AI Interview Software for Video Interviews: A Hiring Checklist for Trades

Mostly yes for scheduling, reminders and transcripts, and no for anything that scores or ranks candidates from their face, voice or answers. Rules vary by place, but the common thread is four duties: tell candidates, get consent to record, test for bias or don't score, and keep a person making the decision.

AI interview software for video interviews comes in two kinds. Helpers book the slot, run the call and write up what was said. Judges rate the candidate. A firm that hires two or three people a year has no way to test a judge for bias on its own applicants, so the line I'd draw is simple: software does the admin, and every judgement stays with a named person.

Follow me on Instagram@sagnikteaches

Helpers and judges: which features a trades firm should switch on

Most video interview products mix both kinds in one plan, which is where small firms get caught. The table sorts the common features by how much can go wrong, with a verdict for a business that hires a handful of tradespeople or apprentices a year.

Connect on LinkedInSagnik Bhattacharya

Nothing in the "use" rows needs a specialist product. Zoom's Pro plan lists unlimited meeting summaries and AI note-taking at $14.16 per user per month billed annually, while its free Basic plan allows three of each a month. Google Meet's "Take notes for me" needs an eligible Google Workspace edition or Google AI plan, and Meet tells every participant that notes are being taken. Microsoft's Teams documentation says intelligent recap needs a Teams Premium or Microsoft Copilot licence. A scheduling page such as Calendly's Standard plan is listed at $10 a seat a month on yearly billing as of October 2026 (check the live page). For one hiring round that is about $24 a month in tools, and usually less, because most firms already have a video-call plan. Setting up the booking side without endless messages is covered in automating interview scheduling, and the consent wording that goes with any AI note-taker is in AI note takers and candidate consent.

Subscribe on YouTube@sagnikteaches

The "avoid" rows are where the legal and ethical weight sits. Scoring and ranking tools only pay off when you have far more candidates than any person can read. A firm that sifts 58 applications and watches six interviews isn't in that position; it needs consistency, and a written scorecard gives you that for nothing. For help building one, see how to build interview questions and scorecards with AI.

What the rules ask of a small employer, in plain terms

I'm not a lawyer and none of this is legal advice. Hiring rules sit in several overlapping areas, and they differ by place and by where the candidate is sitting. What follows is the practical shape of the main rule families, so you know what to put to an employment adviser or your data-protection adviser before you buy anything that scores people.

• Tell people before the interview. Data-protection law in many places expects you to explain what you do with candidates' data, and to say more where a decision is automated. At least one city also requires employers to give candidates notice before an automated tool screens them. Put the notice in the invitation, not in the first minute of the call.
• Get consent to record, and make saying no easy. In some places one participant's consent is enough to record a call; in others everyone on it must agree. When people join from different places it can be unclear which rule applies, so the safe habit is to ask every candidate in writing first and again on the call.
• Test for bias, or don't score. At least one city requires employers who use an automated tool to screen candidates to have it audited for bias within the previous year and to publish a summary of the results. Equality law applies whichever tool you use, including the expectation in many places that you adjust the process for disabled candidates. Assume you stay answerable for what a tool you chose does; a vendor's assurance doesn't move that.
• Keep a person in the decision. Data-protection law in many places restricts decisions based solely on automated processing where they have legal or similarly significant effects; under the strictest versions, people can ask for human intervention, give their view and contest the outcome. The explanatory notes to one major data-protection law give "e-recruiting practices without any human intervention" as an example of exactly that kind of decision.
• Check whether an AI-specific law covers your hiring tools. At least one major AI law classes AI used to recruit or select people, including filtering applications and evaluating candidates, as high-risk, with the main duties for those systems scheduled to apply from 2 December 2027. The same law has banned AI that infers people's emotions at work since February 2025, and the official guidance on that ban says it covers candidates during recruitment.

That is enough to explain why I'd keep scoring out of a small firm's process. Every duty above assumes someone can explain how a tool reached its output, and a vendor's black box doesn't give you that. If you want the longer version of the ethics, the AI interview software ethics checklist goes into bias, emotion recognition and notice in more depth.

