72.4% of Recruiters Went Onsite: The 2026 Interview Playbook

The crutch is being kicked out. Google reinstated whiteboard rounds, Amazon now asks candidates to sign a no-unauthorized-AI pledge before their loops, and Gartner says 72.4% of recruiting leaders are back to interviewing in person. If your last twelve months of prep leaned on a Cluely overlay or a second monitor with ChatGPT open, your next onsite is going to be an expensive lesson.

Why 72.4% of recruiters are back in a conference room

72.4% of recruiting leaders now conduct interviews in person to combat candidate fraud, specifically impersonation and third-party interview completion at a scale that broke video loops. That figure comes from a Gartner survey covered in Computerworld, and Cisco's VP of global talent acquisition Scott McGuckin cited "remote work and advancements in AI" as the direct driver.

The tipping point was measurable. Fabric analyzed roughly 50,000 candidate sessions and found AI-cheating signals rose from 15% in June 2025 to 35% by December 2025. In purely technical roles the rate hit 48%. In sales, 12%. That 4x gap is why engineers are getting hit hardest and first.

The employers moving fastest are the ones you would expect:

• Google: reinstated at least one in-person round for technical hires, brought back whiteboard-style problems, and prohibits AI at coding assessments, phone screens, and onsite loops.
• Amazon: candidates are asked in writing not to use GenAI tools "unless explicitly permitted," with disqualification as the stated penalty. A new "Gen AI Fluency" round has been added to some SWE loops in 2026.
• Cisco and McKinsey & Co.: both reinstated in-person rounds over the past year.
• Apple: per Blind, moving to in-person final rounds starting July 2026.

If you have been treating the in-person interview 2026 shift as a rumor, stop. It is policy at five of the highest-signal employers in tech.

What Amazon's no-AI pledge actually means on the day

Amazon's pledge is enforced by humans watching for specific tells, not by software scanning your screen, which means you cannot beat it by hiding the tool better. You have to actually not use it. Amazon's internal guidance names three flags interviewers are trained to catch in real time:

1. Typing during questions before you have started coding.
2. Reading rather than responding, with the cadence of eye movement that suggests a teleprompter or overlay.
3. Wandering eyes, particularly to a second monitor or off-camera location.

Amazon senior exec Colleen Aubrey is on record backing the shift, and violations "may result in disqualification."

The "Gen AI Fluency" trap

Amazon added a Gen AI Fluency round to some 2026 SWE loops. It sounds like a prompt-engineering test. It isn't. Blind reports from candidates who have taken it describe a round that functions essentially as a DSA interview under a different name: you reason through the algorithm yourself, and separately articulate where and how you would use AI in production. The company wants proof you can code without the assistant while proving you understand what the assistant is for.

The winning prep for Gen AI Fluency is not prompt engineering. It is old-school algorithms and data structures with a clear verbal narrative about where AI helps and where it hurts.

Google's code comprehension pilot changes the study plan

Google's May 2026 pilot replaces one traditional coding round with a code-comprehension round for junior and mid SWEs on select US teams, and this single change breaks the LeetCode-only prep loop. Instead of writing a solution to a new problem, you read unfamiliar code, identify bugs, and explain the tradeoffs.

This is deliberate. Overlays like Cluely and Interview Coder are generation tools. They write code. They are worst at reading someone else's code and finding the off-by-one error in the third helper function. Google's pilot targets the exact skill their outputs cannot fake.

Your prep list for Google in-person coding interview rounds should now include:

• Real pull request review, ideally on an open-source repo you did not write. Practice narrating what a function does and where it breaks.
• Debugging drills on code with intentional bugs.
• Reading code aloud. This sounds silly. Do it anyway. Verbalizing "this loop mutates the input array, which will break the caller" is the skill being tested.
• LeetCode medium/hard, but only after the above. The pattern-matching muscle still matters for the whiteboard round they kept.

Note that Meta already rolled out AI-enabled coding interviews in October 2025, replacing one traditional round. The industry is not converging on a single policy. It is splitting between "no AI, prove the fundamentals" (Google, Amazon, Apple) and "AI-assisted, prove you can use it well" (Meta). You have to prep both.

The recruiter bottleneck nobody is pricing in

In-person loops create a throughput problem that shrinks the number of interviews per opening, which means resume-fit precision matters more than at any point in the last five years. Here is the math using Refolk's index.

In Refolk's index there are roughly 23,187 US technical recruiters against 348,463 US software engineers. That is one recruiter for every fifteen engineers, and those recruiters are concentrated in SF Bay Area, Seattle, and LA, which are exactly the hubs where onsite loops are being reinstated. Top employers of those recruiters in-sample include Google, Snowflake, Blue Origin, and staffing firms Experis and K2 Partnering.

When each recruiter has to physically coordinate an onsite instead of clicking a Google Meet link, their monthly candidate throughput drops. The Gartner recruiting in person shift is not just a policy change. It is a capacity cut.

For candidates, the implication is clear. Fewer interview slots per opening means your resume has to clear a higher bar to even get considered, and generic "software engineer with Python and AWS" language will not do it anymore. This is the exact work Refolk takes off you: paste the posting, get your own resume back rewritten for that specific team, with the cover letter drafted and a fit score telling you whether it is worth the recruiter's shrinking slot.

