The classic resume mistakes have not gone anywhere, and they are covered separately: duties instead of outcomes, one resume for everything, an objective statement, filling the page. This article is about a different set. These are mistakes that either did not exist five years ago or have become expensive recently, because the layer between you and a hiring manager changed and a lot of advice did not.
1. Letting a model write it, rather than edit it
A language model produces a fluent, competent, entirely generic resume in seconds, and a great many people are now doing exactly that. The result is a screening pile in which a large share of documents share a register: the same verbs, the same cadence, the same faint air of a press release.
Fluency has therefore stopped being a signal. It used to be reasonable for a recruiter to infer care from a well-written resume. Reading forty polished, interchangeable ones, what they are now looking for is whatever is specific, and specificity is the one thing a model cannot invent for you because it does not know what you did.
Fix: use a model the way you would use an editor. Give it your real material and ask it to tighten, cut, or fix a clumsy sentence. Do not ask it to supply the substance, and read every line it returns for claims you cannot defend.
2. Optimising for the score instead of the reader
Match scores are now everywhere, and they create a strong pull toward gaming them: adding terms you have a passing acquaintance with, padding a skills list, rewording bullets to echo the posting rather than to describe the work.
This fails in two directions at once. A well-built model caps how much keyword matching can contribute precisely because it is gameable, so the ceiling on that strategy is lower than it looks. And every term you add that you cannot speak to for two minutes is a landmine in the interview you were trying to reach.
Fix: treat the score as a diagnostic rather than a target. When a component is low, ask what it is measuring and whether the underlying thing is actually true of you. If it is true and unstated, state it. If it is not true, the low score is correct information.
3. A modern template that cannot be parsed
Template marketplaces are full of beautiful two-column designs with sidebars, skill rating bars, icons and timeline graphics. They look current, which is exactly the problem: they are optimised for the preview thumbnail, not for the pipeline the file has to survive.
A parser reading across the full page width interleaves two columns into alternating fragments. A skills grid built as a table flattens into an unpunctuated run of words. Contact details in the document's header region are frequently discarded outright, which is how people become uncontactable. Rating bars and badges contain no text at all, so a certification shown as a badge is, to the system reading it, a certification you do not hold.
Fix: keep two versions of the same content. The plain single-column file goes to every upload form; the designed one goes to a recruiter who emails you and into the interview loop. Test the plain one by opening the PDF, selecting all, copying and pasting into a text editor.
4. A skills section that is an inventory
Sixty terms in a block at the bottom, covering everything you have touched since university. This was always weak, and matching that works on meaning rather than literal strings makes it weaker, because a list of nouns describes no actual work and there is nothing there to match against.
Fix: keep the list to what you would be comfortable being questioned on, and name the important items in the role where you actually used them. A tool that appears in a bullet with an outcome attached is worth several that appear only in a list.
5. A job title nobody would search for
Internal titles are a private language. "Growth Ninja", "Customer Happiness Hero", "Member of Technical Staff III" all describe real and often senior work, and none of them matches what a recruiter types into a search box. Semantic matching helps here but does not rescue it, because the recruiter's own search is still literal.
Fix: put the industry-standard equivalent alongside the real one. "Growth Ninja (Marketing Manager)" keeps the accuracy and gains the term. Do not replace it with a title you never held.
6. Listing AI tools as though they were skills
A line reading "ChatGPT, Copilot, Midjourney" under Skills is now extremely common and says almost nothing, because it describes access rather than ability. Everyone applying has access.
Fix: describe what changed. "Built an internal retrieval tool over our support archive that cut average first-response time from 6 hours to 40 minutes" is a claim about you. "Proficient in AI tools" is a claim about the decade.
7. Assuming the human step disappeared
The most consequential mistake on this list, because it is a premise rather than a detail. It is possible to read enough about parsers, scores and automated screening to conclude that the whole process is machinery, and to write a document aimed entirely at the machinery.
Every layer discussed here exists to decide who reaches a conversation with a person. That conversation has not changed at all. A resume optimised for the filters at the cost of being interesting to a human has won the qualifier and lost the match, and it is a failure mode that is easy to fall into precisely because the filters are the part that gets written about.
Fix: after every optimisation pass, read the document as a person. Does it say what you do? Is there anything in it a stranger would find interesting enough to ask about? If not, it will pass the screen and stall at the shortlist.
8. A profile that contradicts the resume
Cross-checking a candidate against their public profile takes about fifteen seconds and is now a routine part of screening rather than an occasional extra step. Which means every discrepancy between the two documents is likely to be seen, and discrepancies are extremely common because nobody updates both at once.
The usual offenders are dates that do not agree, a title that is more senior on one than the other, a role present in one and absent from the other, and a current position that ended eight months ago. None of these is dishonest in intent. All of them read as carelessness at best, and the more generous a reader is feeling, the more likely they are to simply move on to the next candidate rather than ask.
Fix: when you update the resume, update the profile in the same sitting. Dates and titles must match exactly. The profile can be longer and more informal; it cannot disagree.
9. Tailoring without a master document
Tailoring is correct advice, and followed carelessly it produces its own failure. Six per-company versions, each edited directly from the last, drift apart. A role gets trimmed from one and never restored. A number gets rounded up in one version and not another. Eventually you send a document with an eighteen-month gap you created by deleting a job to save space.
Fix: keep one complete master file containing everything, and treat each application as a cut-down copy of it rather than an edit of the previous application. Cutting is faster than writing, the master never loses a detail, and every version stays consistent with every other because they all descend from the same source.
What they all have in common
Every one is a substitution of a proxy for the real thing: fluency for care, score for fit, design for legibility, tool access for capability. The screening layer changed, the proxies changed with it, and the underlying requirement did not. Describe real work, specifically, in a file that can be read.
Check which of these applies to yours
The component breakdown shows which bullets still read as duties, which terms from the posting are missing, and what a parser does with your formatting.
Check your resume free