AI and Résumés: What the Document Can't Tell Anyone About You

Back to Blog

Nobody can promise you a job.

Not from a résumé. There is no set of instructions that, followed correctly, guarantees you a callback — and anybody who implies otherwise is selling something. This is a game of percentages. You are trying to get the best outcome you can in one specific moment, in front of one specific person, and that's all it has ever been.

I want to explain why I'm so sure of that, because it isn't cynicism. Some of it I learned from the wrong side of the desk.

There are two ideas underneath all of this, and I'd rather name them here than have you find them at the end.

On a résumé, the part that matters most is the part no document can check. The facts on it can at least be checked — a reference confirms the dates, a background check confirms the rest, when anybody bothers. What no document you write checks is the half you actually need believed: that you're good with people, that you don't need watching, that the queue doesn't rattle you. On that, you are the sole source, making claims about yourself, with nothing standing behind them. That is one of the two things this article keeps running into. The other is what happens to any piece of advice once everybody follows it.

Hold onto that. It explains the shelf. It explains why team player stopped meaning anything. It explains why a referral moves the needle further than a better-worded bullet. And it explains what a model fundamentally cannot do for you.

The shelf

This was the early nineties. I'd come out of college — where nobody mentioned résumés once, in all that time — and I needed one. So I went to the bookstore, because I no longer knew how.

They had shelves of books about it. Not two or three. Shelves. And my first thought, standing there, was that I didn't think there needed to be so many books. I thought it was a standard format. That's what I'd been taught: there was a class in high school that went through it in detail, whether your name and address go at the top, what you put on the Dear ___ line, whether to whom it may concern was acceptable. We spent real time on those little points.

The books were confident and they disagreed with each other. Too much information. Too little. Be precise. Get to the point. And each had its telltale list — *if you're going for this position, these are the skills they're looking for, so brush up on your Excel.*

I've since gone back and checked what was actually on that shelf, and my memory holds up better than I expected. Yana Parker's Damn Good Resume Guide. Tom Jackson's The Perfect Resume. The Adams Resume Almanac, which came out in 1994 and ran to 767 pages and promised "a review of the pros and cons of the various formats."

And it really was sliced up the way I remembered. One publisher, VGM Career Horizons, ran a whole series of them — a separate book for engineers, for nurses, for social workers, for architects, for people over fifty. There was one for women returning after raising children and one for soldiers leaving the Army. Forty-odd titles, each about 150 pages, each certain.

So I went to the library instead, where I could take out four at a time and try to triangulate.

They didn't agree, and somebody checked. In December 1994 two researchers published a paper in the Bulletin of the Association for Business Communication with the title *"Résumé Content and Format — Do the Authorities Agree?"* — and the answer was that while there was agreement on many things, "there is also disagreement in several areas." The disputes I remember from the shelf are the ordinary ones: one page or two, whether to open with an objective, chronological against functional.

But the shelf argued with itself in plain sight, and you don't need a study to see it. Here is the Adams Resume Almanac, 1994, selling itself in a single sentence:

Read that twice. The formats are a matter of pros and cons — and the strategies will definitely get you noticed. Both promises, one sentence, same book, 767 pages. The disagreement wasn't a flaw in the genre. It was the product, sold alongside the certainty.

There's one more thing about that shelf that I only see now. The biggest career book in the shop wasn't a résumé book at all. It was What Color Is Your Parachute?, and its whole argument was that organising your job hunt around the résumé is the wrong frame to begin with.

Then I noticed the thing that has bothered me ever since.

I couldn't find a book on how to write a résumé to become a doctor. Or a lawyer.

Those are the jobs people most want. If the résumé is the gate, that's where the gate has the highest value — and nobody was selling a key to it.

It isn't because those jobs are exclusive. It's because they get checked.

I want to be careful here, because there's an easy misreading and it's the opposite of what I mean.

Nobody phones a cashier's last employer to ask how they were with change. Nobody calls to ask whether they gave receipts, or whether they were pleasant on a Tuesday afternoon. The reference check, if it happens at all, confirms the dates.

A firm hiring a lawyer checks everything. What school. Did they finish. Where they interned, and under whom. Whether they're admitted. Because that firm is about to put this person's name on a case it can be held legally accountable for. A hospital does the same and more, because the consequences of not doing it are somebody's life. It is a whole verification apparatus, much of it statutory, and the document is almost the least of it. The résumé matters less to them, because so much else has already answered the question.

Which tells you where the industry lives. The rulebook gets sold wherever the checking stops short of the question. A board can certify that a nurse is qualified and still leave an employer guessing about everything else; where there's no board at all, the paper carries the whole weight on its own. Secretary. Junior accountant. Customer service. Cashier. The format industry is thickest exactly where no institution verifies anything.