The Cluely time bomb

Anyone who used Cluely in 2025 is carrying a permanent employability liability, because the 2025 Cluely data breach exposed personal data, interview transcripts, and screenshots from more than 83,000 users. That data does not expire. An employer running a match against future applicants gets a cheap, permanent disqualification signal.

For context: Cluely raised $20.3 million including a $15M Series A from a16z, and its founder Chungin "Roy" Lee was suspended from Columbia in March 2025. The tool used GPU-level overlays (DirectX on Windows, Metal on macOS) that rendered below the framebuffer Zoom and Google Meet captured, so screen sharing saw nothing. It worked. It also stored everything, and then leaked it.

The tools now on employers' banned or detected list include Cluely, Interview Coder, Leetcode Wizard, Aura, LockedIn AI, and Final Round AI. The detection vendors employers are actively buying include Fabric, Sherlock, HackerRank's ML plagiarism engine, CodeSignal Suspicion Score, and Codility Similarity Check.

If you used any tool on the first list, assume future employers can find out, and prep accordingly. Do not lie about it if asked directly. Do rebuild the underlying skill before your next loop. And keep in mind: 61% of flagged cheaters in Fabric's 19,368-interview dataset still passed the scoring threshold, meaning honest candidates lost slots to people who did not get caught. That inequity is the other force pushing employers onsite.

How to prep an onsite interview when the safety net is gone

Rebuild the raw skill, rehearse under camera-off conditions, and know cold which round is testing which behavior. Here is the four-week version.

Week 1: rebuild the muscle

• 90 minutes/day of hand-written or whiteboard coding. No IDE, no autocomplete.
• One DSA topic per day: arrays, strings, trees, graphs, DP, then rotate.
• End each session by explaining your solution out loud to a wall, phone recording. Play it back.

Week 2: add comprehension and debugging

• 45 minutes of code-review drills. Open a real PR on GitHub, read the diff, predict what will break.
• 45 minutes of LeetCode mediums, but time-boxed to 25 minutes each. Move on if you stall.
• Practice one system design problem end to end, on paper.

Week 3: mock loops

• Two full mock interviews with a human, ideally an engineer at a peer company.
• One behavioral loop covering STAR-format answers for conflict, ambiguity, failure, and leadership.
• Read Amazon's Leadership Principles and map three specific stories to each one that matters for your level.

Week 4: logistics and taper

• Confirm travel, arrival buffer, and what the loop actually contains. Ask the recruiter for the round breakdown in writing.
• Sleep. Do not add new material in the final 72 hours.
• Prepare three sharp questions per interviewer, tied to something specific about their team.

The resume work sits underneath all of this. If you are applying to Google, Amazon, and Apple in the same week, each posting needs its own tailored resume because the recruiter has fewer slots to give and the fit signal has to be sharper. Refolk writes each version from your own history against each posting, drafts the cover letter, and scores whether the fit is strong enough to be worth applying at all.

The safe harbor nobody is talking about

Sales, customer success, and revenue-adjacent roles are absorbing the crackdown more slowly than engineering, so switching function for six to twelve months is a real short-term option for candidates who cannot rebuild their coding fundamentals fast enough. The 48% versus 12% cheating gap between technical and sales roles means video loops are staying alive in revenue functions long after they die in engineering.

If you have solid product knowledge, strong communication, and a track record of shipping, a solutions engineering, technical account management, or sales engineering role gets you into a hiring pipeline where the rules have not tightened yet. You still have to be good. You just do not have to fly across the country to prove it.

FAQ

Is the in-person interview shift permanent, or will video come back once detection improves?

Best guess: permanent for final rounds at top-tier tech employers, hybrid for earlier stages. Even if detection vendors like Fabric and CodeSignal Suspicion Score close the gap on overlay tools, the 61% of flagged cheaters who still passed Fabric's threshold shows detection alone is not enough. Onsite loops solve the impersonation problem too, and that is a broader risk than any single overlay tool. Expect early rounds to stay remote with stricter proctoring, and expect final loops at Google, Amazon, Apple, Meta, and their peers to stay onsite through 2027 at minimum.

Should I admit to past Cluely use if an interviewer asks directly?

Yes, if asked directly. Lying is worse than the tool use, and the 83,000-record breach means the truth may already be in a database somewhere. The better move is not to volunteer it, but to have a clean answer ready: "I experimented with it during a hard stretch, stopped, and rebuilt my fundamentals from scratch. I can walk you through a problem on the whiteboard right now." Then do it.

How different is Amazon's Gen AI Fluency round from a regular DSA round?

Functionally, minimally different. Candidate reports on Blind describe it as a DSA problem you solve yourself, followed by a discussion of where AI would help or hurt in production. Prep for it exactly like you would for a standard algorithms round, then add a five-minute rehearsed answer on how you use AI responsibly in your daily work: for boilerplate, test scaffolding, and documentation, not for design decisions or novel logic.

Does resume tailoring actually matter more now?

Yes, measurably. When recruiters run in-person loops, their per-opening candidate throughput drops, which means the shortlist shrinks and each resume needs to signal fit faster. A generic resume that cleared a Zoom-era phone screen will not clear a 2026 onsite pipeline. Tailoring each application to the specific posting is the difference between getting the slot and getting a polite no, which is exactly the loop Refolk automates: paste the job, get a resume rewritten from your own history, a cover letter, and a fit score before you spend a week flying to Seattle.