If you've been wondering how to use an AI interview platform without it turning into a three-month project, this guide is for you. We'll walk through your first AI interview in MSZ Hire, screen by screen, from a blank role to a shortlist your hiring manager can trust. Along the way you'll write the AI-disclosure line Ontario now expects in job postings, choose scoring weights, invite candidates and read a scorecard where every rating points back to the candidate's own words.
Our running example: a Mississauga contact centre hiring customer service agents. Swap in your own role as you go.
#How do you use an AI interview platform for the first time?
Think of it as running a structured interview at scale. You decide what good looks like, the AI interviewer asks every candidate the same core questions, and your team reviews the evidence. The platform does the repetitive part (asking, transcribing, matching answers to criteria). Your people do the judging.
In MSZ Hire, that splits into five stages:
- Build and publish the role (about four screens).
- Add the AI disclosure to your job posting.
- Invite candidates.
- Candidates complete the interview by voice, on camera (or with an approved accommodation).
- Review scorecards, record decisions and build the shortlist.
Let's go.
#Step 1: Fill in the role details
From Roles, choose Create a role. The header says it plainly: AI drafts the interview from your job description, and your team reviews every question and publishes a locked version.
On the Role details screen you'll enter:
- Job title, Department and Location, for example "Customer Service Agent", "Customer Care", "Mississauga, ON (On-site)".
- Jurisdiction. This one matters. The hint reads "Sets notices, consent and retention". Pick Ontario for our example. Quebec, British Columbia, Alberta, several US jurisdictions and "Other Canadian province" are also listed.
- Seniority: Junior, Mid, Senior or Lead.
- Interview length: 15 or 30 minutes. For high-volume, early-career roles, 15 minutes is usually plenty.
- Job description: paste the real one. It needs to be more than a sentence or two before you can continue.
#Step 2: Generate and edit the questions
On the Questions screen, click Generate with AI. MSZ Hire drafts a first set from your job title and description, one card at a time. Each card shows the question, the dimension it measures and the signals it's looking for (things like "ownership", "clarity" or "checking understanding").
The rules on this screen keep the interview structured:
- 4 to 6 core questions. The screen warns you if you're outside that range.
- All four dimensions covered: Technical competency, Problem solving, Communication and Practical experience. If one is missing you'll see "No question covers: …" and you can't continue until you fix it.
- One adaptive follow-up per question is allowed during the interview, so the AI can ask "what happened next?" without wandering off-script.
Now the important part: edit. Rewrite generic prompts into your world. "How do you use the main tools in this role day to day?" becomes "Walk me through how you'd handle a customer whose bill is higher than they expected, using whatever tools you've used before." Reorder with the arrows, delete with the bin icon, or click Add question to write your own.
If you want help writing questions and rating anchors that hold up, our guide to writing a structured interview scoring rubric has ready-made examples for contact-centre, retail and nursing roles.
#Step 3: Set scoring weights and check the recommendation bands
The Scoring screen has two panels.
Dimension weights. Each of the four dimensions gets a slider from 1 to 5, and the screen shows each one's share of the overall score as a percentage. For a contact-centre agent you might weight Communication 5, Problem solving 4, Practical experience 3 and Technical competency 2. The on-screen reminder is a good one: weights describe the role, not the candidate.
Recommendation bands. These are fixed calibration settings for version 1:
| Band | Overall score |
|---|---|
| Strong Hire | 85–100 |
| Hire | 70–84 |
| Leaning Hire | 55–69 |
| No Hire | 0–54 |
The screen labels these honestly: calibration settings, not validated predictors of job performance. And there's a safety rule underneath. If any dimension lacks evidence, the overall score is withheld and the attempt is marked Insufficient evidence. A candidate who runs out of time on one question isn't quietly given a zero.
#Step 4: Pick a persona, check compliance and publish
On Persona & publish, choose the interviewer's tone:
- Friendly Professional: warm and encouraging. Good for high-volume and early-career roles (our pick for the contact centre).
- Formal: concise and neutral. Good for regulated and executive roles.
- Tech Lead: peer-to-peer and precise. Good for engineering roles.
Each has a Preview voice sample. Persona changes tone and voice only, never the questions, evidence standards or thresholds.
