For employers
How to write a great task brief (for employers)
3 min read · Updated
The quality of the work you get back depends heavily on the brief you write. Vague briefs lead to slow applicants, mismatched expectations and rework. A clear one attracts people who can do the job and lets them start with confidence. This guide gives you a structure you can follow for any task or project on Chakri.ai.
Start with the outcome
Open with one or two sentences describing what you need and what it will be used for. "Rank 500 chatbot answers about personal finance for helpfulness and accuracy so we can evaluate a model" is far more useful than "Need AI rating help". Workers who understand the purpose make better judgement calls on edge cases.
Describe the data and the volume
- What the data is: text, images, audio, code or documents.
- How much there is and in what batches.
- Language and region requirements.
- Whether any of it is sensitive. Never share personal data, secrets or client-confidential files unless workers have signed the right agreements.
Define what good looks like
Write the quality bar explicitly. Include three to five worked examples: two good, one borderline and one bad, with a sentence explaining each. Examples cut the number of questions more than any other element.
- Label definitions with plain-language rules.
- What to do when the item is ambiguous, unreadable or off-topic.
- Tone, format and length expectations for written answers.
- Safety rules, including content workers should skip and how to flag it.
State required skills and screening
List the skills that truly matter, such as a programming language, a medical specialty or fluency in a given language. Overlong skill lists discourage good candidates, while a missing one invites mismatches. Chakri.ai offers an optional AI interview and screening questions. Use them to test what the work needs, and avoid questions about protected traits such as age, religion, ethnicity or family status.
Set a fair budget and timeline
Pay is the clearest signal of how much you value the work. Estimate the time one item takes, multiply by volume and convert to a fair hourly rate for the skill level you need. Underpaying leads to rushed work and high churn.
- Choose Task for a single well-defined batch, Project when the work has stages, or Job for an ongoing role.
- For Projects, split the budget into milestones tied to a visible deliverable, such as "first 100 items approved".
- Be realistic about the deadline, including time for review and rework.
- Remember that Tasks and Projects are paid into escrow before the listing goes live; this reassures workers and speeds up good applicants.
How acceptance and payment work
Spell out how you will review work and how long you will take, for example "we review within three working days; one revision round is included". Workers value predictable approval. With milestone payments, released funds should match clearly approved deliverables.
A template you can copy
- Goal: one sentence on what you need and why.
- Data: type, volume, languages, any sensitivities.
- Instructions: step-by-step workflow and tools.
- Quality bar: examples and edge-case rules.
- Skills: must-haves and nice-to-haves.
- Timeline: start date, deadlines and review time.
- Budget: total, currency, milestones if any.
- Contact: how and where workers should ask questions.
Common mistakes to avoid
- Posting before you have tested the task yourself on a handful of items.
- Changing the rules midway without telling workers.
- Asking for free sample work that is actually production output.
- Hiding the true workload, such as undisclosed long videos or huge documents.
- Using the AI "Improve description" button without reading the result. It is a suggestion; check it for accuracy before publishing.
Ready to look for work?
Browse live AI training, annotation and RLHF openings, each labeled with its source platform.
Browse jobs