What I actually looked at, when I was the one looking

I've been on both sides of this. I've handed in applications and I've read them.

From the reading side: a résumé written in pencil gets attention, but not the kind you want. Misspellings won't win anyone over — and that isn't snobbery about spelling. It's that this is supposed to be your best foot forward. If this is the version you chose to hand me, I'm now wondering about the version you didn't.

But here's what I never did. I never sat there like a teacher with a checklist. *Did they use bullet points. Are the bullets in the right place. Are the indents correctly spaced.* Not once. And in all the years I did that job, nobody ever told me to.

What I actually wanted was simpler, and almost nobody writes it down. Most of those jobs, we taught you. Our systems were unique; you'd have to learn ours regardless. So when I saw cashier experience, I wasn't thinking this person can operate a till.

I was reading the number of years. Fifteen years on a register means fifteen years of other people deciding not to move you. A supervisor, an assistant manager and a store manager all had the option, every one of those years, to put somebody else on the front and this person somewhere they wouldn't meet a customer. Nobody took it. That isn't a claim the applicant made. It's a record of decisions made about them by people who had to live with the consequences.

What it establishes is a floor, not a quality. At minimum this person is not a problem in front of customers, because fifteen years is far too long for three levels of management to have missed it. And then it forks, and the two branches look nothing alike. Either they're genuinely good with people — or they can't stand the public, and they're fast, accurate, and never the reason the line isn't moving, which is worth keeping on a register over somebody warmer who miscounts a drawer.

Both of those are worth hiring. They are not the same hire, and the résumé cannot tell you which one is standing in front of you. A question can. That is the whole of what I was doing when I read those pages: getting to a floor, and then finding out which side of it somebody was on.

And this is the ordinary way everybody reads a résumé, which is exactly why nobody bothers to write it down. Hiring a coder, you look at their coding; a cook, their cooking. A record doesn't carry one meaning. It carries a meaning against the opening it's being read for.

Which is also the answer to the objection somebody will raise, that fifteen years in one seat with no promotion is a signal of its own — that the same people who kept them also declined to move them up. That only bites if you're screening for trajectory — which for some jobs is exactly what you're doing, and for this one is not. If what I need is a cashier, fifteen years of being kept says this person will still be on that register in two years, and that is the entire question the record is being asked to settle.

And "I'm excellent at dealing with people" is not a sentence anyone can put in a bullet point and be believed.

And here is what a stack of forty actually looks like

The posting lists the skills it wants. So every résumé lists those skills. A, B and C, then X, Y and Z, on all of them.

And nobody told them to copy it out. The advice is more reasonable than that — if a posting asks for something and you actually have it, make sure they can see that you have it. That's fair. It's what I'd tell somebody. It's what I'd want, reading.

It still produces forty résumés carrying the same list in the same order.

And the advice for when you can't match it is just as sensible: say why you're right for the job anyway. I'm a fast learner. *I've worked with W, which is close enough to X that I'll pick it up quickly.* Reasonable. Frequently true. I've said it myself.

You will read I'm a fast learner eleven times before lunch.

So both routes close. The people who matched all sound like each other, the people who didn't match all sound like each other, and the phrase that was supposed to rescue you goes the way team player went — it stopped being something you were and became something everybody says.

Then you sort by experience and a dozen of them have ten years or more. Some have twelve. Some have fifteen.

At that point, the years aside, what's left to tell them apart is the name at the top, the school, and what they studied.

So you go looking for a person in there. You start reading the bottom section — were they in the band, did they take archery, do they hike at weekends — because after everything above has been standardised, the extracurriculars are the only place any individuality survived.

And the years don't settle it either. Fifteen years doesn't mean good — it can equally mean somebody counting the years down, who does the job and not one thing more. You can't tell which from the number, and the number is most of what you've been given.

Call that homogeneous, because that's what it is. They tell you how many sentences. How many words in the skills section. Which verbs. Those are rules, and when everyone follows them, everyone lands in the same narrow ballpark. It stopped being about what to write and became about how you write it.

Here's the part that stayed with me. When somebody writes their own résumé their own way, humanity comes through it. Their sense of humour comes through it. And anyone who could make me laugh while I was reading résumés — of all the things to be reading — I wanted to meet that person. That's the thing the polishing takes out first.

Why no advice can guarantee you anything

There's a second reason the guarantees are impossible, and it has nothing to do with software.

Hiring managers are not the same as each other.

Some are exhaustive. They want to see everybody who applied, double-check, triple-check, be sure they're getting the best person available. Others work differently: the first person who piques their interest is the person they call, and once the list is full the backlog stays a backlog.

Same résumé. Same words. Completely different outcome, decided by which of those two people happened to open it. No advice controls for that, and no study I found separates the two readers.

