How to use AI across a tech job search

You've probably already used AI somewhere in your search, and you've probably also had a moment of doubt about whether you were allowed to. That doubt is reasonable, and the reason it's hard to resolve is that there's no longer one answer. Meta tells candidates it expects them to use an AI assistant in certain coding interviews. Amazon tells candidates not to use one at all. Anthropic asks you to use Claude on your application and then bans it in the take-home. All three are true at the same time, in the same month.

This article covers what actually changed, and where the ethical and practical line sits. It then walks through eight phases of a search with specific ways to use AI in each one, plus a copy-pasteable prompt at the end of most phases. Read the first two sections before anything else, because everything after them depends on knowing which kind of process you're in.

The people quoted here run hiring at companies you're probably applying to. Brendan Foody is CEO of Mercor, which builds AI-based candidate assessments. Tomer Cohen is CPO of LinkedIn. Dhanji R. Prasanna is CTO of Block. Marty Cagan of the Silicon Valley Product Group has trained product leaders for two decades. Sam Stone is VP of product at EvenUp and Sarah Koo is a senior director of product at Rippling, and both screen PM candidates weekly. Nikhyl Singhal was VP of product at Meta and Google. Dana Ingraham runs agent products at Harvey. Every quote plays the exact moment it was said, and the company policies below are quoted from the companies' own pages, verified in July 2026.

What changed in the last year

Two things happened at once, and they pull in opposite directions.

The first is that applying got almost free. Brendan Foody, whose company builds assessments for exactly this problem, describes the loop plainly.

The employer side of that is measurable. HR Dive reported in January 2026, citing LinkedIn research, that 93% of recruiters said they planned to grow their AI use in 2026. SHRM's 2025 Talent Trends survey of 2,040 organizations found 43% now use AI somewhere in HR, up from 26% in 2024, with 51% using it to support recruiting.

The second change is that AI ability became a thing companies interview for. Tomer Cohen, who runs product at LinkedIn, put it in the context of rebuilding how his organization hires and evaluates people.

Foody's framing for how assessment should respond has become the standard argument on the permissive side.

Not everyone agrees this predicts much. Dhanji R. Prasanna, CTO of Block, runs the new format and is openly unsure it measures the right thing, which is worth holding onto if you're about to spend a month learning Cursor instead of practicing how you explain a system.

The policies contradict each other, so check every time

Here's the part that most articles get wrong by picking a side. As of July 2026, major employers have published policies that flatly disagree, and several of them differ round by round inside the same company.

Meta expects it on select rounds. The Meta careers hiring process page says "Select roles now include an authorized AI assistant within CoderPad during technical interviews" and that "Candidates are expected to use this AI assistant as part of the interview." It also spells out what's in the box: "Candidates use the built-in AI assistant in CoderPad, which includes Claude, ChatGPT, Gemini, and Meta's models." Supported languages include Java, C++, C#, Python, Kotlin, and TypeScript. Meta engineers told CoderPad's blog in October 2025 that "AI has fundamentally changed the way software is developed... Our modernized AI-enabled interview is a better representation of our work and of our mission." One caution: Interviewing.io reports that Meta's own candidate prep materials describe AI use as optional while the careers page says "expected." Don't try to resolve that yourself. Ask your recruiter.

Canva expects it too, and published the clearest engineering-side explanation of why. Simon Newton wrote on the Canva engineering blog in June 2025 that "we now expect Backend, Machine Learning and Frontend engineering candidates to use AI tools like Copilot, Cursor, and Claude during our technical interviews," and that the problems were redesigned for it: "These problems can't be solved with a single prompt; they require iterative thinking, requirement clarification, and good decision-making."

Cognition encourages it. Emily Cohen, Cognition's leader of people and operations, told Business Insider: "I guess this is like asking a kid to take a math test without a calculator." She was specific about scope: "For the bulk of building something similar to what you would do on the role, you can and should use AI tools."

Amazon bans it. Amazon's candidate guidance, reported by IT Pro in March 2025, reads: "To ensure a fair and transparent recruitment process, please do not use GenAI tools during your interview unless explicitly permitted." Non-compliance "may result in disqualification from the recruitment process."

Anthropic does both, in the same process. Its candidate AI guidance page, updated July 2025, encourages AI at the application stage: "Please create your first draft yourself, then use Claude to refine it." Then it closes the door for take-homes: "Complete these without Claude unless we indicate otherwise. We'd like to assess your unique skills and strengths. We'll be clear when AI is allowed (example: 'You may use Claude for this coding challenge')." And for live interviews: "This is all you–no AI assistance unless we indicate otherwise." A company that sells the model bans the model in its own live interviews. If you take one fact from this article, take that one.

