What a Take-Home Task Checks That Your Resume Can't

Everything on your resume is self-reported. A reference is somebody else’s account of you, and an assessment is the employer generating its own evidence directly — the only stage of a hiring process where nobody is relying on a description of your work. That is why the step exists, and it is why the task you are given usually maps onto a specific line of your application.

Which means the first useful question about any take-home task or timed test is not “how do I do well at this” but “which claim of mine is this checking”.

Find the claim it is testing

Read the task, then read your own resume, and identify the line it corresponds to. It is rarely subtle. A task about restructuring a dataset is testing the tool you listed. A written exercise is testing the “stakeholder communication” phrase. A case is testing the judgement your bullets claim you exercised.

Once you can name the line, you know what a good submission has to demonstrate, and you know whether you are in trouble. If the claim was generous — a tool used once, a skill you can talk about but not perform — the assessment is where that surfaces, and it surfaces in a form you cannot talk your way past. This is the practical cost of listing a skill you barely know: the bill arrives as an afternoon of work you cannot complete.

If you find yourself in that position, say so before you submit rather than after. “I have used this at the level of X rather than Y” costs you the role you were not going to get and preserves the one you might.

Scope it before you start

Ask three questions by email, and ask them all at once so it reads as diligence rather than negotiation:

• Roughly how long should this take?
• What does a complete submission look like — a document, a working thing, a set of recommendations?
• Will there be a conversation about it afterwards?

The answers change the work substantially. A task expected to take two hours and a task expected to take a weekend are different assignments even when the prompt is identical, and candidates routinely spend a weekend on the two-hour version and are marked down for over-engineering it.

If nobody answers, pick a bound yourself, state it in your submission, and stop when you reach it. A note saying what you would have done with more time is a legitimate part of the deliverable and often reads better than the extra work would have.

The AI question, stated plainly

The instructions may say no AI, may say use whatever you like, or — most often — may say nothing.

If it says no, that is a rule and not a suggestion. It is the same category of thing as any other stated condition of an application: whether AI assistance is acceptable is the employer’s call to make, and it varies by stage. Ignoring an explicit instruction is a different act from using a tool where none was mentioned, and it is the one that ends processes.

If it says nothing, work to this standard: the submission has to be work you can account for line by line, out loud, without notes. Not because someone will catch you otherwise, but because the follow-up conversation is where the assessment actually happens, and an artefact you cannot explain is worse than a simpler one you can. Under that standard, using a tool to check your spelling or reorganise your own prose is unremarkable, and using one to produce an approach you do not understand is self-defeating.

The failure mode is specific and it is not detection. It is sitting in a review conversation being asked why you chose this structure over that one, about work you did not choose, and having nothing to say. That is the same exposure as a resume bullet you cannot back up, compressed into a single meeting.

Timed tests are a different animal

A timed online test — no lookups, a countdown, fixed questions — measures something narrower: recall and speed under constraint. Preparation for it is almost entirely about the format rather than the content. Find out how long, how many questions, whether you can skip and return, whether wrong answers are penalised, and whether it is one sitting. Do a practice run of any sample provided.

Then treat it as an exam: answer everything you know quickly, flag rather than fight the ones you do not, and stop refining. Most people lose more marks to unfinished sections than to wrong answers.

When to decline

Some tasks are worth pushing back on, and pushing back politely is not a withdrawal:

• Unbounded scope. No time guidance, no definition of done, and no answer when you ask.
• Work that is obviously production work. A specific deliverable the business needs, for a real client, with real data. That is a job, not an assessment.
• A task before any human conversation. Reasonable for some volume processes, but you are being asked to spend hours on a role you know nothing about, and you may reasonably ask for a short call first.
• Anything requiring you to breach a current employer’s confidentiality. You cannot submit their material, and you should not describe it in detail either. The rule that applies to writing about work you cannot describe applies here too, and a good employer accepts a sanitised or invented-context version.

Declining costs you that application. Doing three days of unpaid work for a process that goes quiet costs you more.

Prepare for the conversation, not just the submission

Assume you will be asked to walk through what you sent. Before the meeting, write down for each significant decision: what you chose, what you rejected, and why. Then say it out loud once — rehearsing the walkthrough is one of the few things a model set up as an interviewer is genuinely useful for, since it will keep asking “why that way” until you have an answer.

Submit on time, with a short covering note stating your assumptions, your time bound, and what you left out deliberately. That note is part of the evidence, and it demonstrates the thing most tasks are actually looking for.