From the writer's desk · Resume · AI
The Case Against ChatGPT — at Least for Resumes
From inside a resume writing business: what we've actually been seeing in the documents that come across our desks since ChatGPT entered the chat. The shift is unmistakable, and it's not flattering.
In thirty seconds
- We've watched the documents people send us shift dramatically over the past 18 months. Pre-2024 documents were rough but real. Most documents we see now are polished and impersonal.
- Three patterns keep showing up: identical opening lines, fabricated metrics, and what we call "candidate-shaped erasure" — where the candidate's actual personality has been smoothed out of existence.
- The polish that ChatGPT adds is real. The cost is that the document no longer represents a specific person — it represents the median of what ChatGPT thinks a job application looks like. That's not a competitive position.
- The fix isn't to never use AI. It's to use it for the parts AI is actually good at (research, brainstorming, proofreading) and not for the parts it makes worse (the actual writing).
This is going to sound like a complaint from someone whose business is threatened by ChatGPT. So I'll be upfront: it's not, and we're not. Application volumes are at record highs, the candidates we see are working harder than ever to stand out, and the demand for hand-written documents has gone up — not down — since AI tools became mainstream. We're not afraid of ChatGPT.
What we are is watching it. Closely. Because we read job applications all day, every day, and we've had a front-row seat to what happens when the tool that was supposed to help every candidate instead makes nine out of ten of them indistinguishable.
Here's what we're actually seeing.
The shift, in 18 months
Before late 2023, the resumes and cover letters that arrived in our inbox tended to be a mixed bag. Some were tidy, structured, and quietly impressive. Most were rough — full of typos, inconsistent formatting, the same word used five times in two paragraphs, achievements buried inside duty lists, and the occasional Comic Sans crisis. They were, in a word, human. A real person sat down and tried, and the result reflected how that effort went on that day.
The documents arriving now are different. They're cleaner. The grammar is flawless. The structure is consistent. The metrics are present and accounted for. There are no typos, no redundancies, no formatting accidents. They look, frankly, more polished than the documents we used to see.
And they all sound the same.
The new documents arrive grammatically perfect, structurally tidy, and almost entirely interchangeable. The polish is real. The personality has gone missing.
This isn't a vague impression. It's three specific patterns, repeated across hundreds of documents, all roughly traceable to the moment a candidate decided to "improve" their materials with ChatGPT.
Pattern 1: The interchangeable opener
Pattern 01
Cover letters that all begin the same way
Of the cover letters we've reviewed in the past year, an extraordinary number open with a variant of the same six or seven sentence patterns. Some are direct ChatGPT defaults. Others are what happens when ChatGPT has been asked to "make this better" and the model has rewritten the candidate's genuine first line into AI-house style.
The problem isn't that this opening is wrong — it's that everyone is using it. When a hiring manager opens 30 cover letters in a morning, half of them starting with this construction, the entire batch becomes background noise. The candidate's letter doesn't stand out as polished; it stands out as obviously AI-assisted, which now triggers a different, less generous reading.
Pattern 2: Metrics that don't fit the role
Pattern 02
Confident, suspiciously round percentages on every bullet
Generative AI has been trained that good resumes have quantified achievements, so it inserts them — frequently. The result is bullets like "increased efficiency by 40%" or "boosted productivity by 25%" or "drove $200K in savings" appearing on resumes for roles where the candidate would never realistically have access to those metrics.
This is the moment a credible application becomes an uncredible one. A junior admin officer doesn't usually own client retention metrics. A graduate marketing coordinator rarely "increases conversion by 47%." When the size of the metric doesn't match the seniority of the role, recruiters quietly downgrade the application. We've written about this in detail in the metric mirage, but the short version is: AI-inflated metrics sink applications faster than no metrics at all.
Pattern 3: Candidate-shaped erasure
Pattern 03
The personality has been smoothed out
This is the subtlest pattern, and the one that costs candidates the most interviews. When you put a draft into ChatGPT and ask it to "make this more professional" or "improve the tone," the tool doesn't sharpen your voice — it averages it. Idiosyncrasies disappear. Specific phrasing gets replaced with generic alternatives. The thing that made the writing recognisably yours is exactly what the model treats as friction to be polished out.
Read those two paragraphs. The first one is interesting — there's a person there, a story, a sense of how it actually went. The second is a wall of formal nothing. It's grammatically superior in every measurable way and dramatically less compelling. That's what's been happening to applications across the board: real voices being replaced by their AI-averaged equivalents, and the candidates losing exactly the thing that would have made them memorable.
The cruel irony is that the candidates who use ChatGPT most heavily are often the ones with the most interesting stories. The tool is busy smoothing out the very thing that would have got them the interview.
A pragmatic position on AI
We're not telling you to never touch ChatGPT during a job search. Most of you will, and a flat "don't use AI" rule isn't useful. The more useful framing is: AI is good at some parts of this work and actively bad at others, and the difference matters.
- Researching the company before you apply
- Brainstorming achievements you might have forgotten
- Extracting keywords from a job ad you can use in your resume
- Proofreading your finished document for typos
- Practising answers to common interview questions
- Reformatting clean ATS-friendly structure
- Writing your cover letter from scratch
- Rewriting your resume bullets to "improve" them
- Generating metrics or percentages you don't actually have
- Smoothing out your voice — that's the part that wins
- "Make this more professional" — it strips your personality
- Anything you can't defend in interview
The general rule we'd suggest: use AI as a research and review tool, not as a writer. The substance and voice of your application has to come from you, or it won't work — because the people reading the documents have learned to spot the difference, and they've stopped finding the polished version impressive.
What's actually working in 2026
What we're seeing land interviews, consistently, are documents that read like a person describing real work. Specific situations. Real numbers, where they exist. Sentences that sound the way the candidate would actually talk. Cover letters that name something specific about the company, not "your esteemed organisation." Resumes that feel like the product of an hour with a real interviewer, not 30 seconds with a chatbot.
That's not nostalgia. That's just where the market has moved. When polished is cheap, polished stops being a signal. The signal now is specificity, evidence, and voice — exactly the things AI is worst at producing.
Use the tool where it helps. But don't let it speak for you, because at the moment, it's speaking for almost everyone else.
When the polish is everywhere, voice is the differentiator
We interview you. Then we write the document by hand
An hour with a senior writer who asks the questions that interrogate the real story — not "tell me about your achievements" but the specifics, the context, the way you'd actually describe it to a friend. We write it up in language that's still recognisably yours, calibrated to the level and sector you're applying into. The opposite of an AI-averaged document, which is exactly what the market is rewarding right now. No AI. No offshore. No templates. 4.8 on Google. Trading since 2016.
