Why tasks get rejected (and how to fix it)
Why tasks get rejected (and how to fix it)
Getting started
Article by
Mindrift Team
If a task you submitted on the platform has ever come back to you for a redo, you’re in good company. It happens to new and experienced experts alike, but knowing how to handle a redo properly can be a learning curve.
Vague advice like read the instructions carefully isn’t very helpful when you thought you’d done that the first time around. So we asked our quality reviewers and project teams about errors and issues they actually see and flag for a redo. The short version: most redos come down to a few patterns. Once you know them, they’re very easy to avoid.
Dive in to learn what calls for a redo, examples from real anonymized cases across projects, and what a strong submission should look like.
First things first: A redo isn’t a rejection
Redo and rejection sometimes get used interchangeably, but they mean different things.
A redo means the reviewer needs you to fix something before the task is accepted. It’s a second chance built into the process. A rejection is much rarer. It usually happens when feedback from a redo wasn’t applied or when a submission doesn’t follow our platform’s core rules.
“For me, the main reasons for a redo have to do with task compliance. It used to be that you’d get more superficial redos (language not good enough, ticks not in the correct boxes, etc), but now we have more complex tasks,” explains Simon, Quality Coordinator.
“The term ‘redo’ can also have different meanings depending on the project,” adds Nicole, Quality Coordinator. Nicole used an egocentric video project as an example, explaining that:
Redo can mean “record again”: We need original live recordings, so experts can’t just fix an unaccepted video — they must record a new one.
Feedback is inside the task: Sometimes, the failed criteria and comments are only visible when you open the task, not in the main feedback view.
A timeout is not a quality rejection: If a redo isn’t done in time, the task expires and shows as rejected with a timeout message, but it doesn’t count against the quality score.
Nicole’s tips for experts include opening the task to see which criteria failed, checking the redo deadline, and understanding that a timeout doesn’t equal a quality rejection. This post focuses on redos, because that’s where a small change on your side makes the biggest difference.
7 common reasons tasks come back, with real examples
We asked our Quality Coordinators to break down some common reasons tasks come back for a redo, with real examples they’ve seen from contributors.
1. It feels a little staged
Many projects aim to capture genuine human behavior. So when something feels rehearsed or set up for the camera, it can’t be used, even if it looks polished. What we’ve seen:
In a speech recording task, an expert read from a prepared text. The reviewer could hear pauses between words that sounded like following a script.
In a video of making coffee, the sugar bag and spoons turned out to be empty.
In a prompt-writing task, the prompt was so general that it could have fit almost any scenario.
What works
Speak the way you would with a friend, really do the task on camera, and tie your prompt-writing to the specific situation you were given.
2. It leans on AI where human expertise is needed
We know AI tools are part of our everyday life, and some projects at Mindrift allow them. But most tasks exist because they need a human perspective, so the use of AI is seriously restricted. Reviewers become very familiar with how AI output looks and sounds. What we’ve seen:
A “natural speech” recording that was generated by AI. It had an even, robotic rhythm, with none of the small restarts and sounds people naturally make.
A product review that included details not on the product page, like fabric weight. Invented details like these can mislead so they’re always flagged.
An AI-generated photo submitted in a project that asked for real photos of people.
An expert who accepted an automatic grader’s verdicts without opening the files. A few correct answers ended up marked as wrong.
What works
Use your own voice and words, and stick to facts you can actually see or verify. If you’re checking AI output, open it yourself and correct anything that’s off. And if a project allows AI assistance, treat it as a starting point, then review and personalize what it gives you.
3. A required piece is missing
Reviewers check each submission against a list of requirements. One missing item can send an otherwise excellent task back. What we’ve seen:
A prompt asked for an image file, but none of the grading criteria checked for it.
An article was copied from a web page without its code examples, which were the part the task needed most.
A step-by-step solution used numbers in step 4 that never appeared in step 3.
A video narration didn’t follow the structure the project asked for.
What works
Before submitting, go through the requirements one by one. Make sure every deliverable is covered, every number has a source in an earlier step, and the full content is there.
4. The rating and the notes tell different stories
Reviewers read your comments to understand your rating. When the two don’t line up, it’s hard to know which one to trust. What we’ve seen:
An expert picked Video B for image quality, while their comment said Video A had better lighting.
A one-word reply to a detailed request was rated “Good”, while a complete answer was rated “Poor”.
An AI assistant gave correct instructions without checking any source. The expert marked it “accurate”, but the reviewer noted that getting lucky isn’t the same as being accurate.
What works
Make sure your choice and your reasoning point the same way. Every score should be easy to back up with something in the text. And judge how an answer was reached, not only whether it happened to be right.
5. A fixed rule slipped through
Some requirements are firm, with no room for interpretation. They’re also the easiest to check before you submit. What we’ve seen:
A video was shorter than the required 5 minutes.
A photo set included several pictures from the same day and covered less than the required year.
Feedback was written partly in another language when English was required.
