Which tool to choose
You don't need a specialized tool to generate the questions. You only need a generic conversational assistant, because the job is "read this description and produce targeted questions."
- ChatGPT free version is the right choice for 90% of those who hire occasionally. Even without a subscription you can search the web, upload files and images and have them analyzed: you can therefore upload the PDF of the posting or the resume. The limit: with a free account the best model has a message cap within a window of a few hours, after which the chat switches to a lighter model until the reset. To prepare an interview it's more than enough.
- If you hire often and as a team, consider the paid team plan: it costs about $20-30 per seat per month depending on billing, offers workspace management, higher limits than a single account, and the guarantee that your data won't be used for training. That last point matters when you paste resumes with personal data.
- Tools dedicated to interviews (Kickresume, Interviews.chat, and the like) exist, and most focus on the candidate side, while some also help recruiters prepare more targeted interviews. They add, however, a cost and one more account to do what a generic assistant already does well. Skip them, unless you also want to let candidates practice.
This guide continues on the free version of ChatGPT.
How to do it
The typical mistake is asking "give me 10 questions for a warehouse worker interview." You get the usual magazine list. The method that works starts from the skills and produces behavioral questions, because skills-based interviews start from the idea that the best predictor of future performance is past performance.
Whether in a browser or an app, the path doesn't change:
- Open ChatGPT and create a new chat.
- Keep within reach the role description (the posting you published) and the list of the 4-6 key skills (the concrete abilities the job requires: e.g. "handling difficult clients," "meeting deadlines").
- Paste the prompt below, replacing the parts in quotation marks with your real data.
- Read the generated questions and discard the generic ones.
- Ask the AI to add, for each question, what to look for in a good answer (the evaluation grid).
The operating syntax:
Act as an expert recruiter. I need to prepare a structured
interview for the role of "sales assistant in a clothing
store," 20 hours a week.
The skills I want to evaluate are:
1. handling unhappy customers
2. teamwork during busy periods
3. accuracy in handling the cash register and returns
4. ability to learn new procedures quickly
Generate 2 behavioral questions for each skill, in the format
"Tell me about a time when...". For each question add:
- what should emerge in a strong answer
- a red flag in the answer
- a possible follow-up question
Use exclusively job- and skill-related questions.
Do not include questions about family, age, marital status, health,
religion, origin, or orientation, because in Italy they are prohibited.
Check: you should get 8 questions, each with its grid. If the questions are still vague ("Are you a precise person?"), reply to the AI:
These are too generic, rephrase them by asking for concrete
episodes from the candidate's past.
The good question forces the candidate into a story, not an adjective.
Keep in mind the right measure: most skills-based interviews include between 4 and 8 questions, each aimed at a specific skill. Don't ask for thirty.
A concrete example
Marta runs a small pastry shop in Bologna and needs to hire a counter clerk. She has never done structured interviews. She opens free ChatGPT and pastes the prompt above adapted: role "sales counter clerk in a pastry shop," skills "managing the queue on crowded weekends," "recommending products to the customer," "hygiene and tidiness of the counter," "handling complaints."
The AI returns questions like "Tell me about a Saturday morning when there was a long line: what did you do to keep customers from getting annoyed?", with alongside it what to look for ("prioritizes, stays calm, communicates timing") and a red flag ("dumps the blame on colleagues or customers").
Marta then asks: "Add a 1-to-5 grid to give a score to each answer." She prints the sheet, uses it identically with the three candidates, and at the end compares the scores instead of the impressions. She hires the candidate with the highest score on complaint handling, the most critical point of her counter. This is the real advantage: a process in which every candidate for the same role receives the same predetermined questions and is evaluated with a consistent grid reduces recruiter bias and improves the ability to predict job performance.
When it does NOT work (and how to fix it)
If the AI proposes illegal questions
It can happen that it generates questions about availability to work overtime "even with young children" or about origin. In Italy these are prohibited: Article 27 of the Equal Opportunities Code prohibits, during an interview, asking about marital status, the presence of children, or the intention to have them. The same goes for origin, faith, and political ideas: Art. 8 of Law 300/1970 prohibits the employer from investigating the worker's political, religious, or union opinions. The way out: always reread the list with a critical eye and delete any question that doesn't directly concern being able to do the job. There is no statutory list of prohibited questions; they are prohibited as a consequence of the prohibition of discrimination. If a question doesn't help you understand whether the person will do that job well, don't ask it.
If the questions are all the same and predictable
It means you gave the AI little information. Don't write "sales assistant" and stop: paste the entire posting, describe what the day is really like, what specific problems that role creates. The more detailed the job description, the more the AI can calibrate the questions on the real skills and requirements of the role.
If you don't know how to evaluate the answers you'll receive
Ask the AI to build the grid for you before the interview, with the behaviors that distinguish a 5 answer from a 2:
For each question, create a 1-to-5 evaluation grid.
Describe in words what characterizes a 5 answer, a 3,
and a 1, referring to concrete behaviors and not impressions.
It works because the most solid organizations do exactly this: the questions arise from analysis of the role, are evaluated with behavioral anchors, and are documented in a way that holds up to a possible audit.
If you fear that pasting resumes violates privacy
On the free plan your inputs can be used to improve the models. Solution: don't paste names, contacts, or identifying data of candidates. To generate the questions you need the role description, not the person's resume. If you really have to analyze real profiles on an ongoing basis, that's the case where the paid team plan is justified, for the data guarantee seen above.
A tip from someone who really uses it
Use the AI for the questions, but deliver the same opening in person to everyone and take notes during, not after. The value of the structured method lies in the comparison, not in the single brilliant question: when every candidate faces the same questions in the same order, likes and first impressions have much less room to override the evaluation, and together with a good grid you get comparable data. Keep a "quick notes" column next to each question: at the end of the day, with four candidates behind you, you won't remember who said what.
Frequently asked questions
How many questions should I prepare for a one-hour interview?
Aim for 4-8 behavioral questions plus two or three follow-ups each. Each answer generally takes 2-3 minutes, which brings a complete interview to last between 30 and 60 minutes. If you prepare thirty, you end up doing a rapid interrogation instead of an interview.
Can I have the questions generated directly from the posting file?
Yes. In the free version you can upload the PDF or the document of the posting and ask the AI to start from there. It's also more precise than rewriting by hand: the AI reads the exact requirements you published.
Do I have to tell the candidate that I used AI to prepare the interview?
No, and there's nothing improper about it. Preparing the questions with an assistant is like using a structured interview template taken from a manual: the tool helps you be fairer and more consistent, it doesn't replace your judgment. The decision stays yours. It would be different to have the answers evaluated by an algorithm in your place: there you'd enter delicate territory of transparency and personal data that, for those who hire occasionally, isn't worth crossing.