So where did the rules come from?

Short answer: a machine, and not the one you would guess.

The instructions about how a résumé should look — plain text, no bolding, no italics, no underlining — arrived in the nineties attached to machines, and never to a machine of yours. A 1989 patent takes an optical scanner as given and describes getting a document into a state it can read. By 1998 the job boards were giving a plainer reason: typed into their boxes, formatting *"will not appear correctly online."*

The part that turned out to be interesting isn't the history at all, and it turned into its own article — Read the Label. It's that five résumé sites owned by a single company still tell you how a résumé has to look, and the same company's own pages disagree with each other about what its own templates can do.

For this article you only need the two ends of it, which I can show you: in the nineties the shape of the document was set by somebody else's equipment. Today a shape is still what gets sold to you as expertise. No document connects a 1990s column to a 2026 style guide, and I looked for a long time before admitting that.

But I can tell you what the whole run of it was for, and it isn't what I assumed when I started. Strip the grievance out and every layer of this was somebody solving one real problem: how do you make a piece of paper findable by a person who is looking for it? Fax-clean, then scanner-clean, then parseable, then visible to a search. Each layer went on because the one under it worked. That is not a scam. It's an infrastructure — and it is genuinely baffling from outside, because nobody hands you the rules and half of them were never decisions in the first place.

And then the loop starts

Here's the part I circled for years without being able to name it.

What these companies do is, on its face, reasonable. They go to people who hire and ask them questions — what attracts you to a résumé, what puts somebody on the list you want to know more about. There are a lot of applications and a lot of capable people to sift, and the market keeps shifting, so doing that research repeatedly is genuinely useful. I respect it.

Then it becomes advice. *We asked a lot of people and they said they like X, Y and Z. So make sure X, Y and Z are in your résumé. These are the beats. These are the stopping points.*

And everybody does it. Which produces the stack of forty I described above, where the only variables left are your name, your school and whether you were in the band.

You can follow all of the advice, hit every beat, and still not get on the list. Because once everyone is hitting the beats there's nothing left to distinguish anyone, so the person reading has to start looking somewhere else — at whatever hasn't been optimised yet.

The advice destroys its own value at the exact moment it succeeds. That's not a flaw in the research. It's what happens to any signal once everybody sends it.

And to be fairer to them than I've managed so far

Before any of this, one thing about who's reading. These are companies. They have marketing departments, they are not non-profits, and none of that is an accusation — you can look up who owns what in an afternoon, and they will tell you themselves that they can't guarantee you a job. It's in their terms of service and their privacy policies — not on the page with the template counter on it, which carries no qualifying language at all. But the person reading their pages is often out of work, and marketing read by somebody out of work does not land as marketing. It lands as a life jacket. That's the whole reason to be exact about what's promised versus what's sold — not to score points off a company, but because somebody is going to spend money they haven't got and then feel that they failed. They didn't. They bought a service that does one part of this well and was sold to them as though it did all of it.

So: pros and cons, and you decide. Which means I have to do the pros properly, and I've left out the obvious one, which is that it was worse before.

Standing in front of that shelf, there was no way to tell which of the forty-odd confident books was the good one. You picked one and went with it. What the sites did was generalise it, and that was a real service. They walk you through — this section is for this, that section is for that. They impose a limit, which matters more than it sounds; my own first résumé listed the clubs I was in at school and what I felt I'd learned from being in them, because nothing was stopping me. And they surface the rough basics, which is most of what most people need.

If they weren't there we'd be back in the aisle, lost. And AI now does that job better than any of them: it can do the research, pull the practices common across the big sites, check your draft against those guidelines, handle the formatting, and take something you've already written and say it more clearly. That last one is not the same as writing it for you, and it is genuinely useful.

And here's the concession that matters most, because it means nobody has to be the villain.

There is no way to avoid homogenisation if you are giving out best practice.

Not because the practice is wrong. Because of how it gets executed. Publish the standard, and the people who do the minimum will produce the minimum — and the minimum is by definition the most repeatable thing on the list. Best practice, honestly given, produces sameness by construction. Nobody has to be acting in bad faith for the field of forty to look the way it does.

Which means the flair, the part that's actually you, was always going to be your job. And for anyone who can't see it in themselves — which is most of us about our own work — that is precisely what a coach or one of the free services is for.

Credit where it's due, and then the other half

I assumed "free" on these sites meant you could build it but not keep it. That's true of some of them and flatly untrue of others, and the split isn't the one the marketing suggests.

