Hiring for written communication: interview tasks that actually test it

A polished cover note predicts very little about how someone writes when a decision is waiting. Elif wrote the best application we received and could not write an incident note.

The sample that fooled us

Her application was clear, warm, and structured, and it took her a week to compose. We read it as evidence of a skill we would use every day.

In her first month she wrote an incident note that four colleagues read twice and still could not act on. The sentences were elegant and the ask was missing.

The problem was not her ability. It was our test, which measured careful writing produced with unlimited time and no reader waiting.

Polished is not clear

Polish is mostly revision, and revision is mostly time. Clarity is a decision about what the reader needs in the first line.

A candidate with a week and a thesaurus can produce prose that hides a weak structure. A candidate with forty minutes shows you the structure itself.

We had been grading the surface of the writing. The job requires the skeleton, usually under mild time pressure and without an editor.

Use a real message from the job

The best task we tried was a genuine message pulled from last quarter, with the names and dates changed. It asked a colleague to reorder two priorities.

Candidates had twenty-five minutes to write the reply. The task needed no invention, which removed the advantage of people who enjoy writing exercises.

It also produced comparable answers. Every candidate faced the same conflict, so differences in the responses came from judgement rather than from the prompt.

Give the reader a decision

A good task forces a recommendation. Ours asked which of two launches should slip, with one paragraph of reasons and one of consequences.

Half the candidates described the trade-off beautifully and never chose. That is the failure mode the job punishes most, and it is invisible in a cover note.

We scored the choice first, the reasoning second, and the prose last. Before that change, the prose had been doing all the work.

Time-box the task

We now say plainly that the task is worth forty minutes and that we will not reward extra effort. Most candidates stop at forty minutes and say so.

The honest ones also tell us what they would have added with more time. That sentence is often the most informative part of the submission.

A take-home with no time limit selects for people with free evenings. That is a different filter from the one we claimed to be running, and we never stated it.

Score the structure, not the vocabulary

Our rubric has four lines, and none of them mentions style. Does the first sentence state the ask, are the reasons ordered, is anything missing, is it short enough to read once?

Vocabulary appears nowhere on the rubric at all. A candidate writing in their second language can score full marks with plain, well-ordered sentences.

Since adopting the rubric, our scores agree far more often between reviewers. Disagreement before was almost always about style rather than substance.

The ask belongs in the first line

Two of three candidates buried the request in the third paragraph, after context the reader already had. That habit costs a distributed team hours each week.

We now ask candidates to rewrite a message so the request appears first. The exercise takes ten minutes and reveals more than a portfolio.

Rewriting is also closer to the real job. Most working writing is repair, not composition, and repair is where weak structure becomes obvious.

Test rewriting, not only writing

Give them a bad message and a reader who needs to act on it. The good candidates cut, reorder, and add one number that was missing.

The weaker ones improve the tone and leave the structure untouched. That is a recognisable pattern, and it is hard to disguise under time pressure.

We keep both tasks now, one written from scratch and one repaired. Together they take an hour and replace two rounds of conversational interviews.

What a bad task reveals

A vague prompt produces vague answers, and then the panel argues about taste. We spent a year doing that and mistook the argument for rigour.

The tell is a submission that could have been written without reading the brief. If the answer ignores the constraint, the task was not testing anything.

We now test every task on two colleagues before using it. If their answers are not clearly better or worse, the task is the problem.

Reviewers need a rubric too

Two reviewers score independently before discussing. Our first attempt skipped that step and the louder reviewer won, which is not a hiring method.

Independent scores also expose a bad task quickly. When four reviewers disagree on nearly every submission, the prompt is usually the cause rather than the candidates.

We keep the rubric to one page and rewrite it after every round. A rubric nobody revises slowly stops describing the job.

The native speaker advantage is smaller than assumed

Our strongest written submission came from a candidate who had never lived in an English-speaking country. Her sentences were short and her asks were explicit.

Native fluency helps with idiom and hurts with brevity. Long, fluent, hedged paragraphs read as competent and cost the reader the most time.

We now tell candidates that plain sentences score higher than elaborate ones. Saying it out loud removed a bias we had been quietly applying.

What the task cannot measure

It cannot measure whether someone will write when they are tired, annoyed, or convinced nobody will read it. Those conditions produce most of the writing.

It also cannot measure judgement about when to write at all. Some conversations should be calls, and a good writer knows which ones.

So we still check one thing in references: does this person write things down before they are asked? That question has predicted more than any exercise.

A week of revision tells you about a candidate’s patience, not their working prose. What would your task show if the deadline were forty minutes?