Google is piloting, and hadn't started as of publication. Brian Ong, Google's VP of recruiting, said in an emailed statement reported by Entrepreneur in May 2026: "We're always evolving our interview processes to ensure we're recruiting and hiring the best talent. As a part of that, we're rolling out a pilot for software engineering interviews to be more reflective of how our teams are operating in the AI era." The pilot is scoped to select junior and mid-level software engineering roles in the US, on a code-comprehension round, using Gemini, in the second half of 2026. Google hasn't published a candidate-facing page about it.

Shopify is permissive, on thinner evidence. There's no published Shopify candidate policy. What exists is Farhan Thawar, its VP of engineering, in The Pragmatic Engineer: "You let them use whatever they want... If they don't use a copilot, they usually get creamed by someone who does." He also said, "I love seeing the generated code because I want to ask them, what do you think? Is this good code?" Treat that as one leader's description of how he runs it rather than as a company rule.

Ask your recruiter in writing, and treat the answer as binding

Do this before every round, even when you think you already know the answer. Before every technical round, take-home, and case, send your recruiter a short message asking what's allowed. Get it in email or in the ATS message thread, not on a phone call, so there's a record. Then do exactly what they said, including when what they said is more restrictive than what the company's public page implies.

Where the line is

Strip away the moralizing and the line is simple. Use AI freely for research, tailoring, drafting, and rehearsal. Use it openly wherever you've been invited to. Don't run a covert real-time assistant during a live interview.

The reason to stay behind that line has nothing to do with ethics, which is worth saying plainly, because the ethical case doesn't persuade anyone who's already decided. The practical case is that a covert overlay doesn't work on the interviews you actually want to pass.

Interviewing.io ran a study on this, published in 2024, in which candidates used ChatGPT during technical interviews. Pass rates were 53% for the control group with no AI, 73% on verbatim LeetCode questions, 67% on modified versions of them, and 25% on custom company-specific questions. Read that last number again. On the questions a real company writes for itself, AI-assisted candidates did worse than half the control rate. The sample was 32 interviews in total, which is small, so the exact percentages are rough, though the direction is clear enough to act on. The study also noted something that cuts the other way: "not a single interviewer mentioned concerns about any of the candidates cheating."

The reason the 25% matters more each quarter is that companies are moving toward exactly those custom questions. CoderPad's Amanda Richardson, speaking to Interviewing.io in October 2025, described the shift: "We're seeing the very fast death of the algorithmic take-home. That's over. But I think what we are seeing is people replacing that with more of a project-based take-home if you will." The same piece put 20 to 30% of CoderPad's customers as now using AI inside interviews.

Then there's disqualification. Amazon's language is explicit that using a tool without permission can end your candidacy. Greenhouse launched a product called Real Talent in June 2025, broadly available February 2026, that includes identity verification through a partnership with CLEAR. Vendors are selling detection too: Fabric publishes its own detection rates for AI-assisted interviews, though it sells detection software, gives no denominator or false-positive rate, and hasn't had those numbers audited externally. Treat the specific figures as marketing and the trend as real.

Boyden, an executive search firm, argued that covert assistance defeats the point of the conversation: "Recruiters and hiring managers depend on interviews to evaluate not just what candidates say, but how they think, react under pressure, and communicate. Cluely's invisible assistance can mask weaknesses and inflate competence, undermining the entire vetting process." That's a search firm defending interviews, so weigh it accordingly, though it describes the same thing the Interviewing.io numbers show.

The tools themselves have a short and messy history. Cluely was founded in April 2025 and raised a $5.3M seed the next day, co-led by Abstract Ventures and Susa Ventures. Its founder publicly claimed $7M in ARR and then retracted it: "is the only blatantly dishonest thing i've said publicly online, so this is my formal retraction." TechCrunch reported the actual figure at around $5.2M in March 2026. He also said of the launch, "I can't say if it's a mistake, but maybe we launched too early." That's the company you'd be trusting to sit inside a live interview with you.

Detection tools get it wrong, and that can hurt you unfairly

If you're a non-native English speaker, AI-detection tools are a risk to you even when you write everything yourself. Liang and colleagues, in a peer-reviewed paper in Patterns, tested detectors on TOEFL essays written by humans. The detectors "misclassified over half of the TOEFL essays as 'AI-generated' (average false positive rate: 61.22%)," against 5.19% for essays by native writers, and 97.8% of the TOEFL essays were flagged by at least one detector. The samples were small, 91 TOEFL essays and 88 native-writer essays, and the native comparison set was US eighth-graders, so it isn't a clean apples-to-apples test. It's still the best-evidenced finding in this whole area, and it's the reason to keep your own voice and your own irregularities in anything you write. More on that below.