Checklist one: before you buy anything

1. Write the job in one sentence. "Book interviews and keep notes" describes a helper. If your sentence contains rank, score, rate or predict, you're shopping for a judge. Why: it stops a demo steering you towards features you didn't need. Check: say the sentence to the vendor before the demo starts.
2. Count your candidates. Add up the people you interviewed over the last two years. Why: a few dozen applicants can't show you whether a tool treats groups of people differently, and a vendor's average results say little about your own pool. Check: if your number is in the dozens, rule out scoring features.
3. Name the decision-maker. Write down who makes each hiring decision. If the honest answer is "the software" or "whoever is free", fix that first. Why: every rule above assumes a person who can explain the decision. Check: that name appears on your scorecard.
4. Find out where candidates will sit. List where applicants live as well as where the job is. Why: recording, notice and hiring-tool rules can depend on either place. Check: ask your adviser one question that covers both.
5. Price the hours, not the licence. Time your last hire: scheduling, note-writing and sifting. Why: the real saving from helpers is admin time, and it is modest (the sum is further down). Check: you can name the task each feature removes.

Checklist two: ten questions for the vendor

Put these in an email and keep the replies. A sales call isn't evidence, and a promise made in a chat window is hard to point to later. If the vendor won't answer questions one to four in writing, stop there.

Checklist three: telling candidates and getting consent

1. Put the notice in the invitation. People should read it before they're on camera, not as the call connects. Check: the notice sits above the link in the message.
2. Say what is recorded, who sees it and how long it is kept. Name the person, not "the team". Check: the retention period in the notice matches the deletion setting in the tool.
3. Say what the software does not do. "It does not score you or analyse your face or voice" is a promise you can keep only if you've switched those features off. Check: open the settings page and look.
4. Make recording optional. Offer an unrecorded call, a phone call or an in-person meeting, and say that choosing one counts for nothing against the candidate. Check: you have a person ready to run each option.
5. Keep the answer. A "yes, record" or "no, don't record" reply belongs on the candidate's file. Check: you can find it in under a minute.
6. Check the extra rules for young candidates. Trades firms often interview teenagers for apprenticeships. If someone may be under 18, ask your adviser what you need from a parent or guardian before recording.

A copy-and-paste candidate notice

This is an illustration to take to your adviser, not finished legal wording. Square brackets are the parts to fill in.

Subject: Your interview on [date], and how we use technology in it

Hi [first name],

Thanks for applying for the [role] job at [firm]. Your interview is booked
for [date and time] on [video platform]. The link is below.

Before then, please read how we use technology in our interviews:

- We would like to record the call and have software write a transcript
  and a short summary. [Name] reads both. Nobody else has access.
- The software does not score you, rank you or analyse your face, voice
  or emotions. [Name] makes every hiring decision, not the software.
- We keep the recording and notes for [90 days] after we reach a
  decision, then delete them. Reply to ask for them to be deleted sooner,
  or to see what we hold about you.
- Recording is optional. If you would rather we did not record, or you
  prefer a phone call or a meeting in person, tell us. It will not count
  against you.
- If you need any adjustment (extra time, captions, a different format),
  tell us and we will arrange it.

Please reply "Yes, record" or "No, don't record" by [date].

Thanks,
[your name]

Notice what the wording avoids. It doesn't call the software "smart" or "unbiased", and it doesn't promise anything the settings page can't back up. Each bullet maps to a line in the vendor questions above, so the notice and the contract say the same thing.

Checklist four: running the interview itself

1. Ask every candidate the same six questions in the same order. Keep one on a real job that went wrong and what they did next. Check: the questions are printed on the scorecard before the first interview.
2. Score each answer on a written 1 to 5 scale before reading any AI summary. A workable anchor: 1 is no relevant example, 3 is a relevant example with an unclear result, 5 is a specific example where the candidate's own part and the result are clear. Why: reading a summary first anchors your view on its wording.
3. Ignore the backdrop. A messy van, poor lighting or a signal that drops says nothing about whether someone can install a ducted system. Check: the scorecard has no row for presentation.
4. Give a second try after a technical failure. Candidates on a job site or a phone hotspot will drop out. Reschedule without comment. Check: your invitation says so.
5. Keep the practical test separate. A short trial on real tasks tells you more about a tradesperson than any video score. The HVAC technician job description tutorial shows how to build one that treats applicants fairly.

Here is one candidate's scorecard, filled in for an invented installer interview. The interviewer scored each answer before opening the AI summary, then added the transcript lines that justify the score.

