Apply for a technology job today and there is a good chance the first "interviewer" you meet will not be a person. You might be asked to record answers to a webcam with no one on the other end, play a series of short games that claim to measure how your mind works, or simply submit a CV that an algorithm ranks before any recruiter reads it. Together these are known as AI hiring assessments, and they now sit at the front of many graduate schemes and tech hiring processes.
Employers use them for practical reasons. Popular roles attract hundreds or thousands of applications, and AI tools let companies screen everyone quickly and consistently rather than sampling whoever applied first. Technology companies in particular tend to adopt them early, partly because they build similar tools themselves and partly because their hiring volumes are large. Vendors also argue that structured, automated assessments are fairer than unstructured human screening, because every candidate faces the same questions under the same conditions.
Here is the honest position if you are the candidate: these systems are not mind readers, and they are not unbeatable black boxes. They are structured tests, and like every structured test, they reward people who understand the format, have practised the underlying skills, and stay calm on the day. There are no secret tricks, and the sites promising to help you "hack the algorithm" are mostly selling anxiety back to you. What actually moves your results is familiarity. This guide explains how each major format works, what is genuinely being measured, and how to prepare in a way that improves your performance without pretending to be someone you are not.
AI video interviews: what happens after you press record
The format most candidates worry about is the one-way video interview, with HireVue the best-known platform. Employers reported to use this style of interview include major banks, consultancies and technology firms, typically at the early screening stage.
The mechanics are straightforward. You receive a link, complete the interview in your own time before a deadline, and answer a short set of structured questions on camera. A typical setup gives you around 30 seconds to read each question and prepare, then a fixed window, often up to two or three minutes, to record your answer. Some employers allow a limited number of retakes per question; many do not. Everyone applying for the role gets the same questions, which is part of the point: the format is built on structured interviewing, a method with a long track record in selection research.
What happens to your recording afterwards depends on the employer. Some use the platform purely as a recording tool, with humans watching and scoring every answer. Others use AI-assisted scoring, where the platform transcribes what you said and evaluates the language of your answer against job-relevant competencies such as teamwork, adaptability or problem solving, with humans reviewing results downstream.
One fact worth knowing, because outdated advice about it is everywhere: the major vendors no longer score your face. HireVue publicly discontinued facial analysis, announcing the change in early 2021, after concluding that visual signals added little predictive value and amid sustained criticism of the practice. Current AI scoring, as vendors describe it, focuses on the content of what you say. Guides telling you to engineer your micro-expressions or hold unblinking eye contact are describing a system that mainstream platforms have retired. You should still look at the camera and present yourself well, for the same reason you would in any interview: a human may watch the recording.
What is actually evaluated is closer to a written exam delivered out loud: the substance, structure and relevance of your answers. That is good news, because substance and structure are trainable.
To prepare:
- Learn one answer structure and drill it. STAR (Situation, Task, Action, Result) fits the "tell me about a time" questions that dominate these interviews. Aim to land a complete answer inside 90 seconds to two minutes.
- Prepare five or six real examples from work, study or projects that flex to common themes: teamwork, conflict, failure, learning something quickly, delivering under pressure.
- Practise under recording conditions. Set a timer, record yourself on your phone answering a question cold, then watch it back. It is uncomfortable and it is the single highest-value thing you can do, because the awkwardness of talking to a lens fades quickly with exposure.
- Speak naturally rather than reciting. Memorised scripts tend to come out flat and rushed, and if a retake is offered, resist the urge to chase a perfect take; second attempts often sound more stilted, not less.
Game-based assessments: what the games are really measuring
The second format that unsettles candidates is the game-based cognitive assessment. The best-known examples are the pymetrics games, now part of Harver's platform following its 2022 acquisition, and Arctic Shores' behaviour-based assessments. Employers reported to use them include large graduate recruiters across consulting, banking, consumer goods and tech.
Instead of asking questions, these assessments have you play a series of short games, usually 10 to 12 of them across 25 to 35 minutes. Each game is a repackaged task from cognitive psychology. Depending on the platform, you might inflate balloons that can burst at unpredictable points, recall lengthening sequences of digits, react to arrows or symbols under time pressure, plan moves in a puzzle, or decide how to split money with a simulated partner. Behind the game skin, the platforms are measuring things like working memory, sustained attention, processing speed, planning, impulse control, risk appetite, and how you respond to reward, effort and setbacks.
