows it, and never include code or text under a copyleft licence. Where a task allows outside material, name its source and licence in your notes.
• Don't paste task material into any outside tool or service, including AI tools, translators, grammar checkers and browser extensions.
• Don't share tasks, answers or instructions outside the platform, and don't let anyone else work under your account.

You accept the original-work policy (https://canonset.com/policies/original-work) when you create your account, and it is part of the expert agreement.

Why clients need original work

AI labs pay for human judgment. An AI-written answer teaches their models nothing new and can reinforce their mistakes, so this is the rule we enforce most strictly. A client can return a delivered answer that reads as AI-written or copied within 14 days of delivery, and every delivery's datasheet tells the client that its experts accepted the expert agreement: their own work only.

How we check

Every answer is reviewed by a person. Reviewers also see signals about the work itself: the time it took compared with other experts, hidden known-answer tasks, how much of an explanation was pasted rather than typed (pasting code is fine), answers that closely match another expert's, and writing patterns typical of AI models. Where the client allows it, an AI model gives our reviewers a second opinion. A flag is not a verdict: a reviewer looks at the work before anything happens.

The pre-check before you send

On every task, Check before sending shows you the same signals a reviewer would see as flags, as warnings with what to do:

• The main text is shorter than the task type's minimum.
• Writing patterns typical of AI models, with the phrases quoted.
• Mostly pasted text, with the count of what was pasted.
• A close match with another expert's answer to the same task.

Each warning takes a line, If you're sure, say why, which the reviewer reads beside the flag. Sending with warnings is allowed.

What a confirmed breach costs

A confirmed case ends your access on the first occurrence, and the work isn't paid. On an hourly project this is the only kind of rejection that isn't paid. The decision is final: you can't apply again. If we later confirm that work already paid broke the rule, we may deduct that amount from future payouts or ask you to repay it.

If you think a decision is wrong, email support@canonset.com with your drafts or notes; that's why we suggest keeping them for longer tasks. A person answers within one business day. An answer rejected for breaking the rules has no second look by another reviewer: it goes to our team.

Questions: support@canonset.com.