ChatGPT has become a regular stop for anyone writing or touching up a resume. It writes fast, never tires and produces sentences that sound polished. The catch is that it also produces sentences that sound like ChatGPT, and recruiters now spot them on sight: inflated adjectives, identical bullets under every job, a uniform tone that says nothing specific about the person.
This guide separates what an AI assistant is good for from what it is not, offers prompts for the tasks where it helps, lists what to check before sending the result, and explains the formatting limits that no chat window can solve.
In this guide
What an AI assistant is actually good at
A language model excels at one narrow task: rewriting text that already exists. Given a weak bullet and real numbers, it returns a clearer version with an action verb up front and a result at the end. Given a full resume, it finds repetition and empty phrases. Given the resume and a job posting, it points out which terms from the posting are missing.
The tasks where it earns its keep:
- Rewriting experience bullets in the verb + what + measurable result pattern, from a plain description of duties and the actual figures the person supplies.
- Catching redundancy: two bullets that tell the same story, a summary that repeats the job title, skills listed separately that already show up in the experience section.
- Tailoring to a specific posting: comparing the text against the job description and proposing what to reorder or which term to adopt, following the logic in the guide to tailoring a resume to the job.
- Cutting length: turning a five-line paragraph into two bullets without losing anything a hiring manager needs.
In every one of those cases, the raw material comes from the person and the AI only organizes it. That is the condition for the result not to show.
What not to ask for
Two requests ruin resumes more than any others. The first is asking it to write the whole thing from a couple of facts ("write a resume for an accountant with five years of experience"). The model knows nothing about the actual career, so it fills the gaps with whatever is statistically likely: generic responsibilities, invented achievements with round percentages and a summary that could belong to anyone in the field. That text does not survive a ten-minute phone screen.
The second is asking it to "improve" an experience without data. Faced with "helped customers at a store," a model tends to produce "managed the end-to-end customer experience while consistently exceeding sales targets." If there were no targets and no figures, the bullet is fiction, and fiction gets exposed the moment someone asks about it. It is also unwise to have it decide what to leave out: that call requires knowing the market and the role.
Simple rule: AI may change the shape of what it is given, never add substance. Every number, tool and responsibility on the resume has to come from the person.
Prompts that work
These three prompts cover the useful tasks. Paste the actual resume text rather than describing it, and add the concrete details (figures, tools, team size) the model cannot guess.
To rewrite bullets:
Rewrite the following experience bullets using this structure: past-tense action verb + what was done + result with a number. Use only the data below; if a number is missing, leave the bullet without one rather than inventing it. Maximum 20 words per bullet. Data: [paste duties and real figures].
To catch redundancy:
Read the full resume below and list: (1) bullets that repeat the same idea, (2) sentences that carry no concrete information, (3) skills that are already demonstrated in the experience section and are redundant in the skills list. Do not rewrite anything yet, only flag. Resume: [paste text].
To tailor to a posting:
Compare the resume with the job posting. List the terms, tools and requirements in the posting that do not appear in the resume. For each one, say whether any experience in the resume covers it under a different name and suggest how to phrase it using the posting's wording. Do not add experience that is not in the resume. Posting: [paste]. Resume: [paste].
A well-specified prompt saves three rounds of corrections. A vague one ("improve my resume") produces the generic text that then has to be undone.
What to check before sending
The text an assistant returns needs a second read with four concrete questions. It takes five minutes and prevents most of the damage.
- Fabrications. Every number, tool or client named must be real and defensible in an interview. Models add "by 30%" with alarming ease.
- Generic filler. If a bullet could sit on the resume of anyone in the same field, it says nothing. "Collaborated with cross-functional teams to achieve objectives" is the classic example.
- AI tells. Long dashes in every sentence, adjective triplets ("dynamic, proactive and committed"), words like "synergy," "leverage," "spearhead," "robust," and closers such as "ready to add value."
- Uniform tone. When every bullet under every job has the same length and the same kind of verb, the document reads as manufactured. Let the important jobs carry more detail than the minor ones.
Many of these overlap with the most common resume mistakes, with one difference: AI produces them in bulk and in good prose, which makes them harder to see.
Avoid / Better
The difference between a bullet generated without data and one rewritten from real data:
| Avoid | Better |
|---|---|
| Led strategic initiatives to optimize processes and drive results | Redesigned the purchase-approval workflow and cut average turnaround from 12 to 5 days |
| Managed the end-to-end customer experience with a focus on excellence | Served an average of 60 customers a day at the Austin store and held a 4.8 rating on post-visit surveys |
| Proactive, results-driven professional with strong communication skills | Accounting analyst with 6 years in consumer goods; closed 24 monthly books with zero audit findings |
| Spearheaded the digital transformation of the department | Migrated invoicing from spreadsheets to QuickBooks and eliminated 15 hours a week of manual entry |
The right-hand column contains something only the person could know. That is what AI cannot fabricate, and what a hiring manager is looking for.
The formatting limits
ChatGPT and similar assistants return text. They do not design the page, pick a typeface or margins, or export a clean PDF. Asked for "a formatted resume," they produce Markdown tables or asterisk-laden text that, pasted into Word or Google Docs, has to be rebuilt by hand. Some can generate a downloadable file, but it usually arrives with inconsistent fonts, uneven spacing and a structure that applicant tracking systems misread.
The sensible approach is to treat the assistant as a copy editor and handle layout elsewhere: a restrained one-column document on Letter paper, with sections in the order described in the guide on how to write a resume, and a verified PDF export.
How chooseme's AI does this inside the editor
Copying the resume into a chat, pasting the answer back and copying again has a cost: every pass loses context, and the model never sees the document the way it will actually look. chooseme's resume improvement AI works the other way around: inside the editor, on the real resume, one section at a time.
Suggestions for a bullet are made knowing the title, the company, the dates and the other bullets under that job; the summary is proposed from the experience already entered; and the comparison against a posting runs on the real content, highlighting what is missing and where. Nothing is invented, because the AI only rewrites what is already in the document, and each suggestion is accepted or dismissed individually. Layout is handled by the editor's templates, with a one-column PDF export. It is free, no credit card required.
Frequently asked questions
Can recruiters tell when a resume was written with ChatGPT?
When the text was generated without real data, yes: generic bullets, evaluative adjectives, an identical tone under every job and phrases like "drive synergies." When the AI only rewrote concrete information the person supplied, the result is indistinguishable from a well-written resume.
Is it wrong to use ChatGPT for a resume?
No. It is a writing tool, like a spell-checker or a template. The problem is sending content the model invented, or signing off on bullets that cannot be backed up in an interview.
What prompt should be used to improve a resume?
One that includes the actual text, the concrete facts (figures, tools, team size) and an explicit instruction not to invent. For example: "Rewrite these bullets with an action verb and a measurable result, using only the data that follows." Vague prompts produce generic text.
Can ChatGPT run a resume through an ATS?
It has no access to any applicant tracking system and cannot reliably simulate one. It can compare the resume against a posting and list missing terms. Compatible formatting is a separate matter, covered in the guide on building an ATS-friendly resume.