Questions one to five total 19 out of 25. The decision line, written by the owner with a date: "Shortlist for the paid trial day. Strong safety answer and a clear example of owning a mistake; weaker on handover to the customer." Six rows of evidence like these are what you would show a candidate who asked why they weren't taken forward, and they don't depend on any software's score.

Checklist five: after the call, records and review

1. Read the summary against the transcript for three claims. Qualifications or tickets, availability and notice period, pay expectations. Why: summaries state these confidently even when the candidate hedged. Check: each claim has a transcript line number beside it.
2. Write the decision and the reason in two lines. Put a name and a date on it. Why: that is what you'll want if a candidate asks why they weren't taken forward.
3. Have a person send the message. AI can draft a polite rejection; it shouldn't invent a reason. Check: the message gives no reason you haven't recorded.
4. Delete on schedule. Put the deletion date in your calendar when you book the interview. Check: recordings from the last hiring round are gone.
5. Review every quarter. Count candidates who declined recording, who asked for adjustments, and who dropped out between invitation and call, and log any complaint. Re-read the vendor's privacy settings, because vendors change defaults when they update a product. If you can't explain a number, find out why.

For a hiring process that wraps around these checks, from the advert to the offer, the tutorial on setting up an AI-assisted hiring process for a small team covers the stages either side of the interview.

Scoring two vendors for an eleven-person HVAC firm

Here is the checklist filled in for an illustrative firm with eleven staff, which needs two installers and has had 58 applications. It has demoed two products. Vendor A is a video-interview platform whose pitch is an "AI fit score". Vendor B is a scheduling link plus the video-call plan the firm already has, with transcript and summary switched on.

The owner picks Vendor B. Vendor A isn't ruled out for good: if the firm ever hires dozens of people a year, it would revisit scoring with an adviser and an audit report in hand. For now, nothing that matters is lost, because the owner was always going to watch the six interviews personally.

The sum: what the admin help saves and what scoring would add

Take the same firm and time one hiring round by hand. Sifting 58 applications at two minutes each is about two hours. Twelve fifteen-minute phone screens are three hours. Six forty-minute video interviews are four hours. Booking those six by message, at five messages and about two minutes each, costs another hour. Writing up each interview takes roughly 25 minutes, so 2.5 hours. The total is about 12.5 hours.

With a scheduling link and a transcript plus summary, scheduling drops to a quarter of an hour (setting up the link, plus the odd change), and note-writing drops to about ten minutes per interview because the owner is checking and editing, not typing from memory. That is 10.25 hours, a saving of about 2.25 hours. At a notional $50 an hour the owner's time is worth about $112, against $10 to $24 a month for the tools, depending on whether the firm already has a video-call plan. That is a sensible trade, and it is the whole size of the prize.

Now add AI scoring to the sift. Its best case is to shorten the two hours of reading applications, saving perhaps 90 minutes, and it leaves the interviews untouched, because the owner still has to watch them. In exchange the firm takes on the audit, notice, consent and explanation duties above for a tool it can't test. That is a poor deal at this size.

How a transcript summary goes wrong, and the check that catches it

Take a typical slip from the same firm's second interview. The candidate says: "I've finished the refrigerant-handling course, and the exam is in March." The summary reads: "Holds a refrigerant-handling certificate." It is a small compression, and it changes the offer: a certificated installer and a trainee are different roles at different pay.

The three-claim check from checklist five catches it. The owner searches the transcript for "certificate" and "exam", finds the sentence, and corrects the note before the scorecard is filled in. Make the habit mechanical: any claim that affects pay, safety or what you'd let the person do alone gets a transcript line beside it. Where a candidate hedges, record the hedge.

You can also make the slip less likely before it happens. If your note-taker lets you set its summary instructions, or you summarise the transcript yourself in a chat assistant on a business plan that your candidate notice covers, a tighter instruction changes what comes back:

Summarise this interview transcript for the person making the hiring
decision. Rules:
- Use only what the candidate said. Quote, word for word, anything
  about qualifications, tickets, licences, notice period, availability
  or pay.
- Put the transcript line number after every point.
- Keep any hedge the candidate used ("I think", "the exam is in March").
- Do not rate, score, rank or recommend the candidate, and do not
  describe their personality, confidence or manner.
- End with "Unclear or missing:" and list what to ask at the next stage.