Monster's builder is free, full stop — PDF, Word or plain text, unlimited, and in their words *"There are no hidden fees, and you don't have to enter credit card information to access your document." Kickresume: "…you can download your documents as many times as you want! This applies to both premium and free users."* FlowCV promises "No watermarks or branding" on its free tier. Teal, Indeed and Canva all let you walk away with a real PDF for nothing. If that's what you need, take it — it's a good deal and I'm not going to pretend otherwise.

One more fact, and I'll leave it to you. Those four and Monster are the same company. BOLD's own about page names nine brands it owns — those five, plus CareerBuilder, FlexJobs, Bold.pro and My Perfect Cover Letter — and gives away free under one brand precisely what it charges for under the others.

Somebody measured this happening

I worked that out from the reading side. It turns out you can watch it occur at scale, with numbers.

Researchers at Yale followed a freelancing platform that switched on an AI cover-letter writer, across 5.5 million bids. People who got access to it saw more callbacks at first — that estimate is soft, significant at the 10% level rather than the 5%, and I'll take it as suggestive.

What's estimated more firmly is what happened underneath. The correlation between tailoring a letter and getting a callback fell by 51%. With receiving an offer, 79%. The best writers gained least — "27% smaller gains than the weakest ones." And the individual advantage tapered out after about two months.

Two months. That's the shelf life of an optimisation once everyone can perform it.

But the mechanism is not mine, and I had it as a guess. I'd written that employers probably moved toward signals that are harder to fake. The authors measured it: *"employers shifted toward other signals less susceptible to AI influence, such as workers' past work experience,"* and the correlation between callbacks and workers' review scores rose by 5%. When the document stopped carrying information, the hiring side went looking for somebody else's account of you. That is this entire article, arrived at from the other direction and with a number on it — and I'd rather say so than keep the credit for a hunch.

Including this one

While I'm being honest about it: this article is subject to its own law.

There's a piece of formatting advice further down with a real field trial behind it, and it works right now, for the people it was measured on. It will work less well the more people know it, and if enough people read this it will stop working altogether.

I don't have a way around that and I'm not going to pretend I do. Any advice that works by distinguishing you stops working when it's widely held.

The score nothing rejects you on

There's one more belief underneath all of this, and it's the one that generates the most anxiety: that there is a formula, that following it works, and that software is checking your compliance and dropping you below a line.

Across the five applicant tracking systems whose documentation I could actually read — Greenhouse, Workday, Ashby, Taleo, SmartRecruiters — not one documents rejecting anybody because of a résumé content score. Greenhouse says it outright about its own AI matching layer:

Note what they thought worth promising: including candidates in lower match categories. That's the exact fear, denied by name.

And once you've read a stack of these yourself, the reason it's denied stops being a courtesy and starts being a limit. A parser can see the record. It cannot see the branch. The dates, the employer, the title are all on the page, and a machine reads them about as well as I did. Which side of the fork you are on is not anything a parser can read off the page, in any format, at any keyword density. There is nothing there to score. So a score can only ever be a floor measurement, and you cannot reject somebody for clearing a floor.

Automatic rejection is real. It runs on things you typed into a form — a yes/no question, a checkbox — or on facts the employer already holds. Workday's own worked examples are declining candidates who failed the background check, or who aren't eligible to work in the country. None of that is a verdict on your bullet points.

And where a machine does grade a résumé, look at which direction it runs. Workday's grading tool has an automation attached called Candidate Fast Lane, which *"provides the ability to automatically advance top-rated applicants through the hiring process to the Hiring Manager review phase."* The other three recipes in that group are email automations — two notifications and an outreach mailer — so there is no auto-decline anywhere among them. A low score doesn't reject you. It leaves you in the ordinary queue.

Put the two paragraphs together and the software has drawn the line for you. Everything it acts on against you is a fact with a definite answer — did the check come back, are you eligible to work here. Everything it acts on from a score, it acts on only in your favour. Failing to be readable will still cost you — that's the other floor, and it's real. But the score, the one everybody worries about, has nowhere adverse to go. I can show you the asymmetry. I can't tell you anyone designed it that way.

So whose score is everyone optimising for? It belongs to the company that sold you the checker — and they say so themselves, on the page where they're selling it.

Rezi, which is a résumé builder and gives you a score out of a hundred: *"But let me be clear: real ATS systems don't give your resume an official score." And, in the next breath: "Recruiters won't see the same percentage you see in Rezi, and there's no universal 'resume score' hiring teams use behind the scenes."*

Jobscan, which sells you a match rate: *"Keep in mind, when you submit your resume to a job, the ATS doesn't score your resume. The match rate in our scanner is a visualization tool to help you measure the strength of your resume before you apply."*

A résumé builder and a résumé scanner, both stating that there is no universal score of the kind they are selling you against. Jobscan still describes the product as *"By simulating the 'scanning' process, Jobscan identifies exactly what is missing or unreadable, allowing you to optimize your document before it ever hits a recruiter's database."* Which is accurate, and worth sitting with. The score is real. The thing it's a simulation of isn't.

