6 ChatGPT Resume Mistakes When Tailoring, and the Fix for Each

ChatGPT is a capable resume editor and an unreliable witness. It rewrites fluently, follows structure well and knows the vocabulary of most jobs. It also does not know your history beyond what you paste, and when the job posting asks for something your resume does not show, it tends to produce the most plausible-sounding answer rather than a blank. Most tailoring mistakes come from that one gap. This post names the six that matter, shows how to spot each in the output, and gives a fix you can paste.

#Why these mistakes happen

Three documented properties of language models explain nearly all of them.

They guess when uncertain. The 2025 paper Why Language Models Hallucinate by Kalai and colleagues describes models that "sometimes guess when uncertain, producing plausible yet incorrect statements instead of admitting uncertainty." A posting that rewards impact, next to a bullet with no number, is exactly that situation.

Output varies from run to run. OpenAI's prompt engineering guide notes that "the content generated from a model is non-deterministic". Running the same prompt twice can give you two different resumes, with different errors.

They do better with explicit rules. The same guide says the models "benefit from more explicit instructions around how to accomplish tasks", and Anthropic's guidance on reducing hallucinations recommends giving the model explicit permission to admit uncertainty and telling it to "only use information from provided documents". A prompt like "tailor my resume to this job" gives neither.

#The six mistakes at a glance

#Mistake 1: invented metrics

Symptom. "Increased customer satisfaction by 25%" appears where your original said "handled customer escalations". A recruiter will ask how you measured it.

Fix. Paste this rule into your prompt: "Do not add or change any number. Where a bullet would be stronger with a figure I did not give, write [add: what to measure] instead." Then fill each marker from your own records or delete the clause.

#Mistake 2: changed job titles

Symptom. "Marketing Coordinator" has become "Marketing Manager" because the posting is for a manager. Sometimes the model "clarifies" a title in a way that changes its level.

Fix. "Do not change any job title, employer name or date." If your real title is unusual and you want to signal the role family, add a headline above your experience instead; changing a job title for ATS explains what is acceptable.

#Mistake 3: the generic summary

Symptom. "Results-driven professional with a proven track record of cross-functional collaboration and a passion for innovation." Nothing in it could only be true of you.

Fix. "Write a two-sentence summary. Each sentence must contain one specific fact from my resume: a scope, a result or a named tool. Do not use the words results-driven, passionate, proven track record or dynamic." Resume summary examples has patterns that work.

#Mistake 4: the keyword dump

Symptom. The skills section now lists every tool in the posting, or a bullet reads like a string of nouns: "Leveraged Python, AWS, Kubernetes and AI/LLMs to drive cross-functional stakeholder alignment."

This is the mistake with the clearest cost, because posting vocabulary varies more than people expect. In the ATS Resume AI analysis of all 503 S&P 500 careers sites and 9,260 live job postings, June 2026 (role data), even the most common software engineer keyword, Python, appeared in only 26.6% of full-text postings, and Kubernetes in 13.4%. For product managers, by contrast, "roadmap" appeared in 90.3%. A model that pads your resume with terms "typical" for the role is adding words most specific postings never asked for, and that you may not be able to discuss.

Fix. "Use a term from the posting only where my resume shows I did that work. For posting terms with no evidence, write [confirm: term] and list them at the end." Pull the terms that a specific posting actually uses with the free resume keyword generator.

#Mistake 5: dropped or merged dates

Symptom. In a long resume or a long conversation, an early job disappears, two roles at one company become one, or a date range shifts. This tends to happen when you iterate several times in the same chat, as earlier details fall out of focus.

Fix. Start a fresh chat for each posting, paste the full original every time, and end the prompt with: "Finally, list every role in my original resume with its title and dates, and confirm each appears unchanged in your version."

#Mistake 6: ATS-hostile formatting

Symptom. You paste the result into your document and it carries asterisks for bold, pound signs for headings, a markdown table for skills, or emoji bullets. Some of these survive into the file and confuse a parser or a reader.

Fix. "Return plain text only: no markdown, no tables, no emoji. Use the headings Summary, Experience, Skills, Education." Then paste into your own single-column template rather than formatting inside the chat.

#The combined guardrail prompt

Put all six fixes in one prompt so you do not have to catch them afterwards.

Tailor my resume to the job posting below.
 
Rules:
1. Do not change any job title, employer name or date.
2. Do not add or change any number. Where a figure would help, write
   [add: what to measure].
3. Use a term from the posting only where my resume shows I did that
   work. Otherwise write [confirm: term].
4. The summary is two sentences; each contains one specific fact from my
   resume. No clichés (results-driven, passionate, proven track record).
5. Keep every role in my original resume.
6. Return plain text, no markdown, with the headings Summary,
   Experience, Skills, Education.
 
After the resume:
- list every [add] and [confirm] marker;
- list every role with its title and dates and confirm each is unchanged;
- list any posting requirement my resume has no evidence for.
 
If you are unsure whether my resume supports a claim, leave it out and
say so.
 
POSTING:
[paste]
 
MY RESUME:
[paste]

#Check the output in five steps

1. Put your original and the output side by side.
2. Search the output for every digit and confirm each number is in your original.
3. Compare every job title, employer and date range.
4. Read each skill and ask whether you could answer an interview question about it.
5. Resolve or delete every bracket, then paste the plain text into your template.

For a wider set of tailoring prompts built on the same rules, see ChatGPT prompts to tailor a resume.

#Sources

• Kalai, Nachum, Vempala and Zhang, Why Language Models Hallucinate, arXiv, 2025
• OpenAI, Prompt engineering guide
• Anthropic, Reduce hallucinations