AI Prompts for Resumes and Cover Letters That Don't Sound Like AI

Why most AI resume output is bad

Ask a model to "write a resume for a marketing manager" and you get something that reads well and describes nobody. It is fluent, generic, and full of the exact phrases every other applicant using the same approach has also submitted. Recruiters see the pattern quickly.

The failure is structural, not a matter of picking a better model. A resume is a claim about specific things you did, with specific results, at specific places. The model does not know any of that, so when asked to produce a resume it produces the average of every resume it has seen. Averages do not get interviews.

The useful move is inverting the task. Instead of asking AI to generate content it cannot know, ask it to work on content you supply: sharpen it, restructure it, find what is missing, and tell you what a recruiter will ask about. That AI is genuinely good at, and it is the difference between a tool that helps and a tool that produces a liability.

The rule that keeps you out of trouble

Never let AI invent a fact about your history. Not a metric, not a job title, not a technology you have not used.

This is not only an integrity point, though it is that. It is a practical one: everything on a resume is interview material. A bullet claiming you "increased conversion by 34%" will be asked about, and if the number came from a language model rather than from your work, the conversation ends badly. The same applies to a tool listed because it sounded impressive.

A useful habit: any time a model produces a number you did not give it, delete it and replace it with either your real figure or a qualitative statement you can defend. "Reduced onboarding time from two weeks to three days" is strong. "Significantly improved onboarding efficiency" is weaker but true. An invented 34% is neither.

Prompt 1 — Turn duties into outcomes

The most common resume weakness is bullets that describe responsibilities rather than results. "Responsible for managing the social media accounts" tells a reader what your job description said, not what changed because you were there. This prompt does the conversion without inventing anything.

Copy this and paste your existing bullets underneath it:

• Prompt: "You are an experienced technical recruiter. Below are bullet points from my resume. For each one, tell me: (1) whether it describes a duty or an outcome, (2) what specific information is missing that would turn it into an outcome, and (3) a rewritten version using ONLY information I have given you — leave a clearly marked [BLANK] where you would need a number or detail from me. Do not invent metrics, tools, or scope. Do not add adjectives like 'successfully' or 'effectively'. My bullets: [PASTE HERE]"

Prompt 2 — Extract what a job description is really asking for

Job descriptions are written by committee and mix genuine requirements with wishlist items and boilerplate. Tailoring your resume to all of it wastes the limited space you have. This prompt separates the three.

• Prompt: "Below is a job description. Separate its contents into three lists: (1) genuine requirements — things the person almost certainly cannot do the job without, (2) preferences — things that would help but are clearly secondary, (3) boilerplate — generic phrasing that appears in most job ads and carries no information. For each item in list 1, quote the exact wording from the ad. Then tell me the three things a strong applicant would most need to evidence. Job description: [PASTE HERE]"

Prompt 3 — Find the gap between your resume and the job

Once you have both, the highest-value analysis is the gap. This prompt is deliberately harsh, because a friendly assessment is useless here — the recruiter will not be friendly.

• Prompt: "You are screening applications for the role below and you have 90 seconds per resume. Here is the job description and here is my resume. Tell me: (1) which of the role's genuine requirements my resume evidences clearly, (2) which it evidences weakly or only by implication, (3) which it does not evidence at all, and (4) the single strongest reason you would reject this application. Be blunt — I want the real assessment, not encouragement. Job description: [PASTE]. Resume: [PASTE]"

Prompt 4 — Rewrite for a specific role without lying

Tailoring is legitimate and expected: the same work can be described with different emphasis depending on what the role values. This prompt keeps the emphasis flexible and the facts fixed.

• Prompt: "Rewrite my resume bullets to emphasise what matters for the role below. Rules: you may reorder bullets, change emphasis, and reword — you may NOT add any fact, metric, tool, or responsibility that is not already present in my original. If a requirement of the role is genuinely not evidenced in my history, say so at the end rather than manufacturing it. Keep each bullet under 25 words. Role: [PASTE]. My bullets: [PASTE]"

Prompt 5 — The cover letter that sounds like a person

Most AI cover letters fail in the first sentence, because "I am writing to express my keen interest in the position" is the tell that a template was used. A cover letter's only job is to say something a resume cannot: why this role, why you, and what you noticed about them.

This prompt is built around supplying that specific content rather than asking the model to fabricate enthusiasm.