And now you can see why it has to exist. Nobody on the hiring side needs a number for the branch — they know it comes out of a conversation. But a floor score is unsellable as a floor: *your record is legible and it matches the posting* is not a product, so it gets sold as something else. So the checker sells you a number that carries itself like it has measured the part that matters, when the only thing any machine can see is the part that was decidable all along.

The famous statistics are the same kind of object. *Seventy-five percent of résumés never reach a human* — I went looking for a primary source and couldn't find one: no publication, no methodology, no dataset, a citation trail ending around 2012. Recruiters spend six seconds on a résumé — thirty people, in 2012, studied by a job site's own head of user experience, never peer-reviewed. The same company revised it to 7.4 seconds in 2018.

What the machine actually does, and why it still matters

I've just spent several paragraphs saying no software is judging you. That's true, and it is not the same as saying the machine doesn't matter.

It doesn't judge you. It does have to be able to read you.

I've never really seen a "short list," and I'm not sure the phrase means what people think. My guess is it's mundane: somebody sits at one of these databases and types in what they need. Ten years' experience, and speaks French, because a lot of the clientele do. Whoever comes back in those results is the list.

That's not hitting beats. That's a search query — literal keyword matching on one layer, semantic similarity on another, exactly as the vendors document it.

Which is where the real failure lives, and it isn't a verdict. The free-text half will find your title wherever it sits on the page. The filters won't — sort by years of experience, narrow by job title, and a record that never parsed into those fields isn't in the set being sorted. You are in the system, complete, and not in the results. Nothing was decided about you at all. You were overlooked — and being overlooked is not a decision somebody made, which is why you will never be told it happened.

Notice what that does to the floor. A parse failure isn't a low score. It's that the record never became readable at all — you didn't come up short on the question nobody can answer from a page, you never got as far as presenting the one that was answerable.

So there are two gates, and most of the advice you'll read runs them together.

The first gate is being findable. Somebody types what they need into a database and reads what comes back. You don't write the query and you can't know it — but whether your title sits where a parser expects a title, whether your dates parse, whether the document survives being read by something that isn't a person: that part is yours, all of it.

The second gate is being worth reading, and then worth a conversation. That one is the branch, and you don't own it — and it's worth being exact about why, because the reason is not that somebody judged you and found you wanting.

It's fit, and fit runs both ways. Sit on the hiring side long enough and you learn that the expectation almost always overshoots. You had a picture of who'd walk in; nobody walks in matching a picture. You knew that and you hoped anyway. And then, twenty minutes in, you realise it isn't that they're short of anything — it's that they wouldn't thrive here, and putting them in it would be doing them a disservice. So you keep looking. That never gets written down, because there's no line on the rejection email for you'd have been miserable and I could tell. It's a gut feeling, and it is most of what the second gate is made of.

The two gates fail differently, and the difference is what you can do next. If you were never retrieved, nobody decided anything — which is why it feels like shouting into a room with nobody in it. That is exactly what it was.

That is why the instruction is make it legible before you make it compelling. Legible is the gate you can get through; compelling is the branch, and the branch was never going to survive the trip through a parser however well you wrote it.

And that is the commercial shape of it. Legible has a ceiling — you either cleared it or you didn't. Compelling has none. A product needs somewhere to go next month, and only one of those two provides it. The gate you can actually clear is the boring one.

And it is why I put the one piece of formatting advice with a live field trial behind it down here, at the boring gate — which turns out to be the wrong place for it, as I'll get to. The Behavioural Insights Team sent 9,022 applications to real vacancies and found that presenting experience in years rather than dates raised positive callbacks by 4.8 percentage points — a 14.6% improvement — measured against a CV that showed an unexplained gap.

What's left when everything else has been optimised

Now I have to argue against myself, because there's a version of formatting I know matters, and I know it from the reading side.

If you're going through a mountain of paper and one is on good stock, with a font that's pleasing to the eye, you give it another once-over. From the paper alone. And if there's nothing wrong with it, you don't mind reading it. That's just the truth. Thin paper, obnoxious font, and you're not giving it much time regardless of what's on it.

I was ready to print that as my own testimony and admit nobody had tested it. Somebody had.

Arnulf, Tegner and Larssen, 2010, in the European Journal of Work and Organizational Psychology. Ninety people — half professional recruiters, half students — ranking twelve candidates for one vacancy, each candidate appearing an equal number of times in three layouts: formal on white paper, coloured, and "creative." The authors' phrase is résumés of equal quality. What the rotation isolates is the layout.

