Why ChatGPT fails you when tailoring your CV with AI

It feels like tailoring. Most of the time it is not.

I understand why people do it. You paste your old CV into ChatGPT, Gemini, or Claude, drop in the job description, type something like “tailor this for me”, and hope the machine turns your tired Wednesday night application into something sharp.

It usually does not.

What you usually get is not tailoring. It is polished approximation.

The first draft often looks impressive for about thirty seconds. Then you read it properly and realise it sounds vaguely fake, slightly overconfident, and weirdly detached from the career you have actually had.

That is not proper CV tailoring. It is word shuffling with good manners.

The model does not know you well enough

This is the real failure point, and most people underestimate it.

A generic chatbot like ChatGPT, Gemini, or Claude usually sees one CV and one job description. That is it. Maybe you also paste a few notes if you are feeling diligent. But it still has a paper-thin understanding of your career.

It does not know which project mattered most. It does not know which bullet on your old CV was underselling a hard piece of work. It does not know which job title was misleading, which promotion never made it onto paper, or which part of your background is more relevant than it first appears.

So it does what these models always do when context is thin. It guesses. OpenAI’s own explanation of hallucinations is basically an admission that language models are still pushed towards confident guessing when they are uncertain. Even LinkedIn’s AI-powered resume tips tell users to verify AI-generated suggestions for authenticity because they may contain inaccuracies.

Sometimes that guess is harmless. Sometimes it quietly stretches a claim, invents confidence you did not express, or turns a supporting responsibility into a core achievement. That is where the hallucination problem really shows up on CVs. Not always as a made up technology, but as a subtle lie about scope, ownership, or seniority.

It can twist your words around. It cannot reliably represent a career it barely knows.

The writing gives it away instantly

You can usually spot chatbot CV writing from across the room.

Too much smoothing. Too much symmetry. Too many tidy little leadership verbs. Too much polished certainty in places where a real candidate would be more specific.

Every model has its habits. Certain words. Certain bullet shapes. Certain rhythms. You see the usual rubbish about spearheading, driving, leveraging, optimising, and delivering strategic impact, even when the original experience was much more grounded and believable.

This is landing in a hiring market where Greenhouse says recruiters are drowning in application volume and trying to work out what is real. And it is not just annoyed job seekers saying this. Resume Genius reports that 80% of hiring managers say they can often tell when a resume is AI-written, with unnatural phrasing and vague inflated descriptions showing up as the biggest giveaways. Whether the draft came from ChatGPT, Gemini, Claude, or something else, the same patterns tend to show up.

Once your CV starts sounding like everyone else who asked a chatbot to rewrite it, you lose the one thing that matters most. A credible signal that a real person did real work.

Chatbots are generalists, not serious CV editors

This is the second trap. People assume that because a chatbot can do many things, it must also be good at this one.

But writing a strong CV is its own craft. A good CV editor knows when to sharpen a bullet and when to kill it. They know that one concrete metric beats three fluffy claims. They know how to prioritise evidence, how to match the language of the role without sounding desperate, and how to avoid stuffing in every tool you have touched since 2017.

A general chatbot does not naturally care about that. ChatGPT, Gemini, and Claude are trying to be helpful in a broad, all-purpose way. The result is usually generic uplift rather than serious editing. Even Indeed’s advice on putting a human touch on AI-assisted applications comes back to the same boring truth. Specific examples, measurable results, and language that genuinely reflects your experience matter more than polished filler.

That is why the output often gets more verbose instead of more convincing. It adds adjectives where it should add proof. It broadens claims where it should narrow them. It turns sharp experience into smooth paste.

It gives you too much credit

This one is more dangerous than it sounds.

Most models are deeply biased towards encouragement. They want to be useful, positive, and reassuring. Nice qualities in many contexts. Not so nice when you are trying to decide whether you are actually a fit for a role.