Phase 1: strategy and positioning

Before you tailor anything, decide what you're tailoring toward. AI is bad at this part and good at everything downstream of it, so doing it yourself first pays for itself repeatedly.

Pick one archetype and write it in a sentence: the level, the function, the company stage, and the problem you want to be handed. Then pull three real job descriptions for that archetype and read them for the strategic problem the role exists to solve, which is usually one layer under the responsibilities list. Only then use AI, and use it to check whether your positioning already matches or doesn't. Our guide to positioning yourself for a job covers the archetype work in detail.

Volume discipline belongs here rather than later, because it's a strategy decision. Nikhyl Singhal, who was VP of product at Meta and Google, makes the case that low-volume, high-effort applying is the version that compounds.

Thirty applications you actually researched will outperform three hundred you didn't, and the thirty leave you with an opinion about the market that you can use in interviews.

Phase 2: company and role research

Build a dossier per target company, and let AI do the gathering while you do the judging. A useful dossier has the last ninety days of company news, whatever you can infer about the roadmap from public releases and job postings, the funding stage or the business unit's position, two or three named people you'd work with or near, and one paragraph on the strategic problem the role exists to solve.

Use a tool with live web access for this. A model answering from training data alone will confidently describe a funding round from two years ago as though it happened last week, and repeating that in an interview is expensive.

It also helps to know what the filter looks like from the other side. Michal Peled, a technical operations engineer at HoneyBook, built an agent to do first-pass sourcing and read the actual screening rubric into it out loud.

Every one of those filters is about location, recency of activity, title adjacency, or tenure, and none of them is about how your resume is formatted. That's a fairly typical shape for a screening rubric, and it tells you where your effort is worth spending.

Phase 3: application materials

Tailor every resume against a specific job description. That's the highest-return use of AI in the entire search, and it's also the one most likely to produce something that reads like it came from a machine, so both halves matter. Our resume tailoring guide covers the mechanics.

On what the ATS actually does, the honest answer is less dramatic than the folklore. Enhancv, which sells resume products, surveyed 25 recruiters across Workday, iCIMS, Lever, Greenhouse, Teamtailor, Bullhorn, Phenom, Beeline, and LinkedIn Recruiter in September and October 2025. It found that 23 of 25, or 92%, said rejections are manual or triggered only by eligibility filters, never by formatting or missing keywords, with content-based auto-rejection at 8%, plus or minus 6 points at 90% confidence. That's a vendor survey with 25 respondents, so it's a signal rather than a settled fact, and it points away from the idea that a formatting quirk silently kills you.

What does happen is grading. Workday's own responsible-AI page describes HiredScore as using "an A, B, C, D grading system where A indicates the closest match," while explicitly disclaiming that it's an automated decision tool. So software grades your resume before any human opens it, and the grade shapes what a human sees first. Relevant keywords earn you a better grade, though they won't rescue a resume with nothing behind them.

Then there's the writing itself. Tomasz Tunguz has the line that should govern every AI-assisted draft you send.

A grammatically perfect specimen is what a recruiter has already read many times that morning. Write the first draft yourself, badly if necessary, then use AI to tighten it, then put your own irregularities back in. Anthropic's own candidate guidance describes the same order of operations: first draft yourself, then refine.

Cover letters are worth writing only when the application asks for one, and then only if you have something specific. Take-home instructions are binding, including the ones about AI and the ones about time limits, and interviewers do notice when you ignore them.

Phase 4: outreach and networking

Pick twenty to forty people rather than four hundred, and write to them individually. AI drafts, you edit, and you never let a tool send on your behalf. LinkedIn's own data, published with no methodology, says its AI-assisted messages "see an overall 44% higher acceptance rate and are accepted over 11% faster" than messages drafted without AI. That's the platform reporting on its own feature, so read it as a directional claim about personalization rather than a measured effect.

What's changed more is what a good outreach message contains. Dana Ingraham, who runs agent products at Harvey after a stretch at Airbnb, described what actually gets a reply from her.

Lazar Jovanovic at Lovable says the same thing from the hiring side, and adds that the format itself gets attention.

Take that with the obvious caveat that a Lovable employee likes Lovable apps. The general version holds anywhere, which is that an artifact showing you engaged with their actual product will get further than a paragraph about how much you admire it. See getting your foot in the door at a company for the fuller version.