Design may matter substantially — may is the authors' word, not mine. And the formal one won over the creative; where the coloured arm placed, the abstract doesn't say. The plain one on white: the arm with nothing to sell, since colour blocks, two columns and skill ratings drawn as little filled circles are what a template gives you that a blank page doesn't.

The second finding is aimed at the readers rather than the writers. The professionals were only slightly less swayed than the students; experience doesn't appear to make you immune to it.

And it fits the loop, though not in the direction I expected. Once everyone's text is polished to the same targets, the reader weighs whatever hasn't been polished — and in 2010, on paper, that was the physical object. What surprised me is which arm won: the plain one, the default, the layout every convention already prescribes. Standing out lost. On a screen, something else becomes the residual. That's the honest version of design advice. Not pretty helps. Whatever nobody has optimised yet is where the attention goes — and which way it cuts when it gets there is a separate question. Here it cut toward the plain one.

Which the résumé builders are no help with

Their AI writes content. The design is a fixed template with a settings panel, and they say so. Enhancv: "Enhancv is an AI resume builder with content feedback, not a design tool." And the detail that finishes it — resume.io will not let you change the font. From their own help pages: "Currently, our templates do not support different font styles." To change typeface, you change template.

Thirty years on from the bookstore, still presets.

A general-purpose model will build the document you actually want and hand you the file, which is the genuinely new thing. Keep your content in plain text and render it twice: a plain single-column version for the upload box, and a properly set one for a person. Properly set is doing real work there — not decorated. Generous margins, one good typeface at a readable size, consistent spacing, real selectable text, nothing important hiding in a header or an image.

So can AI get you the job?

*(Worth saying before I answer: this is a site about AI, I use these tools daily, and I used one throughout this piece — for research, checking, and arguing with. I've also never submitted a résumé written by AI or gone job hunting with one. I've researched it thoroughly. I'm not going to tell you I've run the experiment when I haven't.)*

You hear it everywhere. AI will take your résumé writing through the roof. That's what made me want to say something.

And here is why I spent so long on where the rules came from, because it turns out to be the answer.

Everything that infrastructure asks of you is written down, public, and boring — plain structure, the title where a title goes, the words the posting actually used, dates that parse. None of it is secret. It just isn't taught, and most people applying have never been told the standard exists, let alone that they're being measured against it.

A model is very good at the boring public thing, and that is close to the whole of its honest value here. It can read a posting and tell you what it's actually asking for. It can find the conventions, because finding published things is what it does best. It can show you the weaknesses in your own draft, which is hard to do by staring at it yourself. It can retype a twenty-year-old scanned document into something a parser can read, which removes a class of failure that has nothing to do with your qualifications.

It helps you comply with a standard nobody told you about. Say it that plainly and it sounds small. It isn't — it's the whole first gate.

Here's the one hard number, and it's smaller and more specific than the marketing implies. In the largest randomised trial anybody has run on the tools themselves — nearly half a million jobseekers, published last year in Management Science — an algorithmic writing assistant produced 8% more hires and 10% higher wages in the resulting matches. The tool was explicitly not generative; the paper says so. The measured mechanism was 5% fewer writing errors. It was a proofreader, and it worked.

That is close to the whole of the solid evidence. One randomised trial on the tools themselves. Eight percent. Mostly proofreading.

Most of the rest is the industry marking its own homework. There is independent research — the two trials, the design study, the Yale paper, the 1994 audit — and there is very little of it, which is the actual problem. And you can hear the shape of it once you know to listen: nothing can guarantee you a job — and then the but comes. But our research shows.

What it can't do

Can it make you sound like a different person? A better version of yourself?

No. Not really. It's there on paper. You are what you've written.

Now, the fair objection: a model can ask you questions. It can. Nothing stops you pasting your résumé in and saying interview me about this for twenty minutes. That's a real answer and it deserves a real one back.

Watch what happens.

"You were a cashier for fifteen years — so you're good with people?"

Not established. Nobody moved you, so you're at least not a problem at the front — but *good with people* is one of the two branches. The record supports the floor. It does not tell you which side of it somebody is standing on. A model takes it as settled anyway, because that's what the correlation in its training data says — and then it builds you three bullet points about your passion for customer service.

The difference isn't that the model made an inference. I made one too. It's that the model stopped at the first branch and called it an answer, where the whole job is knowing there are two and asking which.

The model can ask. What it can't do is know which answer is the one that matters, because that comes from having done work, and from having sat across from other people who've done work. It has data about jobs and no experience of having had one. That isn't a criticism of the technology — a model only knows what you tell it about you. It cannot be surprised by you, because nothing has happened to it.

Put that beside everything above. For thirty years the résumé has been written for readers who could only see what was on the page — scanners in 1989, parsers in 1998, search boxes and embeddings now. AI is the best tool anyone has ever had for that job.