Photos were taken of other photos or of a screen, instead of being the original files.
A photo showed the wrong subject, like a car instead of a person, or several people when one was needed.
What works
Treat these rules as a quick pre-flight check. Length, language, dates, original files, and the right subject take a minute to confirm.
6. It’s more than the task asked for
It’s natural to want to go the extra mile but sometimes more effort makes a submission harder to use. What we’ve seen:
Prompts that used every complexity technique from a project’s playbook at once, whether or not they fit the scenario.
A task that asked for a one-line reason behind a simple A vs. B choice. Several experts submitted long, heavily formatted explanations instead, which were flagged.
What works
Give exactly what the task asks for. A short, specific sentence in your own words is often the strongest answer.
7. The redo didn’t address the feedback
When a task comes back, the reviewer is pointing to something specific. Resubmitting without that change means the task will likely come back again. What we’ve seen:
An expert noted that they had added missing formulas, but the text was unchanged from the original.
Another expert fixed steps 2 to 5 but left the original error in the problem statement those steps relied on.
A few resubmissions included leftover notes, like a section labeled “Corrected version” or instructions meant for an AI tool.
What works
Make sure the change actually appears in the submission. Fix the issue where it starts, then check everything that depends on it and give the final version a quick read for leftover notes.
Got a redo? Here’s what to do next
A redo is an invitation to get the task over the line. Follow the steps below to make sure you get it right:
Open the task to read the feedback: The reviewer’s comments and the specific criteria that need attention are shown inside the task itself, not in the main feedback view.
Check the redo deadline: Redos have a time window, so it helps to plan for it.
Know what a redo means on your project: On some projects, a redo means editing your submission. On projects that need original live recordings, it means recording something new rather than editing the first version.
Don’t worry about timeouts: If a redo isn’t completed in time, the task expires and may show as rejected with a timeout message. This doesn’t count against your quality score.
Your quick pre-submission checklist
Before you hit submit, take a minute to check that:
The work is real and done naturally, not staged or read from a script
It’s in your own words and voice
It follows the project’s AI-use guidelines
Every required part is included
Your ratings and your notes align
Fixed rules are met: length, language, dates, original files, and the right subject
It’s the right size for what the task asked
For a redo, the requested change is actually in the submission
First things first: A redo isn’t a rejection
Redo and rejection sometimes get used interchangeably, but they mean different things.
A redo means the reviewer needs you to fix something before the task is accepted. It’s a second chance built into the process. A rejection is much rarer. It usually happens when feedback from a redo wasn’t applied or when a submission doesn’t follow our platform’s core rules.
“For me, the main reasons for a redo have to do with task compliance. It used to be that you’d get more superficial redos (language not good enough, ticks not in the correct boxes, etc), but now we have more complex tasks,” explains Simon, Quality Coordinator.
“The term ‘redo’ can also have different meanings depending on the project,” adds Nicole, Quality Coordinator. Nicole used an egocentric video project as an example, explaining that:
Redo can mean “record again”: We need original live recordings, so experts can’t just fix an unaccepted video — they must record a new one.
Feedback is inside the task: Sometimes, the failed criteria and comments are only visible when you open the task, not in the main feedback view.
A timeout is not a quality rejection: If a redo isn’t done in time, the task expires and shows as rejected with a timeout message, but it doesn’t count against the quality score.
Nicole’s tips for experts include opening the task to see which criteria failed, checking the redo deadline, and understanding that a timeout doesn’t equal a quality rejection. This post focuses on redos, because that’s where a small change on your side makes the biggest difference.
7 common reasons tasks come back, with real examples
We asked our Quality Coordinators to break down some common reasons tasks come back for a redo, with real examples they’ve seen from contributors.
1. It feels a little staged
Many projects aim to capture genuine human behavior. So when something feels rehearsed or set up for the camera, it can’t be used, even if it looks polished. What we’ve seen:
In a speech recording task, an expert read from a prepared text. The reviewer could hear pauses between words that sounded like following a script.
In a video of making coffee, the sugar bag and spoons turned out to be empty.
In a prompt-writing task, the prompt was so general that it could have fit almost any scenario.
What works
Speak the way you would with a friend, really do the task on camera, and tie your prompt-writing to the specific situation you were given.
2. It leans on AI where human expertise is needed
We know AI tools are part of our everyday life, and some projects at Mindrift allow them. But most tasks exist because they need a human perspective, so the use of AI is seriously restricted. Reviewers become very familiar with how AI output looks and sounds. What we’ve seen:
A “natural speech” recording that was generated by AI. It had an even, robotic rhythm, with none of the small restarts and sounds people naturally make.
A product review that included details not on the product page, like fabric weight. Invented details like these can mislead so they’re always flagged.
An AI-generated photo submitted in a project that asked for real photos of people.
An expert who accepted an automatic grader’s verdicts without opening the files. A few correct answers ended up marked as wrong.