Referrals are worth the effort. Referred candidates are roughly 7% of applicants and account for 30 to 50% of hires, which tells you the channel is dense rather than telling you your personal odds.

Phase 5: interview preparation

This is where AI earns the most, and where almost nobody uses it hard enough. Do it in three layers: content, then mock interviews, then delivery. Most candidates do the content layer and skip delivery, then lose on a communication rating they never get to see.

For content, generate a company-specific briefing rather than reading generic question lists. Reid Robinson, a PM at Zapier, described building these for his wife's search.

For behavioral rounds, write your stories yourself and then have AI interrogate them. Our behavioral interview guide has the question set. What matters is asking for the right kind of feedback, and Hilary Gridley of Whoop has the best version of that.

Apply that to every story you plan to tell. Paste it in, ask the model to restate your point and your evidence, and see whether what comes back is what you meant. If it isn't, your story is the thing that needs fixing.

Take-homes and prototypes

If the take-home allows AI, the failure mode has shifted from not finishing to not owning what you submitted. Sam Stone, VP of product at EvenUp, watches candidates fall through the structure they didn't write.

Sarah Koo at Rippling sees the same thing in prototypes, and her example is the one to rehearse against.

So before you submit anything AI helped you build, walk every screen and be ready to say why each element is there, or delete it. Stone also points out that the hours you put in have stopped predicting the result.

Delivery is the layer to protect. Record yourself answering, then watch it. AI can transcribe and count your filler words, and a person or a tool built for rehearsal can tell you whether you sounded like you believed what you were saying. If you're short on time, our 24-hour interview prep plan sequences this properly.

Phase 6: during the live interview

The rule here is that you use almost nothing in real time unless you were invited to.

Some things are perfectly fine. Keep a one-page card of your own notes in front of you, the same way you'd bring notes to any meeting. Ask your interviewer clarifying questions. Record or transcribe if you ask first and they agree, which they often won't for confidentiality reasons, and no is a complete answer.

An overlay listening to your interviewer and feeding you answers, a second person or device off-screen, or any tool you didn't disclose will end the process if it's noticed, and the 25% figure above is the reason it probably wouldn't have helped you anyway.

Where a company invites you to use AI, the interviewer is usually grading how you work with it. Meta's format grades problem solving, code development, verification and debugging, and communication, which means narrating what you asked the model and why you accepted or rejected what it gave you is part of the performance. Canva's stated point is that its problems "require iterative thinking, requirement clarification, and good decision-making," so treat the assistant as a colleague whose output you're responsible for checking. Say out loud when you don't trust an answer, and then verify it.

Phase 7: after the interview

Send a thank-you note within 24 hours, under 150 words, referencing one specific thing from the conversation. AI can tighten it for you, though it can't supply the specific detail, so take two minutes of notes right after the call while you still remember what was said.

Then debrief yourself honestly, which is the step almost everyone skips. Write down each question you were asked, what you actually said, and where you felt yourself losing the thread. Feed that to a model and ask what a hiring manager would have concluded, rather than asking whether you did well. If you're waiting and hearing nothing, we have a separate piece on asking for feedback after an interview.

Phase 8: the offer and the negotiation

Get the offer in writing before you discuss it. Then benchmark it against more than one source, because any single source is one company's sample.

Most people leave money on the table by not asking. A Resume Genius survey of 1,000 US full-time workers, run through Pollfish, found that 55% accepted their initial offer without question, and that 78% of those who did negotiate say they ended up with a better offer. That's a resume vendor's survey, with the methodology disclosed, so treat it as an indication rather than a measurement.

Pew Research gives a more careful picture of what negotiating actually gets you. In a February 2023 survey of 5,188 employed US adults, about two-thirds of those who negotiated got more than the initial offer, but only 28% got the full amount they asked for, and 35% ended up with only what was first offered. Just 30% asked at all. So the realistic expectation is a partial win, and the thing that most affects your outcome is whether you ask at all.

AI is genuinely useful in three places here: assembling the benchmark data, drafting the ask so it's one number with a reason attached, and rehearsing the conversation including the pushback. It's useless at deciding your walk-away number, which you should write down before the call and not revise during it. Our salary negotiation scripts and the guide to negotiating equity in a job offer cover the specific language.

Eight things not to do with AI

How this changes by role

Product management

The PM loop changed more than any other, and Marty Cagan describes what replaced the old case question.

So learn one prototyping tool well enough to build under time pressure, and prepare answers to Cagan's four questions about compliance, sales, and customer demand, because those are where the round is actually won. Combine that with the Koo warning above and you have the whole brief: build it, then test it, then be able to defend every element on the page. Our product sense interview guide covers the reasoning half.