It is still that job. It will make the file better. It cannot make the file bigger than you.

The thing you can't write about yourself

There's a phrase I've come to hate, and the reason is the whole point.

Team player.

It's been used so wrongly by so many people that it's overshadowed what it meant. It is not *nudge nudge, can you carry the team through a bad season.* That's not it at all.

It means: can you trust your teammates, and can your teammates trust you?

Can everybody focus on their own portion of the job and know — wholeheartedly — that the part you're filling is being handled with as much care as they'd bring to it themselves. Your teammate has your six, just like you have theirs. Everyone does their job, so nobody spends attention on the person next to them wondering whether it's getting done. And now you get to focus on yours.

And it runs the other way, which is the half people leave out. When somebody on the team is having a hard time of it — when life has taken its turn with them and decided not to pull any punches — everybody pitches in and carries the weight until they're back on their feet. Not because it's anywhere in the job description. Because that's what the thing is — and you already know they'd do it for you. Not that they might. That you know.

That is a job requirement, and you cannot credibly write it about yourself.

Not because you're not allowed. Because anybody can type "team player" — that's exactly how the phrase got ruined. It has to be observed, or vouched for, or discovered in a conversation. It is unprovable by you, on your own paper, by construction.

And it's the thing I actually wanted to know, every single time.

How it comes out

Here's what I've watched break it open, across two careers of reading these and then meeting the people behind them. What follows is a pattern, not one afternoon — I've seen it run the same way often enough to trust it, and that is the only standing I'm claiming for it.

After you've done something for years it stops feeling like a skill. It's the part of the job you don't think about. So it comes out flat: Yeah, I was a cashier. I rang people up.

Then somebody whose actual work is helping people with this looks past the flat sentence and asks a real question. Not are you good with people. Something with somewhere to go.

What are you hoping to get out of moving into an office role?

Well — better money, obviously. But more than that, it's a slower pace. And I'd get to be the face of the place for the couple of minutes somebody's standing there. Put them at ease if they've come in agitated. I like that part. And honestly, I've been standing up for ten years and I'd like to sit down.

None of that was on the résumé. It wasn't being withheld — it just isn't the sort of thing that survives translation into Cashier, 2015–2025.

And keep asking, and sometimes what comes out has no box on any form.

A man comes through the same line week after week, caring for his father through chemotherapy. And at some point she tells him she lost her own mother the same way — not to make it about her, but because she can see he's decided nobody in that building has the first idea what he's carrying, and she wants him to know that somebody does.

That isn't a feeling somebody had once. It's a thing they do. They put their own worst year out on the counter, on purpose, when they can see somebody standing in theirs. In ninety seconds. With a stranger. Between transactions.

That is as good a predictor as I have ever found of how somebody will be with your customers, and there is no field for it on any application, and no keyword, and no beat to hit. And they would never think to write it down, because to them it isn't a skill. It's just what you do when somebody comes through looking like that.

It only exists out loud, and only if somebody asks.

And no — nobody writes that down. You don't put I once told a customer about my mother on a résumé, and no decent coach would suggest it. What they do is help you find the sentence underneath it. The true, ordinary, sayable version:

That's it. Two plain lines, nothing dramatic about them, and both of them true. You would never have written them about yourself — from where you're standing that isn't a quality, it's just Tuesday. Somebody else heard the story and recognised what it was evidence of. On the page they still read as a claim like any other, and no wording fixes that. What changed is that you now know what to say when somebody finally asks.

Which is the whole thing, in one move. A model can phrase. Give it those two sentences and it will polish them nicely. What it can't do is sit across from you, hear *it was fine, you get to know them*, and know that there's something under it worth going after.

Those little points are what somebody else starts writing down for you. Because that's what moves a hiring manager — not just that you can do the job, but that the people already there would be glad you turned up.

Where to find that person, free

One thing first, because I don't want this misread. If you already know your own strengths and weaknesses, this section isn't for you. Some people have a clear-eyed account of what they're good at and how they work around what they're not. They don't need any of this. This is for the person who's looking for a way to get better results and can't see what they already have.

For the conversation itself:

Your public library. Many systems pay for a service that takes your draft and returns it — in Chicago's words, *"…within a business day you will receive a copy returned with feedback from a writing expert"* — plus live coaching hours, which is the part that matters here. Free with a card.

American Job Centers, about 2,300 of them. No income test and no unemployment claim for basic career services. Whether résumé help is offered is a local decision rather than a federal guarantee, but plenty of states promise it in writing.

If you served, or you're a military spouse: Hire Heroes USA, at no cost, remotely, nationwide, with rewriting the résumé as the core service.