On the right, the Compliance applied automatically panel lists what your jurisdiction switches on. For Ontario you'll see:
- AI disclosure text for the job posting
- Outcome notice within 45 days
- Consent and AI disclosure before start
- Text-mode alternative and accommodations
- No facial or voice biometrics
Check the summary (role, level, length, number of questions), then click Publish v1. Not ready? Save draft keeps your work. Publishing locks the version, so every candidate invited to v1 gets the same interview.
#Step 5: Write the Ontario AI-disclosure line for your posting
Since 1 January 2026, Ontario employers with 25 or more employees must say in a publicly advertised job posting if they use AI to screen, assess or select applicants. Everyone counts towards the 25, including part-time and casual staff across all your Ontario locations. Internal postings are excluded, and the province says a general statement is enough.
An AI interview that scores answers against criteria is assessing applicants. That stays true even though a recruiter makes the final call, so don't treat "a human decides" as a reason to skip the disclosure.
Here's a line you can adapt for the Mississauga posting:
We use artificial intelligence to help assess applicants for this role. Shortlisted applicants will be invited to a structured first interview conducted by an AI interviewer, answered by voice on camera. Our hiring team reviews every result and makes all hiring decisions. If you need an accommodation (such as typed answers or an interview without video) or would prefer to speak with a person, let us know.
While you're editing the posting, check the other Ontario requirements that started at the same time: a pay range (the spread can be no more than $50,000, with an exemption for roles above $200,000), a statement on whether the vacancy currently exists, no "Canadian experience" requirement, and a reply within 45 days to applicants you interviewed. Keep copies of postings and related records for three years.
#Step 6: Invite candidates
From the published role, open Invite candidates. Each candidate gets a secure, single-attempt link, and the interview clock starts only when they press Start, not when the email lands.
- Role (published version): choose "Customer Service Agent · v1".
- Deadline: the latest allowed start time. A week is a sensible default for hourly roles.
- Add people lets you enter a name and email per row, with Add another for more.
- Paste list takes one person per line in the format
name, email, and shows how many lines are valid before you send.
Click Send invites, or Copy link if you'd rather share it through your own channel. If you use an applicant tracking system, connect your ATS so invites and results flow without copy-and-paste.
#Step 7: See what candidates see
Before you invite a single real applicant, take the interview yourself. You can try the candidate interview demo right now. Here's the flow:
- Welcome. The candidate sees who invited them, roughly how long it takes, that they'll answer by voice on camera, and that humans decide. The notice explains that the camera and microphone are recorded for the hiring team to review, that recordings are kept for 180 days and then deleted, that the AI never analyses faces, expressions or voices, and how to request an accommodation or a human interview.
- Quick setup. A short camera and microphone check. If the browser blocks either, they're told how to allow it, or how to request an accommodation (typed answers, no video, or a human interview).
- Consent. A checkbox confirms they understand the interview is conducted by AI and reviewed by the hiring team, with the notice version and jurisdiction shown.
- Interview room. The interviewer asks each question aloud, live captions appear as the candidate speaks, and a progress bar and timer sit at the top. They press Done with my answer to move on.
- Submitted. The confirmation tells Ontario candidates they'll hear about the outcome within 45 days, and asks for a quick star rating.
#Step 8: Review the scorecard and build your shortlist
Open a completed interview from Interviews. At the top: the candidate's overall score in a ring, the AI recommendation band and the model and rubric version used. Below that, the four dimension cards show a 1–5 score each, the weight, and quoted evidence. Click a quote and the transcript on the right scrolls to that exact moment and highlights it. For voice interviews you can play the audio from there.
Then:
- Read the Strengths and Areas of concern tabs.
- Add Private notes (only you can see these).
- Click Record decision. Agree with the AI, or choose a different band. A different band needs a reason of at least 10 characters ("Strong de-escalation example in Q3 the rubric under-weighted"). The AI score stays unchanged for the audit trail.
- Click Mark as reviewed so your team knows it's done.
Your shortlist is simply the candidates you've reviewed and advanced, each with a reason a colleague can read later. That's the real win: not speed for its own sake, but decisions you can explain.
#What should you check after your first ten interviews?
Before you scale up, read ten transcripts end to end and compare them with the scores. Do the quotes support the ratings? Did any question confuse people? Open Analytics & fairness to see score parity by mode, so you can confirm text-mode candidates aren't scoring differently from voice. If something looks off, edit the role and publish a v2.
Then do it again for the next role. The second one is always faster, because by then you know what a good question sounds like in your organization.