The scoring model is different from a traditional test, and this matters for how you prepare. Vendors are clear that most of these games have no universal "right answers". Rather than passing or failing you against a fixed bar, the platform builds a profile of your traits and compares it with a profile associated with success in the specific role. Cautious behaviour in the balloon game is not inherently bad; it simply fits some roles better than others. This is why trying to fake a personality is a poor strategy: you do not know which profile the role is tuned for, and behaving unnaturally across a dozen different games is far harder than answering one questionnaire dishonestly.
Familiarity still helps, though, and for an ordinary reason rather than a sneaky one. If the first time you ever see a working-memory game is during the real assessment, you spend your opening rounds figuring out the interface while the clock runs, and your early performance reflects confusion rather than ability. Candidates who have seen the game formats before settle into their natural rhythm faster, and the assessment reads their genuine capability more cleanly. On the cognitive games specifically, the ones measuring memory, attention, speed and logic, practice does what practice always does: it sharpens the underlying skill. Regular timed work on reasoning tests builds exactly the mental muscles these games sample.
To prepare: get comfortable with the common game mechanics in advance, do regular short sessions of timed cognitive practice in the weeks before, and then play the real assessment honestly, in one sitting, rested and undistracted.
AI CV screening: how ranking works before a human reads anything
The least visible AI in hiring is the one that reads your CV. Applicant tracking systems have been standard at large employers for years, and it is widely reported that the overwhelming majority of major companies use one. What has changed recently is the intelligence layer: older systems matched keywords fairly literally, while newer AI-assisted screeners are described by vendors as reading a CV more contextually, extracting skills, experience and qualifications and scoring how well each candidate matches the job description, then presenting recruiters with a ranked list.
The practical implications for you are simpler than the technology:
- Tailoring is no longer optional. A single generic CV sent everywhere will sit low in rankings for every role. Mirror the genuine language of each job description where it truthfully applies to you, especially for skills, tools and qualifications.
- Keep the formatting boring. Clean headings, standard section names, real text rather than images or elaborate templates. Parsing has improved, but a plainly structured document gives the software nothing to misread.
- State the obvious. Software cannot infer what you leave implicit. If the role asks for Python, SQL or a certification and you have it, it should appear in plain words.
- Do not keyword-stuff or paste hidden text. Ranking is a first pass, not the decision; a human reads what the system surfaces, and a gamed CV fails at exactly that moment.
Screening is the one stage of AI hiring where "practice" means preparation of your documents rather than of yourself. Fifteen focused minutes tailoring each application beats fifty untailored submissions.
AI and coding assessments: where this meets traditional skills testing
If you are applying for engineering roles, AI is reshaping the coding assessment too. Asynchronous coding tests on platforms such as CodeSignal, HackerRank and Codility have long worked like take-home aptitude tests: standardised problems, automatic scoring, completed in your own time. What is newer is AI inside the assessment itself. Some platforms now offer formats where candidates are expected to work with an AI coding assistant, on the reasoning that engineers use these tools daily, so hiring should measure how well you direct them, judge their output and debug what they produce.
Two things follow. First, read the rules of each assessment carefully, because expectations now vary sharply: some tests prohibit AI assistance, others build it in, and using it where it is banned is treated as cheating. Second, notice what the AI-assisted formats are really measuring: problem decomposition, logical reasoning and critical evaluation of output. Those are aptitude skills wearing an engineering costume, and they respond to the same kind of structured practice as any reasoning test.
What AI assessments mean for how you prepare
Strip away the interfaces and a pattern emerges. A game-based assessment sampling your working memory and pattern recognition is measuring what an abstract reasoning test measures. Games and case-style questions involving numbers, trade-offs and probabilities draw on the skills a numerical reasoning test trains. Rule-finding and sequence games run on the same machinery as a logical reasoning test. And an AI video interview scored on the content of your answers rewards precisely what verbal reasoning practice builds: structuring information, picking the relevant from the irrelevant, and expressing conclusions clearly.
This is the most useful thing to understand about AI hiring, and the reason it is less alarming than it first appears. The scoring layer is new; the abilities being scored are not. Employers have measured numerical, verbal, logical and abstract reasoning for decades because they predict job performance, and the AI formats are largely new ways of sampling the same abilities. That means preparation transfers. Time spent on classic aptitude practice raises your floor across game-based assessments, structured interview answers and coding screens alike, and unlike format-specific tricks, it does not expire when a vendor updates its product.
If you are facing an AI-led process, the strongest use of your preparation time is a short daily habit of timed reasoning practice, started as early as you can. You can take a free practice test on any of the linked pages above and get an immediate sense of where you stand.
Your preparation checklist for an AI-run hiring process
Work through this in the week before any AI assessment:
Environment
- Quiet room, tidy background, and a session booked when you will not be interrupted.
- Test your camera, microphone and internet beforehand; use a laptop or desktop rather than a phone where allowed.
