For hiring managers
AI interview questions for customer support candidates
Support candidates often say they use AI to draft replies. That claim tells an interviewer little. Some agents send fluent replies that miss the customer’s question or state an old policy. Others reply faster and more accurately, and they know which sentences to confirm before sending.
The twelve questions below follow the six skills Emeri measures, with two questions for each skill. For every question, the page describes a strong answer and a weak answer. Two interviewers can then score the same answer the same way.
How to use these questions
- Ask for a real example
- Each question works best when the candidate describes work they have done. When an answer stays general, ask what they did the last time, and what happened next.
- Ask every candidate the same core questions
- Pick four to six questions before the interviews start. The same questions make answers comparable across candidates.
- Score the behavior, not the vocabulary
- A candidate who names many tools or uses the right terms has not shown the skill. Score against the strong and weak notes under each question.
- Pair the interview with a short work sample
- If the job uses AI, let the candidate use it in the sample. Ask for the requests they wrote and a note on what they changed. An interview shows how a candidate talks about the work. A sample shows the work.
Asking well
Asking for what you need clearly enough to get useful results. What this skill looks like for customer support professionals.
Question 1
A customer writes that their data export keeps failing and they plan to leave at the end of the month. A teammate asks an AI tool to "write a reply that keeps this customer." What is missing from that request?
Support requests often name a mood or a goal and leave out the problem. This question shows whether the candidate puts the customer’s issue at the center of the request.
A strong answer
Puts the export failure first and asks for a reply that addresses it. States what the team can offer, such as a workaround or an escalation to engineering. Names the channel, the length, and the next step for the customer. Says the reply must not offer a discount or a fix date that nobody approved.
A weak answer
Adds words such as "warm" or "persuasive" and leaves the export problem unstated. Or focuses the reply on reasons to stay and treats the failed export as a side issue.
Question 2
Tell me about a saved reply or macro you built with AI help. How did you word the request so it worked for different customers?
Macros reach many customers, so a weak request repeats its errors at scale. This question tests whether the candidate can write a request that holds up across cases.
A strong answer
Describes placeholders for details that change, such as the customer’s name, plan, or region. Asked for separate versions for email and chat. Tested the macro on several past tickets with personal details removed. Changed the request when a version ran too long or skipped a step.
A weak answer
Built one generic macro and rewrites it by hand for each ticket. Or cannot say what the request contained, only that the tool produced something usable.
Giving good context
Giving AI the background, data, and limits it needs. What this skill looks like for customer support professionals.
Question 3
You pick up a ticket from a colleague on another shift. The thread has nine messages and several internal notes. What do you give an AI tool before you ask it to draft the next reply, and what do you leave out?
A model has not read the thread and does not know what the team already tried. This question tests whether the candidate selects the material a new agent would need.
A strong answer
Includes the customer’s messages, the fixes already attempted, and any commitment the colleague made. Adds the current account status and the relevant policy text. Leaves out payment details and personal data the reply does not need. Checks the internal notes against the account, because notes can be wrong or out of date.
A weak answer
Pastes only the latest message, so the draft repeats a fix the customer already tried. Or pastes the whole thread with every note and personal detail and lets the tool sort it out.
Question 4
Our product changes every few weeks. How do you keep AI drafts from describing menus, settings, or plan features that no longer exist?
A model may hold an older picture of the product, or a general picture of similar products. This question tests whether the candidate supplies current product facts on purpose.
A strong answer
Supplies the current help article and recent release notes, with their dates. Names the customer’s plan, because features differ by plan. Tells the model to use only the supplied steps and to say when they do not cover the question. Follows the steps in the product before sending when the release is recent.
A weak answer
Relies on the tool to know how the product works. Or works from an internal document without checking when it was last updated.
Checking the work
Spotting mistakes, checking sources, and knowing what to trust. What this skill looks like for customer support professionals.
Question 5
An AI draft tells a customer they can get a full refund within 60 days. You are not sure that matches our policy. What do you do before sending?
A wrong policy statement in writing becomes a promise the company has to handle. This question tests whether the candidate confirms a policy claim at its source.
A strong answer
Opens the current policy and confirms the window and its conditions. Checks that this customer’s account meets them. Reads the rest of the draft for other commitments, such as timelines or credits. If a macro or a chatbot produced the wording, reports it so the source gets corrected.
A weak answer
Sends the draft because the customer will be glad to read it. Or changes "can get" to "may be eligible for" and sends it without checking the policy.
Question 6
Our chatbot answers some customer questions before an agent sees the ticket. If you owned a weekly review of those answers, how would you run it?
Chatbot errors reach customers with no agent in between. This question tests whether the candidate reviews for accuracy and resolution, not only for tone.
A strong answer
Samples answers across topics and gives extra attention to billing, cancellations, and account access. Compares each answer with the current policy or help article. Checks whether the customer’s question was answered. Reads conversations where the customer returned or asked for a person. Logs each error type and sends fixes to the owner of the source content.
A weak answer
Reads a few answers and judges whether they sound friendly. Or relies on the satisfaction score alone and does not open the conversations behind it.
