There is no longer one answer to this question, and that is the whole story. In 2026 the rules genuinely differ by employer, and two of the largest have landed in opposite places.
Google is piloting a coding round where candidates use Gemini as an approved assistant. Meta began rolling out AI-enabled coding interviews in October 2025. Amazon's candidate guidelines say not to use generative AI tools during interviews unless explicitly permitted, and that failing to follow this "may result in disqualification from the recruitment process." Canva goes further in the other direction and expects candidates to use AI, then grades how well they direct it.
So the correct answer to "can I use AI in a coding interview" is: find out for this specific company, in writing, before the round. Guessing is now the actual risk.
Key takeaways
- The rules diverge by employer. There is no industry standard and there probably will not be one soon.
- Google's pilot covers junior and mid-level software engineering roles on selected US teams, starting with Google Cloud and the platforms and devices unit, using Gemini as the approved assistant.
- Google's new round is "code comprehension": reading, debugging and optimising an existing codebase rather than writing an algorithm from scratch.
- Interviewers in AI-allowed rounds evaluate AI fluency, described as prompt engineering, output validation, and debugging. Using the assistant badly is now a way to fail.
- Amazon's position is the opposite: no generative AI unless explicitly permitted, with disqualification on the table.
- When in doubt, ask the recruiter in writing. An explicit yes is the only safe basis for using AI in a live round.
- Unauthorised real-time assistance is a separate thing entirely and remains a bad idea regardless of company.
The policy map, as it stands
| Company | Position | Detail |
|---|---|---|
| Piloting AI-assisted rounds | Gemini as approved assistant, junior and mid-level SWE, selected US teams | |
| Meta | Rolling out AI-enabled coding interviews | Began October 2025 |
| Canva | Expects AI use | Grades how well candidates direct it |
| Amazon | Prohibits unless permitted | Guidelines state disqualification may follow |
This table will age. Treat it as evidence that the question has no universal answer rather than as a lookup you can rely on in six months.
Why Google changed the round, not just the rules
The interesting part of Google's pilot is not that AI is allowed. It is that the interview itself changed shape.
The new format is a code comprehension round: you are given an existing codebase and asked to read it, debug it, and improve it, with Gemini available. Interviewers assess what the reporting describes as AI fluency, meaning prompt engineering, output validation, and debugging.
That is a coherent response to a real problem. If the job now involves working alongside an AI assistant most of the day, then an interview that bans the assistant is measuring a task nobody actually performs. And the traditional whiteboard algorithm question was the single most compromised format, because it is exactly the kind of self-contained problem a model solves instantly.
So rather than defending a question that AI had broken, they changed the question to one where the assistant is present and the difficulty moves elsewhere: to understanding unfamiliar code, spotting when the model is confidently wrong, and knowing what to keep.
Note the scope, though. This is a pilot, on selected teams, for junior and mid-level roles. It is not Google's whole process, and it does not generalise to your interview at a different company.
Why Amazon went the other way
Amazon's guidelines tell candidates not to use generative AI during interviews unless explicitly permitted, and note that failing to follow this may result in disqualification. Reporting on their internal guidance indicates interviewers are briefed on signals to watch for: typing while a question is being asked, reading rather than responding, eyes tracking away from the camera.
The reasoning is different but also coherent. Amazon interviews heavily on behavioral signal through its leadership principles, and those answers are supposed to come from your own experience. A model can generate a fluent story about a time you disagreed with a manager. It cannot generate one that is true about you. If the thing being measured is your actual history, an assistant is not augmentation, it is fabrication.
The two positions are less contradictory than they look. Google is allowing AI on a round about working with code. Amazon is barring it on rounds about what you personally have done. Both are protecting the integrity of the specific thing being measured.
What this means for how you prepare
You now have to prepare for two possible rounds, and they need different practice.
If AI is allowed
The skill under test is not coding speed. It is direction and verification.
Practice reading code you did not write. Pull an unfamiliar open source repository and give yourself twenty minutes to explain what a module does and where a bug would most likely hide. This is the actual test in a comprehension round, and almost nobody practises it.