• Prompt: "Write a cover letter of 200 words or fewer for the role below. Use only these inputs: my three most relevant experiences [LIST], the specific thing about this company or role that genuinely interests me [WRITE ONE SENTENCE — be honest, 'the problem they are working on' is fine], and my one concern about my own fit [WRITE ONE SENTENCE]. Rules: no opening line about 'expressing interest'. No adjectives about myself — show, do not claim. Address the concern in one sentence rather than hiding it. End with a specific, low-pressure ask. Tone: plain and direct, like a capable person writing to another capable person. Role: [PASTE]"

Prompt 6 — Make it sound like you, not like a model

If you have already drafted something with AI, this pass is what removes the machine texture. It works on any AI-drafted text, not only applications.

• Prompt: "Below is a cover letter draft. Rewrite it to remove AI-writing tells while keeping every fact identical. Specifically: delete any sentence that could appear in any application for any role; remove tricolons (lists of three) unless the three items are genuinely distinct; cut adverbs and self-praising adjectives; vary the sentence length — right now they are probably all similar; replace abstract claims with the concrete detail that supports them, or delete the claim if there is no detail. Return the rewrite, then list the three specific changes that improved it most. Draft: [PASTE]"

Prompt 7 — Prepare for the questions your resume invites

Every bullet is a promise to talk about something. This prompt finds where you are exposed before an interviewer does.

• Prompt: "Read my resume as an interviewer preparing questions. Identify: (1) the three claims most likely to be probed, and the follow-up question each invites, (2) any claim that would be hard to substantiate if challenged, (3) gaps in the timeline or role progression that will be asked about, and (4) the one question I am least likely to have prepared for. For each, give the question as an interviewer would actually phrase it. Resume: [PASTE]"

Prompt 8 — Explain a gap or a career change

Career gaps and changes are common and mostly unremarkable; what damages an application is a defensive or vague explanation. This prompt produces something short and matter-of-fact.

• Prompt: "I need two sentences explaining [GAP OR CAREER CHANGE] for a job application. Here is what actually happened: [WRITE THE TRUTH PLAINLY]. Write it in a way that is honest, unapologetic and brief. Do not spin it as a positive if it was not one — matter-of-fact is more credible than triumphant. Do not use the words 'journey', 'passion' or 'pivot'. Then give me one sentence I could use if an interviewer asks a follow-up."

Prompt 9 — A LinkedIn summary that is not a resume

A LinkedIn summary written as a condensed resume wastes the format. It is read by people deciding whether to start a conversation, which is a different job from deciding whether to interview you.

• Prompt: "Write a LinkedIn About section of 120 words or fewer. Inputs: what I actually do [ONE SENTENCE], the kind of problem I am best at [ONE SENTENCE], and what I want to hear about [ONE SENTENCE]. Rules: write in first person. No third-person narration. Do not list job titles — the profile already shows those. No words from this list: passionate, driven, results-oriented, seasoned, proven track record. Open with what I do, not with a claim about who I am. End with the specific kind of message I would want to receive."

Prompt 10 — The follow-up that is worth sending

Most post-interview thank-you notes say nothing and are read in two seconds. The only version worth sending adds something the interview did not have room for.

• Prompt: "Write a follow-up email of 120 words or fewer after an interview. Inputs: the specific thing we discussed that I want to build on [ONE OR TWO SENTENCES], and one thing I wish I had said better [ONE SENTENCE]. Rules: do not thank them for their time in the opening line. Lead with the substantive point. Address the thing I answered poorly in one sentence, without over-explaining or apologising. No restating of my qualifications. End without pressure."

Prompt 11 — Prepare for the compensation conversation

AI cannot tell you what a role pays in your market. It can rehearse you, which is the part most people skip and then improvise badly under pressure.

• Prompt: "Act as a hiring manager negotiating compensation. I will state a number and reasoning; you push back the way a real manager would — budget constraints, internal bands, comparisons to other candidates. Do not be adversarial for effect and do not fold immediately. After three exchanges, stop and tell me: which of my responses was weakest, what I said that undercut my own position, and the one sentence I should have used instead. My situation: [ROLE, YOUR NUMBER, YOUR REASONING]."

Prompt 12 — Ask for a referral without the awkwardness

Referral requests fail by being either too vague to act on or too demanding to say no to. A good one makes helping easy and refusing painless.