- Light your face from the front, put the camera at eye level, and close every other tab.
Structured answers
- Prepare five or six flexible real examples and know each one in STAR shape.
- Rehearse a 60-second answer to "tell me about yourself" that connects you to the role.
- Practise finishing answers cleanly inside two minutes rather than trailing off.
Practice under recording
- Record at least three full mock answers on camera, under a timer, and review them.
- Note one thing to fix per attempt (pace, filler words, structure) rather than everything at once.
Cognitive practice
- Do short, timed sessions of numerical, logical, verbal and abstract reasoning practice on alternating days.
- Familiarise yourself with common assessment-game formats so the real interface holds no surprises.
- The night before: stop. Sleep is a genuine performance input for cognitive testing, and cramming is not.
What you cannot game, and why that is good news
An honest guide has to draw this line clearly. Some things about AI assessments are not beatable, and chasing them wastes preparation time you could spend on things that work.
You cannot reliably fake a trait profile across a battery of games, because you cannot know which profile the role rewards and because sustained pretence across a dozen varied tasks tends to produce inconsistent, low-quality data rather than a flattering picture. You cannot script your way to a strong video interview, because over-rehearsed answers sound like what they are, and because the recording is often reviewed by humans who have watched thousands of them. And you cannot fake your way through the whole pipeline even if one stage goes well, because AI screening is almost always the first filter rather than the final decision. The candidates it surfaces still face human interviews, where inconsistency with your earlier answers is conspicuous. Verification is also tightening across the industry: platforms increasingly describe identity checks, proctoring features and follow-up live stages designed to confirm that the person who performed is the person who turns up.
The good news is what this implies. Because the systems are hard to trick, employers can trust them, and that means genuine preparation is rewarded rather than drowned out by gamesmanship. The candidate who has practised the underlying skills, knows the formats and answers authentically holds the real advantage. Preparation and pretence are different things; only one of them survives contact with a second interview.
Frequently asked questions
Are AI interviews watched by real people?
Often, yes. Many employers use one-way video platforms as recording tools with human reviewers, and even where AI-assisted scoring is used, humans typically review shortlisted candidates. Assume a person may watch every answer you record.
Does HireVue analyse my facial expressions?
Not any more, by the company's own account. HireVue announced in early 2021 that it had discontinued facial analysis, and current scoring is described as focusing on the content of your spoken answers. Present yourself well for the human viewers, but you do not need to perform for a face-reading algorithm.
Can you fail a game-based assessment like pymetrics?
There is usually no fixed pass mark. These platforms build a profile of your traits and compare it with the profile associated with the specific role, so results are about fit rather than a score out of ten. You can, however, be screened out for a given role, which is why performing naturally and attentively matters.
Do the games have right answers?
Mostly no, for the behavioural games; the trade-offs are the measurement. The cognitive games, covering memory, attention, speed and reasoning, do reward stronger performance, and those skills improve with practice.
Is it worth practising for assessments with no right answers?
Yes, for two reasons. Familiarity with the game formats means your real attempt measures your ability rather than your confusion with a new interface. And the cognitive components draw on reasoning skills that timed practice demonstrably sharpens.
Should I use ChatGPT to write my interview answers?
Use it to brainstorm and rehearse, not to script. AI can help you structure examples and generate practice questions, but reading prepared text on camera sounds unnatural and interviews later in the process will test whether your answers were really yours. Where a coding assessment explicitly permits AI tools, that is different: follow each assessment's stated rules exactly.
How do I know if AI screened my CV?
You usually will not be told directly, although employers in some jurisdictions increasingly disclose automated processing. The safe assumption for any large employer is that software parses and ranks your CV before a recruiter reads it, so tailor every application as if that is true.
Can I retake an AI video interview if it goes badly?
That depends on the employer's settings. Some allow a limited number of retakes per question, others allow none, and completing the interview generally submits it. Check the instructions before you start, and treat your first take as the real one.
How long do AI assessments take?
Typical one-way video interviews involve a handful of questions and take 20 to 30 minutes. Game-based assessments commonly run 25 to 35 minutes across 10 or so short games. Complete them in one calm sitting rather than squeezing them between commitments.
What is the best way to prepare if I have an AI assessment this week?
Fix your environment, prepare and rehearse a small set of structured examples on camera, and do short timed sessions of reasoning practice each day. You can start now with a free practice test in numerical, verbal, logical or abstract reasoning, and build from there.
Facing an AI-led hiring process? The abilities behind the algorithms are the same ones aptitude tests have always measured. Take a free practice test today and turn an unfamiliar format into a familiar one.