So the model starts flattering you. It frames every adjacent skill as if it were direct experience. It treats partial exposure like ownership. It tells you that your background maps beautifully to a role when the honest answer is more complicated.

That might feel good for five minutes. It is terrible for decision making.

A decent job search system should sometimes tell you no. It should tell you that the fit is weak, that a requirement is a real gap, or that you are about to spend an hour tailoring for a role that is not likely to land. Generic chatbots are not very good at that kind of honesty.

Then you inherit a formatting job

Even when the wording is half decent, you still inherit the practical mess.

Now you have a block of text in a chat window or a Markdown export. Fine. What next.

You still need to compare it against your old CV and work out what changed. You still need to move the useful bits into your real document. You still need to fix bullet formatting, bold text, spacing, hierarchy, dates, indentation, and all the annoying little layout issues that make a CV feel coherent.

If you are using Word or a visual template, this gets tedious fast. You end up pasting line by line, second guessing which edits were actually improvements, and wasting half your time on document admin.

So the tool that was meant to save effort often just moves the effort somewhere less obvious.

Before you send anything, I think it is worth doing one blunt final pass:

• Can I defend every claim in an interview without wincing?
• Did this add proof, or just add polish?
• Does this sound like my real voice, or like a chatbot doing management theatre?
• Did it copy the job description’s phrasing without actually improving relevance?

What I think works better

I do still think AI can help with CV tailoring. I just do not think a generic chatbot window is the right shape for the job.

What works better, at least for me, is a workflow built around a proper master profile. Real achievements. Real project detail. Real preferences about tone and writing style. Real criteria for the kinds of roles you actually want.

First, pick evidence. Do not generate it.

The first job is not “write me a CV”. It is “pick the strongest evidence”. Which bullets actually match this role. Which projects matter. Which metrics prove the point. Which requirements are real strengths and which ones are still gaps.

That part should be constrained. The system should be selecting from verified source material, not inventing from scratch. If you let the model freestyle too early, you are back in the same mess as the chat window.

Second, let code do the boring deterministic work.

This is where people underestimate the engineering problem. You can try to force the same process through ChatGPT, Gemini, Claude, or an agentic wrapper around them, but it is awkward. You burn tokens getting the model to re-read context, re-decide structure, and re-attempt tasks that should have been handled by the workflow itself.

That is expensive, error-prone, and usually worse than it looks. Python or application logic can handle structure, filtering, ordering, comparison, and guardrails far more reliably.

Third, use AI last for light polish.

Tighten phrasing. Align keywords. Clean repetition. But do not rewrite the substance of your career, and do not turn a thin fit into a fake fit.

That is the strategy I trust, and it is also the design logic behind OutRung. The interesting bit is not “AI writes your CV”. It is that the AI should mostly select and lightly refine, while the heavier lifting stays anchored to a structured profile and a stricter workflow. That is more efficient, less error-prone, and much better suited to this problem than hoping a chat window will improvise the whole system on demand.

For anyone doing this manually, I would still follow the same logic.

• Build a proper master record of your experience before you tailor anything.
• Score the role honestly before you ask AI to rewrite a single line.
• Pull forward the evidence that genuinely matches before you touch the wording.
• Use AI last for light polish, not first for invention.
• Treat any AI output as a draft to verify, not a truth to trust.
• Keep your final CV in a format you can control without fighting the layout every time.

That is less magical than chatbot demos. It is also much closer to something that actually gets interviews.

The honest version

If you ask ChatGPT, Gemini, Claude, or another generic chatbot to tailor your CV, it will usually give you something that looks cleverer than it is.

It does not know you well enough. It writes in a voice that gives itself away. It flatters too much. It misses what makes a CV credible. And even when it produces a usable paragraph, you still inherit the awkward work of turning that paragraph into a real application.

That is not a tailoring system. It is a convincing shortcut to more admin.

If you are serious about your next move, you need more context, more structure, and a lot less blind trust in a cheerful text box.