AI for job interviews: what to check and who decides

AI for job interviews asks initial questions using a script for the role, sets tasks and assesses answers against specified requirements. The manager still needs to clarify points of uncertainty, see the candidate in action and decide whether they are suited to the particular job. A useful outcome of this interview is a set of answers with clear reasons for their assessment, showing what still needs to be discussed.

A business owner wants to spend fewer hours on hiring. That is a reasonable aim: explaining the job’s terms to every applicant and listening to general answers takes a lot of time. But if few people reach the meeting after automated screening, the cause can lie in the screening itself. A requirement turned out to be unnecessary, a question was unclear, or missing information became a negative assessment.

I think the most troubling mistake here is a rejection that nobody noticed. The company will see the person it hires at work. A candidate ruled out by a poor question simply disappears from the list. So you need to review both the candidates AI recommends and the reasons why the others never reached the manager.

What does AI assess in an interview?

AI assesses what the candidate has said or done during the interview: the experience they described, their answers to questions and their solutions to the tasks set. Answers alone cannot establish whether past achievements are genuine or how the employee will perform in future.

Employers benefit from separating three tasks. Gather information about the person. Get their solution to a work issue. Compare the answer with the job requirements. If all this is called “candidate assessment”, the manager will stop distinguishing between an observation and the applicant’s own account.

The convenience of AI lies in the repetitive part of the interview: the same requirements can be applied to different candidates while retaining the content of their answers. But you first need to choose the requirement. If a salesperson will need to find new buyers, an account of serving an existing customer base leaves the question of finding buyers open. Many years of experience do not fill that gap.

Why does a high score not mean someone is ready for the job?

A high score shows how well the answers match the chosen scale. It cannot be read as the probability of a successful hire or permission to give the person all the duties of the role.

Start by finding out what earns the score. Mentioning the required action, explaining its purpose or carrying it out? In an answer about working with a customer base, a candidate lists calls, emails and reminders. This is enough to assess whether they know the sequence. To assess their ability to reach an agreement with a buyer, you need the conversation itself: a general list of actions does not provide that evidence.

Another distinction is between a mandatory requirement and an added advantage. A well-written answer is useful for a salesperson. But if their main work involves negotiations by phone, good writing does not settle the question of their speaking skills. By adding all the criteria into one total, the company itself allows one strength to compensate for the absence of another.

The business owner only needs to make one decision here: which actions are required before the person starts work and what evidence is accepted for each. You should keep the other preferences separate. How requirements narrow the search is discussed in “How to find a salesperson: unnecessary filters”.

Do not set a passing score just to produce a shortlist. First look at the answers that would fall on either side of the threshold. If the difference comes down to text length or familiar terminology, that is what the threshold selects for. The ability to do the work still needs to be established.

How do you prepare an AI interview script?

An AI interview script is built around the actual work: duties, information available to the employee, decisions they are allowed to make and criteria for a suitable answer. The job title alone is not enough: two salespeople have different tasks from the outset if one finds customers and the second handles enquiries.

Before setting up the interview, the manager fills in a short brief. This can be passed to the recruiter along with the job listing:

• Work: which area the person will take on — finding buyers, incoming enquiries or existing customers.
• Independent action: what they must do without the manager’s help from the first stage of work.
• Available information: product description, terms of sale, materials and help from specialists.
• Task: what result the candidate must prepare using this information.
• Answer criteria: exactly what must appear in the result and which actions are unacceptable.
• Unknowns: which conclusions this task cannot support and what a person will add to the assessment.

Then you need the material itself. Take an anonymised enquiry from your company that relates to the salesperson’s future work. Remove contact details, names and confidential terms. Keep the information needed to understand the request. The person preparing the task should try it themselves: otherwise, they may unwittingly require knowledge that only company employees have.

Here is the task text to use with such an enquiry:

Read the buyer’s request and the description of our offer. Prepare a first reply. Separately list what information you lack to propose terms, who you will get it from and what you will tell the buyer before you receive it. You may use the attached materials. Do not add information that is not in them.

This task can show whether the candidate distinguishes between what is known and unknown, connects their questions to the buyer’s request and makes any promises without a basis. It does not show how they conduct spoken negotiations. You should record this limitation straight away to avoid later reading too much into the written answer.

