The era of solo player AI is over. Adapt is the integrated coworker that empowers every team to work AI native together.

Anyone can tag @Adapt in Slack to answer a quick question, schedule a task, prepare for a board meeting, or build the dashboard you’ve been waiting on for weeks.

Built for business users, powerful enough for engineers.

Get $100 free credits for your team

^{Free credits awarded when you use your work email}

You have been told the machine is objective. That is the entire pitch. Humans are tired, biased, and inconsistent, so companies handed the first pass to software that applies the same standard to every applicant at the same depth, whether you are resume number four or resume number four hundred.

Researchers at IIT Jodhpur decided to test that. They took one Product Manager resume, copied it ten times, and ran it past Claude, GPT, and Gemini with an identical job description. Same candidate. Same role. Same instructions. The average scores came back 69.4, 50.5, and 72.8 out of 100.

That is a 22 point spread on a resume that never changed a word. Depending on which vendor the employer happened to buy, the exact same candidate is a strong yes or a clear no. And when the researchers benchmarked all of it against three human recruiters with six to ten years of experience, the machines disagreed with the humans too, significantly, in every single condition they tested.

Here is why this matters for your search, and it is not the reason you think. This is not a "the system is rigged, give up" issue. It is the opposite. Once you understand that your score is a reading rather than a measurement, several things you have been doing become obviously wrong, and a few things you have been avoiding become obviously right.

When you get filtered out in the first pass, your instinct is to conclude something about your resume. Too junior. Wrong keywords. Bad formatting. So you go rewrite it.

Slow down. In a system where the identical document scores 50 at one company and 73 at another, a single top of funnel rejection carries almost no diagnostic signal. You cannot debug a scoring engine from one data point, especially when you never see the score, the rubric, or the model.

The rule I would use: never rewrite your resume based on fewer than ten rejections from comparable roles. One no is noise. A pattern across ten is data. Most people burn weeks rebuilding a document that was working fine, and the rebuild loses the specifics that were actually doing the work.

This is the most useful finding in the study and nobody is talking about it. The researchers ran a "reduced context" condition where the models got a stripped down job description with the responsibilities, location, and company details removed. That condition produced the widest disagreement of the entire experiment.

Worse, the scoring did not just get noisier. It got inflated. GPT's average score jumped from 64.3 to 82.9 when the job description got thinner, roughly a 29 percent bump for candidates it knew less about. The human experts did the opposite. Given less information, they got more conservative.

Practical translation: a vague, three bullet job posting is not just lazy writing. It is a signal that whatever is screening you on the other end is running with worse information and producing less stable results. Weight your effort toward postings with real detail, real responsibilities, and a real company description. Those are the reqs where a strong application is most likely to be read as strong.

Look at the spread. Across conditions, the AI scores clustered between roughly 50 and 83, with standard deviations of 7 to 15 points. The human recruiters ranged from 30 all the way to 90, with deviations up to 25 points.

Read that again, because it is the whole ballgame. Machines bunch everyone toward the middle and high middle. Humans actually discriminate. They use the top of the scale and the bottom of it.

If you are genuinely excellent, algorithmic screening is working against you. It is not filtering you out, it is flattening you into the same 70 something band as the merely adequate. Which means the single highest leverage move in your search is not resume optimization. It is any route that puts a human in front of your candidacy sooner. A referral, a recruiter conversation, a hiring manager who already knows your name. Not because the machine is unfair, but because the machine cannot tell the difference between good and great, and a person can.

Your portfolio could be losing more to taxes than you might realize. On August 6, Range’s CFPs and CPAs share the portfolio moves that can help you maximize your after-tax returns — join us live, and bring your questions for Q&A.

Reserve your free spot

^{This webinar is for informational purposes only and does not constitute investment advice or a recommendation to buy, hold, or sell any security. Forward-looking statements involve risks and uncertainties. Past performance is not indicative of future results. Range defines "high earners" as households with income over $300k.}

Here is the part that should make you feel better. Underneath all the variance, there was one point of near universal agreement.

When the researchers asked each evaluator to explain its scoring scheme, relevant work experience came out on top for essentially everyone. GPT put 25 to 30 percent there. Claude weighted technical skills and relevant experience at 30 to 35 percent. One human expert put 45 percent on relevant work experience alone. Every other category, education, certifications, achievements, extracurriculars, split the remainder.

So stop spending your energy on the volatile stuff. Formatting tricks, keyword stuffing, and clever section headers are fighting over the 10 percent that moves unpredictably. Relevant experience density in the top third of page one is the input that scores well no matter who or what is reading.

Concretely: your most role relevant job should be visually dominant, described in outcomes, and impossible to miss in a five second scan. If the role you want is different from the role you have, the translation work belongs at the top, not buried in a summary line nobody weights.

Because scoring shifts with the context supplied, the same resume submitted against a different req at the same company is genuinely a different evaluation. Different job description, different detail level, sometimes a different model entirely if that team uses different tooling.

People treat a first pass rejection as a permanent verdict from the organization. It is not. It is one reading, from one configuration, on one day. If a second role at that company fits, apply to it. Do not agonize, do not reference the earlier rejection, just submit a version tuned to the new description.

You can reproduce the core finding in about ten minutes, and it will tell you more than another round of resume advice will.

Take a job description you are targeting. Paste your resume and the full posting into an AI assistant with this prompt:

Note the score and the weights. Now open a fresh chat and run the same thing again, but delete everything from the job description except the title and three core requirements. Compare.

The gap between those two numbers is your exposure to screening noise. If your score swings fifteen points or more, your resume is leaning on context that a thin posting will not supply, and you need the relevant experience to speak for itself without help. Also read the weighting breakdown carefully. If the model is allocating heavily to a category where you are thin, that is a real signal worth acting on, unlike the rejection email that told you nothing.

Most people feel behind. But the people that don’t, aren’t smarter. They’re just better informed.

The Future Today is a daily news briefing for people who want clarity.

In one concise newsletter, you’ll get the most important tech news, learn why it matters, and what it signals about what’s coming next.

One email. Five Minutes. Stay ahead of 99% of the world.

Join Free Today →