Putting it together
Turning prompts, tools, and steps into a repeatable workflow. What this skill looks like for customer support professionals.
Question 7
How would you use AI to hand off your open tickets at the end of a shift? The next agent should not have to reread every thread.
Shift handoffs recur every day, and a missed commitment becomes a broken promise to a customer. This question tests whether the candidate can build a process that is consistent and checked.
A strong answer
Uses a saved request with fixed fields for each ticket. The fields cover the customer’s question, what has been tried, what was promised and by when, and the next step. Runs it one ticket at a time. Checks each summary against its thread before handing off, with most care on promises and deadlines. Writes the steps down so other agents can use them.
A weak answer
Puts every open ticket into one request and passes on the result unread. Or writes a new request each shift, so the handoff format changes from day to day.
Question 8
Which parts of support work would you keep out of an automated AI process, and why?
Knowing where automation does not fit is part of the skill. This question tests judgment about risk, volume, and how easily a result can be checked.
A strong answer
Names identity checks, exceptions to policy, complaints with legal or safety issues, and customers in distress. Explains that these need a person’s judgment or carry a high cost when wrong. Contrasts them with high-volume, fixed-format work that is quick to check. Gives an example from their own queue.
A weak answer
Says every ticket type could be automated with a good enough tool. Or says AI should never touch support work, with no distinction between task types.
Getting real work done
Using AI to finish real work, not just make a first draft. What this skill looks like for customer support professionals.
Question 9
Walk me through a difficult ticket you resolved with AI help. What did the AI do, and what did you decide?
The resolution is what the customer experiences. This question tests whether the candidate owns the judgment in AI-assisted support work.
A strong answer
Separates the parts clearly. AI may have summarized the thread, drafted a reply, or translated a message. The candidate chose the resolution, confirmed the account facts, and set the tone. Says how the ticket ended, including whether the customer wrote back.
A weak answer
Cannot separate their decisions from the tool’s. Or describes sending the AI reply with light edits and closing the ticket.
Question 10
During an outage, the queue grows faster than the team can answer it. How would you use AI to keep replies fast and accurate until the incident ends?
Outages are when agents take shortcuts and speculate about causes. This question tests whether the candidate uses AI for speed while holding replies to confirmed facts.
A strong answer
Works from one approved status statement from the incident owner. Uses AI to adapt that statement to each ticket’s details. Gives no fix time beyond what the incident team has posted. Updates the source text when the status changes. Separates tickets that are not about the outage and tags affected tickets for a follow-up.
A weak answer
Lets AI write each reply from scratch, including guesses about the cause. Or sends one bulk reply to every new ticket, including customers whose issue has nothing to do with the outage.
Keeping it safe
Protecting private information and using good judgment. What this skill looks like for customer support professionals.
Question 11
A customer writes from a new email address and asks us to move their account to it. An AI draft agrees and lists the steps. What do you do?
Account takeover attempts often arrive as ordinary support requests. This question tests whether the candidate keeps identity decisions out of the AI draft.
A strong answer
Does not send the draft. Follows the team’s identity verification process, which a model cannot perform from message text. Treats the request as a possible takeover attempt until verification is complete. Replies with the verification steps through the approved channel and records what happened on the ticket.
A weak answer
Sends the draft because the customer knew the account details. Or asks the AI tool whether the request looks genuine and acts on its answer.
Question 12
A customer asks in chat whether they are talking to a person or a bot. How should the team handle that question, and why does it matter?
Customers who find out they were misled about a bot stop trusting the support channel. This question tests whether the candidate understands disclosure as part of safe AI use.
A strong answer
Says the customer gets an honest answer that follows the company’s disclosure rules. Offers a route to a person. Knows that some places have legal rules on disclosing automated agents. Would raise it with the team lead if the bot is set up to avoid the question.
A weak answer
Says it does not matter who answers if the answer is correct. Or suggests the bot should change the subject so the customer stays in self-service.
For what each skill looks like in this role at every band from E1 to E4, read AI skills for customer support professionals. These questions are written for interviews. They are not taken from the Emeri assessment.
Questions
What AI skills should I look for when hiring for customer support?
Look for all six skills Emeri measures: asking well, giving good context, checking the work, putting it together into workflows, getting real work done, and keeping it safe. In support, give extra weight to checking the work and keeping it safe. Policy statements and customer personal data are where AI-assisted replies most often cause harm.
How should I run a written ticket exercise for support candidates?
Give the candidate a realistic ticket, a short policy excerpt, and access to an AI tool if the job uses one. Ask for the reply, the requests they wrote, and a note on what they checked before sending. Score whether the reply answers every question in the ticket and states the policy correctly.
Are these questions taken from the Emeri exam?
No. These questions are written for interviews and are never exam items. The Emeri assessment uses its own questions and scenarios, graded against the same six skills.
How do I confirm a candidate’s Emeri credential?
Every Emeri credential has a public verification page. It shows the band, the assessment date, and a profile across the six skills. The employers page explains how to read one. No account is needed.
Evidence beyond the interview
Emeri assesses the same six skills with graded work, and each credential has a public verification page. Read how to check a candidate’s credential, or see how teams use Emeri assessments.