> ## Documentation Index
> Fetch the complete documentation index at: https://docs.staffer.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Run AI interviews that are worth doing

> Write the two instructions that decide what an AI interview covers, pick the right method, and read the results well.

An AI interview is only as good as what you tell the interviewer. There's no question bank: two free-text boxes decide what the conversation covers. This guide covers how to fill them, which settings matter, and what to do with the results.

For every field and setting, see [AI interviews](/companies/ai-interviews).

## Decide what the interview is for

Pick one job for the interview to do. Good ones:

* Check the two or three criteria that profiles rarely prove, such as how someone approaches a messy problem.
* Hear how a candidate explains their own work, before a person spends 30 minutes on a first call.
* Give every applicant the same first conversation, at a time that suits them.

If you can't say in one sentence what you want to learn, the interview will wander.

## Write the instructions for the AI

**Instructions for the AI** is what you'd tell a human interviewer before the call. Only the AI sees it. Include:

* The two or three topics to cover, in order.
* What a strong answer sounds like, so it knows when to dig.
* When to push back, for example "if an answer stays general, ask for a specific project".
* What to avoid, such as salary negotiation or questions about personal circumstances.

Example:

> You're interviewing for a senior frontend role. Cover three topics: a component library they built or maintained, how they handle accessibility, and a disagreement with a designer and how it ended. For each, ask for one concrete example. If the answer stays general, ask what they personally did. Don't discuss salary or notice period.

## Write the instructions for the candidate

**Instructions for the candidate** is shown before they start. Set expectations: how long it takes, what it covers, and that there are no trick questions. Candidates who know what's coming give better answers.

> This is a 15-minute conversation about your recent frontend work. We'll ask about a component library, accessibility, and working with designers. Take your time. There are no trick questions.

## Pick the method

* **Chat** suits written roles and candidates who prefer to type. It doesn't record audio.
* **Voice** is closest to a phone screen. Pick a **Voice** and a **Speaking speed** that sound natural in the **Language** you chose.
* **Video** adds a camera recording on top of the conversation. Use it only if you'll actually watch it.

AI interviews run in English (US) or Norwegian.

## Set the length and tone

* **Max length (minutes)** defaults to 15. When the time is up, the interviewer wraps up. Keep it short enough that three focused topics fit comfortably.
* **Interviewer tone** has three options: **Warm** (curious, encouraging), **Neutral** (professional, structured) and **Rigorous** (probing follow-ups). Warm is the default and suits most first conversations. Rigorous suits technical deep dives.

## Test it yourself

Click **Try as candidate** and do the interview as if you were applying. It runs with your current settings, including unsaved changes, and nothing is kept afterwards. It does use credits. Adjust the instructions until the conversation covers what you need, then click **Save** on the **Process** tab.

## Read the results

When a candidate finishes, their **Process** tab shows the transcript, the recording for voice and video, and an evaluation with a summary, strengths, concerns and a recommendation of **Advance**, **Uncertain** or **Reject**.

Treat the evaluation as a second opinion. The candidate's match score doesn't use it: the score reads the transcript itself, as evidence against your criteria. The decision is yours. See [Move candidates through the process](/companies/run-the-process).

## Be straight with candidates

Before they start, candidates have to accept a notice that the session is recorded, that a transcript is kept, and that an AI evaluation is produced for your team, which they don't see. Recordings, transcripts and evaluations are deleted six months after the application closes.

Mention the AI interview in your first message too, so it doesn't come as a surprise.