Practice catching the model being wrong. Ask an assistant to solve something slightly off-spec and see how long it takes you to notice. Confident, plausible, subtly incorrect output is the failure mode you are being assessed on. If you accept it in a room, that is the finding.
Narrate your prompts. In an AI-allowed round, the interviewer is watching how you direct the tool. Saying "I'm going to ask it for the edge cases first, because that's where I expect this to break" is signal. Silently pasting and accepting is also signal, of the wrong kind.
Know when not to use it. Reaching for the assistant on something you could obviously do yourself reads as dependence. Part of fluency is judgment about when the tool is worth the round trip.
If AI is not allowed
The requirement is the old one, and our guide on practicing for a coding interview still applies. Write code without autocomplete. Talk while you solve. Get comfortable being stuck out loud rather than silently.
The one addition for 2026: expect to be asked how you use AI in your work, as a conversation topic even in rounds where you cannot use it. That is a different question with its own traps, and we covered it in how to answer "how do you use AI in your work".
The question you should actually ask
Before any technical round in 2026, send the recruiter a version of this:
"Quick question before the round: is the use of AI assistants permitted during the interview, and if so, which ones? I want to make sure I follow your process correctly."
Three reasons this is the right move. It gets you a written answer, which protects you if it comes up later. It reads as conscientious rather than suspicious, because it is the question of someone who wants to follow the rules. And it is the only reliable way to know, because the policy is now company specific and often round specific.
If the answer is no, or you get no answer at all, assume no.
The line that has not moved
None of this changes the thing that was already true. There is a difference between an interview where the employer has decided the assistant is part of the test, and an interview where you have quietly decided that for them.
The first is a legitimate format, and it is probably where a chunk of the industry is heading. The second is misrepresenting your ability to someone making a decision about you, and it now carries real risk: detection tooling has improved, employers have added in-person rounds specifically to close the gap, and the consequence when it surfaces is not a lower score but removal from the process.
We wrote about where that line sits in is using AI during a live interview cheating. The short version has not changed. Prepare with AI as much as you like. Perform as yourself, unless you have been told in writing that the assistant is invited.
FAQ
Can I use AI in a Google coding interview?
In the pilot, yes, with Gemini as the approved assistant. The pilot covers junior and mid-level software engineering roles on selected US teams, beginning with Google Cloud and the platforms and devices unit. It is not the whole of Google's process, so confirm with your recruiter which format your specific round uses.
Will Amazon really disqualify me for using AI?
Amazon's candidate guidelines state that generative AI tools should not be used during interviews unless explicitly permitted, and that failing to adhere may result in disqualification from the recruitment process. Treat that as the operative rule.
How would they even know?
Interviewers are briefed on behavioral signals: typing during questions, a consistent lag before answers, reading rather than speaking, eyes tracking off camera. Some employers also use integrity monitoring tools. Detection is imperfect, but the relevant point is that the consequence when it does surface is exit from the process, which is a poor trade against a modest boost on one answer.
What is "AI fluency" and how is it assessed?
In AI-allowed rounds it refers to prompt engineering, validating what the model returns, and debugging its output. Practically: can you direct the tool toward the right problem, notice when the answer is wrong, and fix it. Accepting confident but incorrect output is the main way candidates fail these rounds.
Should I mention that I use AI at work?
Yes, and expect to be asked. Being straightforward about how you use it, and where you do not trust it, reads far better than pretending you work without it. The trap is describing yourself as dependent on it rather than as directing it.
Does this apply to take-home assignments too?
Policies on take-homes vary just as much and are often stated in the brief. Read it. Where it is silent, ask. Where you are permitted to use AI, say so in your submission notes and describe what you used it for, which almost always reads better than leaving it ambiguous.
Practice being asked the fourth question
Whichever kind of round you get, the thing that decides it is depth. AI-allowed rounds test whether you can tell when the model is wrong. AI-free rounds test whether the knowledge is yours. Both come apart at the same place: the follow-up question after the answer you rehearsed.
Round Zero runs a live mock interview that asks follow-ups when an answer is thin, then scores the transcript with your own words quoted back as evidence. The free tier includes one full scored session. Find out where your depth runs out before an interviewer does.