Dress for the Person Who Watches the Recording, Not the AI

Does the AI See What You Are Wearing?

Neither of the two platforms quoted here reads your clothing. HireVue's 2025 statement says its AI relies only on what the candidate says, and that it does not assess facial expressions, body language, or background and surroundings ¹. Vervoe's candidate help page says that for audio or video questions its AI reviews only the transcript, and does not assess facial expressions, voice tone, accent or any other personal attributes ³.

What the HireVue statement describes instead is a pipeline: speech-to-text transcription, then a language model, then a score benchmarked against expert human raters' evaluations of answers to similar competency-based questions ¹. What gets read is the sentence you said, after a recognizer has turned it into text.

HireVue and Vervoe each wrote those pages about their own systems. Neither page is an audit run from outside the company.

HireVue took the visual channel out years before that statement. HireVue's chief data scientist wrote in January 2021 that the company would not use any visual analysis in its pre-hire algorithms going forward ². A same-day press release dates the removal itself to early in 2020, for all new assessments ². That covers HireVue's own new assessments from that point and nothing else: not older client models, and not whichever platform sent you the link, unless it has published something of its own.

Finding out which system sent your link is a separate job, and What an AI interview scores covers it.

Who Watches the Recording After You Upload It?

A person at the employer, on their own schedule. Vervoe says the video and audio are transcribed and provided to the employer along with the recording ³. Unilever tells applicants to its 2026 Future Leaders Programme, in writing, that it does not use AI as a screening tool and that the digital interview submission is reviewed by a real person ⁴.

That is one programme in one market, and an employer describing itself rather than an audit.

It still matters, because it is an employer putting in writing that a person does the reading. Reading what employers are told about the same round is the other half of this. The employer-side piece on whether the async video round is worth keeping when answers are scripted argues the format separates candidates on delivery and hardly at all on substance.

One thing that moves a human rating has been measured, and clothing was not in it. In an experiment with 114 participants, interviewees were rated lower over videoconference than face to face, and perceived social presence and perceived eye contact contributed to the gap ⁷.

Ratings made from recordings also came in lower than ratings made live ⁷.

That was a lab, and a wardrobe was never one of the things it varied.

When the Platform Has Published Nothing, Keep It Plain

When the platform has published nothing, plain and unremarkable is the safer default, on evidence that is suggestive rather than settled. Bayerischer Rundfunk's data team ran several hundred clips through Retorio, a German video-interview tool, using an actress who repeated the same answer and ten further test subjects whose videos were edited one factor at a time ⁵.

Glasses cut nearly 10 points from the actress's conscientiousness score. A headscarf moved her results by nearly 20 points. A picture frame and woodchip wallpaper added behind a test person shifted scores by up to 15 points, while repeating the same answer unchanged left the results stable ⁵.

Retorio did not dispute the effect. It said the impression a person makes on others is part of the concept ⁶.

The company told the reporters that its algorithm is trained to measure the impression people make, learned from 2,500 people rating how they perceived more than 12,000 others on the Big Five dimensions ⁶. A model built that way reproduces what its raters reacted to, appearance included. That is the honest reason to keep the picture uneventful when you cannot tell what is reading it: it puts less job-irrelevant material in front of whatever is, and a plain shirt earns nothing beyond that.

Two limits belong with that finding. It is journalism rather than peer-reviewed research: a handful of test subjects, no statistical testing, one vendor, and scores read off its own scales in late 2020 and published in February 2021 ⁵. And the headscarf result is a criticism of the tool, not an instruction to anybody. Nobody should be asked to take off what they wear, and a system whose output moves when they do is describing its own defect.

Keep Two Rules and Give the Evening to the Answers

Contrast with your background, and wear nothing you will adjust mid-answer. Those two cover most of what the long dress guides are reaching for. A reviewer should be able to see your face against the wall behind you, and you should not spend the end of an answer thinking about a collar.

• Match the company, not the platform: whatever you would wear to a video call with that team is right for the recording too.
• Fine stripes can shimmer on a compressed feed. That one is a codec artifact rather than a judgment about you.
• Test the frame once on the practice question, if the platform offers one, and look at the picture rather than at the outfit.
• Check the retake policy before the first real question, because it decides how much of any of this is recoverable.

Nothing on that list takes an evening.

No source cited here measures what colour, cut or formality does to an interview rating, human or automated. That absence is the honest reason the advice is short: what can be said covers visibility and comfort, and stops well before it reaches colour.

The hour you were going to spend on the outfit buys more against the questions. Preparing for a one-way video interview in an hour starts with the competencies hidden in the posting and one worked example for each, which is the part of the round a reviewer can still be surprised by.

If the invitation names the platform, look for that vendor's own candidate page.

The two quoted above say plainly what their systems read, and checking takes less time than choosing a shirt.

See how it works >