• Prompt: "Write a short message asking for a referral. Inputs: how I know this person and how well [ONE SENTENCE], the specific role [LINK OR TITLE], and the one thing about my background that makes the referral defensible for them [ONE SENTENCE]. Rules: under 100 words. Make it easy to say no — give them an explicit out. Do not ask them to vouch for anything they cannot personally observe. Include the one line they could forward without editing. No flattery about their career."

What AI genuinely cannot help with

Being clear about the limits keeps the useful parts useful.

• It cannot know your achievements. Everything specific has to come from you; the model can only shape what you supply.
• It cannot judge whether you should apply. It has no view on the team, the manager, or whether the role is a step forward for you.
• It cannot make a weak application strong. If the requirements genuinely are not evidenced in your history, better phrasing does not close that gap — a different role, or a project that builds the missing evidence, does.
• It cannot tell you what a specific company values. Job ads are written for legal and HR reasons as much as descriptive ones. People who work there can tell you; a model is guessing.
• It does not know current applicant-tracking behaviour. Advice about keyword stuffing for ATS circulates widely and is largely folklore; write for the human who reads it after the filter.

The tells that give an AI-written application away

Recruiters are not running detection software on most applications. They are pattern-matching against the hundreds of documents they have read, and AI-written text has a texture that becomes obvious once you know it. Removing these is worth more than any prompt on this page.

• The interchangeable opening. "I am writing to express my keen interest in the [role] position at [company]" appears in an enormous share of applications and says nothing. Open with the substantive point instead.
• Praise adjectives with no evidence. "Highly motivated", "results-driven", "detail-oriented" are claims anyone can make and nobody can check. Replace each with the specific thing that demonstrates it, or delete it.
• Lists of three, everywhere. Models default to tricolons — "strategy, execution, and delivery" — and once you notice it in a document you notice it in every paragraph. Vary the count.
• Uniform sentence length. Human writing has short sentences next to long ones. AI prose settles into a medium-length rhythm that reads smoothly and feels synthetic in bulk.
• Restating the job description back. A letter that lists the role's requirements and asserts you meet each one reads as a form filled in, not a person writing.
• Concluding paragraphs that thank in advance for consideration. Harmless individually, but combined with the rest it completes the template.
• No specifics about the company. The single strongest signal that a letter was mass-produced is that nothing in it could not have been sent to anyone else.

A short workflow that ties it together

These prompts work best in sequence, because each one's output is the next one's input — which is exactly the shape of a prompt chain.

Run the job description through prompt 2 to get the real requirements. Run your resume and those requirements through prompt 3 to find the gap. Use prompt 1 on the bullets that came back weak. Use prompt 4 to re-emphasise, then prompt 5 for the letter and prompt 6 to de-machine it. Finish with prompt 7 so you know what you have invited.

Done manually that is six copy-pastes between conversations, and it is where most people give up around step three. Chaining the steps so each output feeds the next removes the copy-pasting and makes the whole sequence repeatable for the next application — which matters, because you will be doing this more than once.

What to do with the output before you send it

Every prompt on this page produces a draft, and a draft is not an application. Three passes turn one into the other, and they take about ten minutes in total.

First, the fact check. Read every line and ask whether you could defend it in an interview. Any number you did not supply, any tool you have not used, any scope you did not have — delete it. This is not a formality: invented specifics are the single way AI assistance turns from an advantage into a liability, and they surface at exactly the worst moment.

Second, the voice check. Read it aloud. The sentences you stumble over are the ones that are not yours, and they are usually the ones a model added to sound polished. Replace them with how you would actually say it, even if that is plainer. Plainer reads as more credible.

Third, the specificity check. Find every sentence that could appear in someone else's application for a different role and either make it specific or cut it. A shorter letter with three concrete things beats a longer one with ten generic ones, and it is faster to read, which matters when the reader has forty more.

One habit worth keeping: save the version you actually sent, next to the job description you sent it for. When the same role type comes up again — and it will — you are starting from something already tailored and already true, rather than from a blank page and a model that knows nothing about you.

The honest summary

AI is a genuinely good editor and a genuinely bad author for job applications. It will find the duty statement hiding as an achievement, the requirement you did not evidence, and the sentence that appears in every cover letter ever written. It will not know that the project you are underselling was the hardest thing you did that year.

Use it for the interrogation and the tightening. Supply every fact yourself. And keep the one test that matters: could you defend every line of this in an interview? If not, cut the line — not because a model wrote it, but because you cannot stand behind it.

Related articles