Why Do You Want to Work at Anthropic? (A 200 to 400 Word Answer)

Anthropic asks this question in writing on the job application, and the form states that great answers are often 200 to 400 words. The length is an instruction, not a hint. Write a short essay.

The field on the application reads close to this:

Why do you want to work at Anthropic? (We value this response highly. Great answers are often 200 to 400 words.)

Two things follow from that wording.

First, a person reads the answer. It is part of the screen, not paperwork.

Second, Anthropic's candidate AI guidance asks you to draft this response yourself, then refine it with Claude. The first version should be your own words.

That order is worth following. Reviewers read a very large number of these answers. Text drafted by a model reads as generic, because a model can only write what any applicant would write.

What the reviewer is checking

What they look for	What that looks like in your answer
Specific interest	You name something you read, used, or built
Mission understanding	You describe the work in your own words
Role fit	Your skills connect to the job you applied for
Honesty	Plain sentences, real reasons, no marketing voice

There is one useful test. Could you swap in the name of another AI lab and send the same paragraph? If yes, the answer is not finished.

A structure that fits the word count

Opening, two or three sentences. The concrete reason this company is on your list.
Middle, two short paragraphs. What you have read or used, and how it connects to your own work.
Close, one paragraph. What you would contribute, and what you want to learn.

A full sample answer

This sample runs about 270 words, inside the range the form suggests. Replace every bracket with your own facts.

I want to work at Anthropic because safety and capability are treated as one job here rather than two teams in tension.

I have used Claude almost every working day for [your real use, for example reviewing changes to a large service]. The benchmark numbers are not what convinced me. What convinced me is that the model says when it is unsure, and explains a refusal instead of going quiet. That is a deliberate product choice, and it matches how I like to build things.

I then read Anthropic's published work on interpretability and on responsible scaling. [Name the one you actually read, and say one thing you took from it.] Both treat safety as engineering with tests and commitments attached. I have spent [number] years on systems where a silent failure costs real money, so that framing is familiar.

My own strength is [your strongest area, for example data pipelines, evaluation tooling, or developer platforms]. On [one project] I [one concrete result, with a number]. I care about the parts nobody demos: tests, logs, and making the next person's change safe.

What I want from the role is narrow. I want problems where being careful is the point, and colleagues who will tell me when I am wrong.

I am not applying to every AI lab. I am applying here because the stated mission and the published work agree with each other. I would rather be one careful engineer on that problem than a fast one somewhere else.

Why that answer works

It names one real use. A daily use with a named task cannot be copied by another applicant.
It names published work. Reading one paper and saying what you took from it beats listing three titles.
It carries a number. A result with a number is checkable. A claim without one is decoration.
It says what the writer wants. Wanting correction from colleagues is a specific request, not praise.
It answers "why here". The last paragraph rules out the other labs on purpose.

Mistakes that make a reviewer skim

Praise with no object. "A leader in AI" fits every company in the field.
Repeating the website. Quoting the mission proves you can read, not that you agree.
Only prestige or pay. Wanting strong colleagues is fine. It cannot be the whole answer.
No mention of safety. An answer only about scale misses what makes this company different.
A fully generated answer. Refining your own draft with Claude is allowed. Starting from a generated draft is not, and it reads that way.

If you are asked the same question out loud

Keep the spoken version to 60 to 90 seconds. Use one reason, one link to your own work, and one thing you want to build. We cover the live version in answering this question in the interview.

How to Prepare

Write the first draft in one sitting, then cut it. Aim for 400 words, then remove anything that could apply to any lab.
Read one Anthropic paper or policy properly. One source you can discuss beats five you can name. The company culture, mission and values page is a useful starting point.
Use the short field version if the form is small. Some applications give a smaller box. See the short "Why Anthropic?" answer guide.
Know what comes after the form. The round by round interview process shows where this question returns.
Practise the other behavioural questions too. Work through Grokking Modern Behavioral Interview and the common Anthropic behavioural questions.
Read it aloud before you submit. If a sentence sounds like a press release, rewrite it as you would say it.