If you're over fifty: the AARP Foundation's BACK TO WORK 50+ is free coaching, and membership isn't required — "We serve AARP members and non-members alike."

If you have a disability: your state's Vocational Rehabilitation office, in every state. Counselling and job-related services are free by regulation.

The arithmetic nobody selling you anything will do

Across 165 million applications, one recruiting platform's benchmark of its own funnel — self-reported, and I'm holding it to the same standard I held the seventy-five percent — puts it at about eight percent getting past the first screen and half of one percent receiving an offer. Referrals convert at eleven times the rate of applying cold.

What the referral number means is simpler than it looks, and it is not about who you know in some clubby sense. It's about what it costs the person doing it.

Say a friend or an old coworker mentions their place is hiring, and agrees to walk your résumé into HR. That person has to keep going back to work with those people. If you're hired and it goes badly, some of that lands on them. So when they put your name forward they are spending something they can't easily get back, and everybody in that building knows it. That's why it carries weight the document never can — and it holds at one remove too, for somebody outside the company who knows people inside it.

And a letter of recommendation is the third version of it. A good one speaks to your character in a way you cannot speak to it yourself — because the moment you say it about you, it stops being evidence and becomes a claim. That's the team player problem again, and it has the same solution: somebody else says it.

So a referral does informally what the bar and the medical board do formally: somebody with something to lose stands behind you. It is the one thing the document cannot do for itself.

Nobody says that out loud. Nobody writes it down. And you will almost certainly never find out it happened. It's one more thing that was settled before your document was ever the point.

What I'd say at the end

The shape was never really yours. In the nineties a machine set one. Today somebody sells you another and calls it expertise — $19.99, normally $50, special just for you.

And the advice keeps eating itself. Whatever everyone is told to do stops working the moment everyone does it, which is why the guidance can never stop being republished and can never quite be right.

So no — nobody can guarantee you anything. Not because the advice is dishonest, but because of the thing I started with. There is one specific person on the other side of it, and they are not the same as the next one. The manager who reads all forty and the manager who stops at the first interesting one open the same document on the same afternoon and reach different conclusions, and nothing you did caused that. It's percentages.

And percentages cut the other way too, which is the part I'd want somebody to hear. Even if the next person does exactly what you did, they're not you. The thing that makes you worth hiring — that the queue doesn't rattle you, that people are glad when you're on shift, that your teammates never have to check your work — was never in the file. Not because you wrote it badly. You'd stopped noticing it years ago, and nobody had asked.

Which is where the whole problem lives. The part of the résumé that matters most is the part no document can check, so it's the place where being right helps you least: you can only claim, claims are cheap, and everybody has read the same advice about how to phrase them. How you present it moves the odds a little, and one good trial says so. What moves them most is something that verifies you from outside — a person who watched you work, somebody with something to lose putting their name next to yours, a conversation where the thing you can't say about yourself finally gets said. Which is where the hiring side went, on that platform, the moment the writing stopped telling them anything.

Which is what What Color Is Your Parachute? was saying on that shelf, in a book there that wasn't a résumé book at all. It took me the length of this article to arrive back at it.

A model cannot do any of that. It has not seen you work and it cannot vouch for you. Ask a machine to tell you who you are, and it will hand back — very neatly formatted — exactly what you told it.

So here is what I'd actually do.

Get through the first gate, because it's the one that's yours. Make the record readable — plain, parseable, the title where a title goes. It is not the interesting half, and clearing it is the difference between being passed over and never having been in the room at all.

And use the machine for that. Not because it will hand you the job — nothing guarantees you that, and anybody promising otherwise is selling something. Because the standard is written down, you weren't given it, and a model will hand it to you in an afternoon. Any percentage you can get is the whole of what's on offer here. That was true before any of this was automated. It is what I meant, nine thousand words ago, by a game of percentages.

Then work the second gate knowing you don't own it. Go find a person. Not the same person twice — one to help you see what you have, because you stopped noticing it years ago, and one who'll say your name in a room you're not in.

And if you clear the first gate and still hear nothing, go anyway. Walk it in. Ask somebody who already works there. Not because it usually works — it often won't — but because of what it does to the silence.

Because there are only two ways silence ends, and neither tells you anything on its own. Never found: nothing happened, and there's nothing in it to learn. Found and passed over: still silence, and you will almost certainly never be told why. But go and ask, and you find out which one you were in. And if you were found: either it was fit — which was never yours to fix, and is not a verdict on you — or there's a specific skill they need that you don't have.

And that second one is the best news in this article, even though it doesn't sound like it. A missing skill is a thing you can go and get. It's not a mystery and it's not a judgement on who you are. It's an answer, and the whole rest of this process is built to never give you one. Nobody was ever going to put that in an email. You have to go and stand in front of somebody to hear it.