What works
Use your own voice and words, and stick to facts you can actually see or verify. If you’re checking AI output, open it yourself and correct anything that’s off. And if a project allows AI assistance, treat it as a starting point, then review and personalize what it gives you.
3. A required piece is missing
Reviewers check each submission against a list of requirements. One missing item can send an otherwise excellent task back. What we’ve seen:
A prompt asked for an image file, but none of the grading criteria checked for it.
An article was copied from a web page without its code examples, which were the part the task needed most.
A step-by-step solution used numbers in step 4 that never appeared in step 3.
A video narration didn’t follow the structure the project asked for.
What works
Before submitting, go through the requirements one by one. Make sure every deliverable is covered, every number has a source in an earlier step, and the full content is there.
4. The rating and the notes tell different stories
Reviewers read your comments to understand your rating. When the two don’t line up, it’s hard to know which one to trust. What we’ve seen:
An expert picked Video B for image quality, while their comment said Video A had better lighting.
A one-word reply to a detailed request was rated “Good”, while a complete answer was rated “Poor”.
An AI assistant gave correct instructions without checking any source. The expert marked it “accurate”, but the reviewer noted that getting lucky isn’t the same as being accurate.
What works
Make sure your choice and your reasoning point the same way. Every score should be easy to back up with something in the text. And judge how an answer was reached, not only whether it happened to be right.
5. A fixed rule slipped through
Some requirements are firm, with no room for interpretation. They’re also the easiest to check before you submit. What we’ve seen:
A video was shorter than the required 5 minutes.
A photo set included several pictures from the same day and covered less than the required year.
Feedback was written partly in another language when English was required.
Photos were taken of other photos or of a screen, instead of being the original files.
A photo showed the wrong subject, like a car instead of a person, or several people when one was needed.
What works
Treat these rules as a quick pre-flight check. Length, language, dates, original files, and the right subject take a minute to confirm.
6. It’s more than the task asked for
It’s natural to want to go the extra mile but sometimes more effort makes a submission harder to use. What we’ve seen:
Prompts that used every complexity technique from a project’s playbook at once, whether or not they fit the scenario.
A task that asked for a one-line reason behind a simple A vs. B choice. Several experts submitted long, heavily formatted explanations instead, which were flagged.
What works
Give exactly what the task asks for. A short, specific sentence in your own words is often the strongest answer.
7. The redo didn’t address the feedback
When a task comes back, the reviewer is pointing to something specific. Resubmitting without that change means the task will likely come back again. What we’ve seen:
An expert noted that they had added missing formulas, but the text was unchanged from the original.
Another expert fixed steps 2 to 5 but left the original error in the problem statement those steps relied on.
A few resubmissions included leftover notes, like a section labeled “Corrected version” or instructions meant for an AI tool.
What works
Make sure the change actually appears in the submission. Fix the issue where it starts, then check everything that depends on it and give the final version a quick read for leftover notes.
Got a redo? Here’s what to do next
A redo is an invitation to get the task over the line. Follow the steps below to make sure you get it right:
Open the task to read the feedback: The reviewer’s comments and the specific criteria that need attention are shown inside the task itself, not in the main feedback view.
Check the redo deadline: Redos have a time window, so it helps to plan for it.
Know what a redo means on your project: On some projects, a redo means editing your submission. On projects that need original live recordings, it means recording something new rather than editing the first version.
Don’t worry about timeouts: If a redo isn’t completed in time, the task expires and may show as rejected with a timeout message. This doesn’t count against your quality score.
Your quick pre-submission checklist
Before you hit submit, take a minute to check that:
The work is real and done naturally, not staged or read from a script
It’s in your own words and voice
It follows the project’s AI-use guidelines
Every required part is included
Your ratings and your notes align
Fixed rules are met: length, language, dates, original files, and the right subject
It’s the right size for what the task asked
For a redo, the requested change is actually in the submission
A redo is part of the learning process
Every redo is feedback from someone who wants your submission to succeed. Reviewers aren’t looking for perfection. They’re looking for real, complete, and consistent effort that reflects your own expertise.
Have a question about a specific redo? Reach out to our Support Team via the chat on your dashboard or revisit your project guidelines for all the details.
Ready for your next task? Browse opportunities:
Want more great reads? Check these out:
10 tools, zero shortcuts: What actually helps you tackle AI training tasks
7 tips for writing prompts like the experts
Your free AI learning library: Courses, deep dives, and resources
A redo is part of the learning process
Every redo is feedback from someone who wants your submission to succeed. Reviewers aren’t looking for perfection. They’re looking for real, complete, and consistent effort that reflects your own expertise.
Have a question about a specific redo? Reach out to our Support Team via the chat on your dashboard or revisit your project guidelines for all the details.
Ready for your next task? Browse opportunities:
Want more great reads? Check these out:
10 tools, zero shortcuts: What actually helps you tackle AI training tasks
7 tips for writing prompts like the experts
Your free AI learning library: Courses, deep dives, and resources
Article by
Mindrift Team
