Questions, Topics or Scenarios? Writing AI Interview Content That Gets Real Signal

Should an AI interview use questions, topics or scenarios? Use a question when every candidate must hear the same words, such as a notice-period or eligibility check. Use a topic when you want to test a skill and need the AI to probe with follow-ups. Use a scenario when the candidate needs something in front of them to react to, such as a code snippet, a customer email or a case brief. Most strong interviews mix all three.

The content type you choose decides what kind of answer comes back. A fixed question gets you a comparable answer and nothing more. A topic gets you depth, if you write it well and give it room to probe. A scenario gets you evidence of how someone works, not just how they describe their work. This guide covers when to use each, how to write them, and how to spend a limited follow-up budget where it produces the most signal. If you haven't built the interview yet, start with how to create an AI screening interview from a job description.

Questions, topics and scenarios at a glance

Each content type gives the AI a different job. Here is what changes.

All three can sit in the same interview. Each one counts as a single item, and the interview's total question count is the number of items plus all their follow-ups.

When should you use a fixed question?

Use a fixed question when wording matters more than depth. Three cases fit:

• Eligibility checks. Notice period, location, shift availability, willingness to relocate. You need a clear answer, not a conversation.
• Compliance. Anything that must be said precisely, such as a disclosure or a consent check.
• One anchor question. A single behavioral question you want every candidate to answer in identical words, so you can compare responses side by side. This restores the core of a structured interview inside an adaptive format.

Avoid fixed questions for skills. With no follow-up, a rehearsed or AI-generated answer passes unchallenged. "Tell me about a time you handled a difficult customer" as a fixed question tells you the candidate prepared a story. As a topic, it tells you whether the story holds up when the AI asks what they would do differently.

Two writing rules keep questions clean:

• Ask one thing per item. "What's your notice period and are you open to relocating?" often gets half an answer. Split it into two questions.
• Read it aloud before you save it. If it sounds stiff spoken, it will sound stiff from the AI.

When should you use a topic, and how do you write one?

Use a topic for anything you would normally probe in a live screen: skills, experience, decision-making. A topic is a brief, not a script. The AI writes its own opening question, listens to the answer, then decides what to ask next. That is the point of the format. Two candidates with different backgrounds both get tested on the same skill, through questions that fit their own experience.

The questions the AI asks are only as good as the brief you write. A strong brief covers three things: the skill, the situation it shows up in at your company, and what a strong answer includes.

Keep each brief to two or three sentences. Leave out rules that apply to every item, such as tone or pacing. Those belong in the interviewer's general instructions, covered in the guide to building an AI interview from a JD.

How many follow-ups should each topic get?

Match follow-ups to how much the skill matters. A follow-up is the AI's chance to ask for a specific example, test the reasoning behind an answer, or push into an edge case. Each one also adds time to the interview, so give the most to the skills that carry the most weight.

Recruiters building conversational AI interviews on Equip set an average of 2.5 follow-ups per topic. Treat that as a reference point rather than a target. An even spread is usually a sign that follow-ups were set by default instead of by priority.

The budget is tighter than it looks. A 15-minute Quick Interview on Equip allows 3 to 5 items and 5 to 10 questions in total, so four items leave room for six follow-ups at most. A 30-minute Comprehensive Interview allows 5 to 10 items and up to 25 questions. Either way, spend unevenly. If an item is weighted heavily, give it enough follow-ups to justify that weight. A make-or-break skill scored at the highest difficulty but given one follow-up has a big influence on the final score and very little evidence behind it.

Follow-ups also do two jobs beyond depth. They make cheating harder, because an answer read off a second screen falls apart when the AI asks the candidate to expand on what they just said. (More on why traditional proctoring fails against ChatGPT.) And they affect how structured your interview is. Probing and follow-up are one of the main open questions in structured interview research, according to a review in Personnel Psychology (Levashina et al., 2014). Keeping the follow-up count fixed per topic, for every candidate, keeps the comparison fair even when the follow-ups themselves adapt.

When should you use a scenario?

Use a scenario when the skill is easier to show than to describe. Reviewing code, reading a dashboard, replying to an angry customer email, prioritizing a messy backlog. A candidate can talk well about any of these and still struggle to do them. A scenario puts the material on the candidate's screen in a side panel while the AI discusses it with them.

A scenario has two parts. The candidate view is the material itself: text, a code block, a table or an image. The AI context is optional guidance only the AI sees. Use it for anything the AI cannot work out from the material alone, such as what a diagram shows or which issue you planted in a code snippet.

def get_active_users(users):
    result = []
    for u in users:
        if u["last_login"] > datetime.now() - timedelta(days=30):
            result.append(u)
    return result

Three rules make scenarios work:

• Keep the material short. The candidate should be able to read it in under a minute. A scenario that needs five minutes of reading is a take-home test, not an interview item.
• Set one task. "Review this" or "Reply to this customer" gives the AI a clear thread to probe. Two tasks in one scenario split the answer.
• Write the AI context as what a strong answer notices. That turns the context into scoring guidance as well as background.

Scenarios are also the natural home for testing how a candidate works with AI. Put a realistic task in the candidate view, let them use an AI tool, then use follow-ups to ask how they prompted it and what they changed in the output. The practical guide to AI interviews covers this approach in more detail.

Let candidates write, not just talk

Some answers should not be spoken. Nobody wants to hear a SQL query read out loud. On Equip, turning on the Workspace adds a code editor and a text area to the candidate's screen, and each item gets an answer format: spoken, written, or the candidate's choice. The report then shows the final code alongside timestamped snapshots linked to the recording, so you can see how the candidate got to the answer, not just where they ended up. For more on screening engineers, see Equip's guide to technical recruitment.

Example: mixing all three in a 15-minute screen

Here is a first-round screen for a backend engineer with two to four years of experience. It uses all three content types and stays inside a Quick Interview's 10-question cap.

The notice-period check needs a clear answer, not depth, so it is a fixed question. The incident topic tests the skill that decides the hire, so it gets the highest weight and two follow-ups. The code review is written because you want to see the fix, not hear it described. The API design topic gets one follow-up because it is the lowest priority of the skills, and that leaves one spare slot if you later want to add depth elsewhere.

Mistakes that cost you signal

• Writing topics as labels. "Communication" gives the AI nothing to probe. Write the situation and what good looks like.
• Testing a skill with a fixed question. You get a prepared answer and no way to check it.
• Spreading follow-ups evenly. Two follow-ups on every topic means your most important skill gets the same scrutiny as your least important one.
• Heavy weight, thin evidence. A top-weighted topic with zero or one follow-up lets a single answer swing the score.
• Images with no AI context. If the scenario relies on a diagram, describe it in the AI context, or the AI cannot discuss it properly.
• Asking for code out loud. Use a written answer format for anything where the exact output matters.

Still deciding whether the round needs an adaptive conversation at all? Compare one-way vs conversational AI interviews first.