The Onsite Is Back: Prepping for In-Person, No-AI Interview Rounds

Last updated: September 2026.

Quick answer: Several major employers, Google and McKinsey among them, reinstated in-person interview rounds in 2025 to counter AI-assisted cheating in remote loops, and the practice has kept spreading since. If your upcoming interview includes an onsite, the most important preparation is practicing out loud, from memory, with no AI assist. The rest is logistics and stamina. This guide covers both.

The shift is documented and already affecting candidates across tech and consulting. Google's CEO publicly confirmed the move. McKinsey followed. The stated driver, at both companies and across the broader wave of employers that tracked the same logic, is the same: remote interview integrity became hard to maintain as AI-assisted cheating in video-based assessments climbed sharply through 2025.

For candidates who have grown accustomed to a second screen, or whose prep has been mostly reading rather than speaking, this changes what "prepared" actually means.

Why Did In-Person Rounds Come Back?

A candidate on a video call can run a model on a second screen or use whisper-mode tools to receive prompted answers during the call. Both became common enough that employers started treating remote results as unreliable signals of actual ability.

The logic behind the shift is direct: a candidate in the same room at a whiteboard has none of those options. Proximity handles the cheating problem without requiring sophisticated detection. That is what drove the move at Google and McKinsey, and the trend spread from there.

What Does an Onsite Actually Look Like?

An onsite is a full sequence, usually three to six sessions back-to-back, each with a different interviewer. Here is a representative structure for a software engineering role:

Consulting onsites replace the coding rounds with cases, sometimes presented as written packets rather than verbal prompts. Product management onsites often include a product critique or a short presentation. The formats differ; the dynamic stays the same. You are assessed across multiple hours, by multiple people, with no safety net and no way to buy time.

The hardest part for most candidates is not any single session. It is staying sharp and consistent from the second session to the fifth.

What Do Onsite Interviewers Actually Watch For?

The rubric covers the obvious: did you solve the problem, did you answer the behavioral question fully. But there is a second layer that only shows up in person.

Interviewers watch how you handle getting stuck. If you hit a wall in a coding round, freeze, and emerge thirty seconds later still frozen, that reads differently than someone who says "I'm not sure this is the most efficient approach, but here's where I'd start" and keeps working. The latter demonstrates how you operate when things are hard.

They also notice transitions. If you get stuck in a coding round and then walk into a behavioral interview two minutes later, can you reset? Candidates who carry stress from one session into the next are visible. The ones who compartmentalize stand out.

And they notice engagement. An interviewer asking "what's your reasoning there?" wants to see whether you can articulate your thinking and adapt when pushed. That exchange is the actual signal they came to see, more than the final answer.

The No-AI Drill Checklist

The only way to build these skills is to practice them the hard way: out loud, with a timer, no notes, no model.

Behavioral preparation

• Write down ten experiences that each show one of: judgment under pressure, taking ownership of a failure, a conflict you navigated, or something you taught yourself from scratch
• Practice each one out loud targeting two minutes per answer, and time yourself, because most candidates run either 45 seconds (too thin) or five minutes (too long)
• After each answer, immediately ask yourself "why did you make that call?" and answer that out loud for another 60 seconds
• Run at least three full behavioral sessions with no notes and no pausing

Technical preparation (SWE and quant roles)

• Work problems on paper rather than a keyboard — slowing down forces deliberate thinking
• Talk out loud the whole time, especially when you are unsure
• Practice "working through failure" explicitly: at the 20-minute mark on any problem you have not solved, say out loud what you would try next and why
• Time every problem; 35 minutes is a common onsite budget per round, so stop and review at the cutoff whether or not you finished

Day-of preparation

• Confirm the full address, floor, and the name of your contact at the front desk
• Map your backup transit option in case your primary route fails
• Eat before you arrive — you will talk for hours and hunger degrades thinking in ways that sneak up on you
• Silence your phone completely and leave it in your bag during each session

The reset drill (usually skipped)

• Deliberately practice an emotional reset under time pressure: work a hard problem until you get stuck, then immediately pivot to a behavioral question and answer it in full. The point is building the habit of switching gears when you are frustrated and mid-task.

The Biggest Preparation Mistake

Over-preparing for content and under-preparing for delivery.

Knowing 150 LeetCode problems does not help much if you freeze when asked to work one out loud with someone watching. Knowing every STAR story you have does not help if you cannot feel when two minutes is up and it is time to stop. Content is the entry ticket; delivery is what differentiates candidates once they are in the room.

Delivery is a physical skill. You build it by doing it physically. Speaking out loud to an empty room feels uncomfortable and slightly absurd. Do it anyway, repeatedly.

How Out-Loud Practice Transfers to the Real Thing

Speaking practice works because it builds the pathways you actually use in an interview. Saying an answer is harder than writing one. Thinking out loud under a time limit is harder than thinking quietly with no clock. Reading a guide and feeling prepared is a real feeling that does not accurately predict your performance in the room.

PokeBot's mock interview tool runs structured, timed sessions across behavioral and technical formats, with feedback on what you said and how you said it. The reps are close enough to the real thing to make the actual interview feel familiar.

For behavioral questions, the STAR format gives you the right structure. The mistake is memorizing scripted answers, which sound rehearsed and collapse under follow-up questions. The right approach is knowing your stories well enough to run them from memory and cut them when the interviewer signals they have what they need.

If you want to work on your closing, the list of questions to ask at the end of an interview gives you a solid bank to practice with. Onsites typically end with the hiring manager session, and how you close that conversation leaves a real impression.

One Week Out: What Actually Matters

With one week before an onsite, content review is mostly done. Spend the week like this:

1. Two full mock behavioral sessions, timed, no notes, no AI
2. Three to five coding problems on paper, worked out loud start to finish
3. One clean pass through your strongest stories for this specific role, checking that each one is tight and concrete
4. Logistics confirmed: address, backup transit, outfit decided

The week before an onsite is not for learning new things. It is for converting what you already know into something you can actually perform under pressure.

Score your resume free, create your PokeBot account