How do you read a candidate’s assessment after an AI interview?

Read the assessment alongside the task and the original answer. For every conclusion, the requirement, the relevant part of the answer and the reason for the assessment must be visible; missing information is recorded as a separate result, without automatic rejection.

Three different situations can easily end up under one label, “does not meet requirements”: the candidate broke a clear condition, did not answer the question asked or was never asked about the required skill. The manager’s next steps differ in each case.

Asking for clarification of the task should not in itself lead to a worse assessment. If a future salesperson is required to understand the buyer’s request first, it is strange to ask them to guess during the interview. At the same time, clarifying the initial conditions and suggesting a ready-made solution are different kinds of help. The interview record must make clear exactly what the person received.

Use the following instruction to review the results:

For each requirement, give the corresponding question, the exact part of the answer and the conclusion. Separate the results: the requirement was met in the task; there is a specific mismatch; there is not enough information. Do not infer personal qualities from a missing answer. Do not assess skills for which there was no task. List the questions remaining for the manager separately.

This is a requirement for the content of the review. Having this instruction does not itself guarantee that AI’s answer will be correct: the manager checks questionable conclusions against the original text.

Where is a manager needed after AI?

The manager is needed when there is something still to observe, a contradiction to resolve or a hiring decision to make. The final meeting should add information that is missing from the automated interview.

Repeating the whole questionnaire brings back the original demands on your time. It is better to take one unresolved question. If all you have is a written message to a buyer, have a short spoken conversation about that request. If the candidate described experience of finding customers, ask them to reconstruct the sequence: where they got the contact, how they opened the conversation and how they got it to continue. Without the names of former customers or confidential commercial information.

Discuss the work itself separately. Who helps the salesperson with the product? Which decisions do they make independently? Who do they turn to when they do not have enough information to answer? The candidate is also choosing a company, and a positive assessment of their answers does not mean they have agreed to the actual duties.

After hiring, the unknowns do not disappear. If the interview showed a well-written email, the next stage of work should show how the person handles real enquiries. The article “The new salesperson is busy all day. When will they start selling?” takes this further.

What if the candidate uses AI themselves?

State the rule for using AI before the task: whether it is allowed, for which actions and what the candidate must do independently. Assess the way of working that the role will actually require.

If the salesperson will prepare emails with AI, allow the tool and ask them to explain the result: where the terms came from, which promises they removed and why they asked the buyer those questions. The assessment then focuses on their ability to take responsibility for the message they have prepared. A polished text without an understanding of its content will not be a basis for hiring.

If a quick spoken answer without prompts is required, set aside a separate conversation with the manager for it. One submitted piece of writing has not shown you that skill. Looking for signs of “language that is too perfect” is useless for this purpose: even establishing the email’s authorship with complete accuracy will not show how the person talks to a customer.

When does AI really reduce the work of hiring?

AI reduces the work of hiring when it takes over the repetitive initial questions and the manager uses the results for a shorter, focused conversation. If every conclusion has to be reconstructed from scratch or no one reaches a meeting after screening, first examine the script and the reasons candidates do not progress.

Before launch, agree on three outcomes of the initial interview: invite the candidate to the next stage, request missing information or refer an answer that is hard to assess to a person. Tie a final rejection to a specific mandatory requirement. This gives the recruiter an action to take on an incomplete answer other than removing the candidate from the list.

When the job description, initial questions and introduction to the work have to be repeated for everyone, Robot Recruiter helps put this part together — RENTROP’s AI recruitment service: a web page for the vacancy with a chat adviser, an interview based on a script for the role with situational tasks and answer assessment, and AI training to help a new employee settle into the job. The hiring decision remains with the manager.

Compare the time spent handling applications from start to finish. Add the recruiter’s time spent on clarifications, the manager’s time spent in meetings and the time spent correcting assessments. Divide that total by the number of candidates for whom you were able to reach a reasoned decision. Alongside this, keep a record of the number of unfinished interviews and the reasons: otherwise, a shorter list will look like a saving even when people simply could not get through the screening.

A rejection must also have someone responsible for it who can explain the reason. Saying “that is how AI assessed it” leaves the company without an answer to the main question: what work required by the company was this person unable to demonstrate? Until there is an answer, a CV marked as closed means only that the company has stopped considering it.