Engineering

Engineering is where the bifurcation is sharpest, and where asking in writing matters most, because a Meta round and an Amazon round call for opposite behavior. Prepare both ways: keep your fundamentals sharp enough to work without assistance, and get fluent enough with an assistant that you can narrate your prompting and catch its mistakes out loud. The Prasanna quote above is your reassurance that the fundamentals still carry most of the weight. Our engineering interview guide goes deeper on the formats.

Design

Hiring managers now read your portfolio partly for whether you can work with these tools, so show at least one piece where AI was in your process and say what you did with the output rather than what the tool produced. Keep the reasoning visible. The failure mode is a portfolio full of generated visuals with no evidence of decisions behind them.

Marketing and data

Visible AI fluency is close to table stakes in both. The strongest version is a specific workflow story: a process you rebuilt, what it used to cost you, what it costs now, and what broke along the way. That's a better signal than a list of tools, because it survives follow-up questions.

Tools, and what each one is for

Prices move constantly and none of the pricing claims in circulation held up to checking, so this is a list of purposes rather than a buying guide. Verify current pricing yourself before you commit to anything.

What might already be out of date

Everything above was verified in July 2026, and a few parts of it will move before the end of the year.

Company policies change quarterly, and they're the load-bearing part of this article. Google's pilot had not started as of publication, so Google's software engineering interviews may already work differently than described here. Meta's careers page says candidates are "expected" to use the AI assistant while Interviewing.io reports Meta's own prep materials call it optional, and that tension was unresolved at publication. Anthropic's guidance page was last updated in July 2025, and Amazon's language dates from March 2025. None of that replaces asking your recruiter, which is why that instruction sits where it does in this article.

Tool pricing drifts and the vendor landscape churns, so treat the tool list as categories rather than recommendations. Many "best AI job search tools" roundups are published by the vendors themselves or by their affiliates, which is worth remembering when one of them reads unusually enthusiastic.

Detection technology moves in both directions, so whatever is undetectable this quarter is a research problem next quarter. The numbers throughout this article come from mixed-quality sources and are labeled in the text where they appear: a 32-interview study from 2024, a 25-recruiter survey by a resume vendor, engagement figures published by LinkedIn about LinkedIn's own product with no methodology, and detection rates published by a company that sells detection. Where a widely repeated statistic couldn't be traced to a real source, it isn't here at all, which is why a few numbers you may have seen elsewhere are missing.

Frequently asked questions

Is it cheating to use AI on my resume?

No, and several companies now tell you to. Anthropic's candidate guidance asks you to write the first draft yourself and then use Claude to refine it, which is a good default everywhere. What gets you in trouble is claiming experience you don't have, and that's a lie whether a model wrote it or you did.

Can I use AI during a job interview?

Only if that specific company has invited you to for that specific round, and you should get the answer in writing from your recruiter. Meta expects it on select technical rounds, Canva expects it for backend, ML, and frontend engineering candidates, and Cognition encourages it. Amazon prohibits it and says non-compliance may result in disqualification. Anthropic bans it in take-homes and live interviews while encouraging it on your application. There's no safe general answer, which is the whole point of this article.

Will an AI-detection tool flag my resume unfairly?

It might, particularly if English isn't your first language. The peer-reviewed study by Liang and colleagues found detectors misclassified over half of human-written TOEFL essays as AI-generated, at an average false positive rate of 61.22%, against 5.19% for native-writer essays. The samples were small. The practical response is to write in your own voice, keep your own quirks, and use AI to tighten rather than to generate.

Do applicant tracking systems automatically reject my resume?

Much less than the folklore suggests, on the evidence available. Enhancv's survey of 25 recruiters found 92% said rejections are manual or triggered only by eligibility filters. Grading before a human looks is real, though, so relevant keywords help your position in the queue, and eligibility questions about work authorization and location genuinely can knock you out.

How many applications should I be sending?

Fewer than you probably think, with much more work in each one. Singhal's argument is that ten researched applications teach you something you can use on the eleventh, while a hundred untailored ones teach you nothing.

My interviewer used AI to write the job description. Doesn't that make this even?

Probably, and it isn't a defense if you break a stated rule. Both sides are using these tools heavily, which is what Foody describes above. The asymmetry that matters is that the employer sets the rules for their own process, and you agreed to them when you accepted the interview.

Should I tell them I used AI?

Volunteer it when you were invited to use it, since narrating your process is often part of the grade. You don't need to announce that you used a spell-checker on your cover letter. If you're ever asked directly, answer honestly, because getting caught in a lie afterwards can cost you the job itself.

The whole search on one page