A sample of what comes back (illustrative):

Refrigerant handling: "I've finished the refrigerant-handling course,
and the exam is in March." (line 23)
Notice period: "Four weeks, three if they'll let me go early." (line 114)
Pay: gave a range (line 116).
Comes across as a confident communicator.
Unclear or missing: whether they hold a full driving licence for the
van; whether they have fitted ducted systems or only split systems.

Two fixes before it goes on file. The fourth line breaks the rule against describing the person, which models still do now and then; delete it, because a judgement about manner is exactly what the scorecard keeps out. Then open the transcript at lines 23 and 114 to confirm the quotes, since line numbers drift if a transcript is exported again. The "Unclear or missing" list is the most useful part: both questions belong on the paid trial day.

Transcription errors are a second source of trouble. Accents, background noise from a van or job site, and speech differences can all lower accuracy. Test your set-up on your own team's voices before the first interview, and read a sample of any transcript against the audio. If a transcript is poor for a particular candidate, rely on your own notes rather than the machine's.

Candidates who can't or won't do video

Four situations come up in trades hiring, each with a straightforward answer.

• The applicant who asks for a phone call. Say yes, use the same questions and the same scorecard, and treat the result exactly as you would a video interview. A phone option written into the invitation costs you nothing.
• The applicant who stammers or has a speech difference. Offer extra time and don't let an automated transcript stand in for your own judgement of the answer. Ask them what would help, and then do that.
• The teenager applying for an apprenticeship. Check what you need from a parent or guardian before recording, as in checklist three, and consider an in-person meeting with a parent present instead.
• The candidate calling from a van. Signal and noise will make a video call unreliable. Move to a call at a better time, or meet on site. Neither says anything about their skills.

If you only manage three things before your next hiring round, make them these: switch scoring off, send the notice with the invitation, and write down the name of the person who decides. The rest of the checklists tighten those three, and none of it needs a bigger budget than a scheduling link and a video-call plan.

Questions trades owners ask about AI interviews

Do I need consent to record a video interview?

It depends on where you and the candidate are. In some places one participant's consent is enough to record a call; in others everyone on it must agree, and when people join from different places it can be unclear which rule applies. The safe habit is to ask in the invitation, ask again at the start of the call, and keep the candidate's answer on file.

Can AI software reject candidates automatically?

I'd advise against it. Data-protection law in many places restricts decisions made solely by automated processing that have legal or similarly significant effects, which can include turning down a job applicant, and some places add audit and notice duties for hiring tools as well. Let software sort admin, and let a named person decide who goes forward and who doesn't.

Is an AI note-taker on a hiring call acceptable?

Usually, with notice and consent. Treat the recording, transcript and summary as the candidate's personal data: say they exist, say who can see them, set a deletion date, and check whether the note-taker shares notes by link or uses recordings to train its models. Business plans often differ from free ones on both points, so read the privacy settings before the first call.

What should I ask a vendor about bias?

Ask for the latest independent audit: who ran it, when, what method it used and which groups it covered. A dated report you can read is evidence; 'our model is bias-free' is a slogan. Also ask whether scoring can be switched off for your account. If the vendor won't put its answers in writing, treat that as your answer.

What can I use instead of AI scoring?

A written scorecard. Ask every candidate the same six questions, score each answer on a 1 to 5 scale with a one-line description of each level, and add a short practical trial for the shortlist. It costs nothing, and you can explain it to a candidate, an adviser or a regulator in plain words.

Further reads

• How to Use AI in Hiring for a Small Shop: Ads, Questions, Scorecards — Same hiring steps for a small shop: ads, questions and scorecards.
• How to Evaluate the AI Features in Your Recruitment Software — Judge the AI features inside recruitment software you already own.
• Apprentice Training Plan for Trades Firms: Sign-Offs and Reviews With AI — Plan the first months once the apprentice has been hired.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the text of a major AI regulation, its official implementation timeline and the official guidance on its prohibited practices; data-protection law on automated decisions and its explanatory recitals; a city's published page on automated employment decision tools; a media-law guide to recording conversations; Zoom and Calendly pricing pages; Google Meet Help ('Take notes for me'); Microsoft Learn (Teams intelligent recap licensing).