How to Write a Resume With AI [21+ Tips for 2026]

AI writes a plausible resume in about 4 minutes, and plausible is the problem: the same 4 minutes are available to every other applicant, and screeners now read the output all day. Used lazily, AI produces the most average document in the pile. Used the way this guide lays out, in 26 tips, it drafts, tailors and keyword-matches while the judgment stays yours.

What AI is actually good at (and where it lies)

Three jobs suit it: turning rough notes into structured first drafts, rephrasing a clumsy line 5 different ways, and comparing text against a job posting. One job ruins resumes: generating content from nothing, because a model prompted with only a job title produces confident, generic, occasionally false claims. MIT's career office keeps a blunt resume toolkit for exactly this reason: the content standards do not change because a model typed it.

It also helps to know where the failures come from mechanically. Models complete patterns, so a missing number gets filled with a plausible one, a 5-month stint rounds toward a year, and a modest title drifts toward the impressive version of itself, all delivered in the same confident tone as the true parts. Nothing in the output flags which is which; only your records can. That is why every drafting tip below feeds the model your material, and every editing tip audits the result against it.

AI also does not change what a resume is. The sections, the reverse-chronological order and the final checks are the same as writing the resume by hand; the tips below only change who types the first draft.

Before you prompt: tips 1 to 4

1. Build the raw-material file first. Ten minutes in a blank note: every job with dates, and every number you can attach to your work. Budgets, team sizes, percentages, volumes, and the messy half-remembered wins too. AI is only as honest as this file, and every later tip assumes it exists.

2. Paste the job ad into the chat. The single highest-value move in this whole guide. With the posting in context, every draft leans toward the employer's own vocabulary and priorities; without it, you get boilerplate for a job title, phrased like everyone else's boilerplate for that job title.

3. Give the model a role and a reader. Open with one line: you are a recruiter who screens 200 resumes a week for [role]; write for a 7-second skim. Output style follows the reader you define, and without a defined reader the model writes for nobody, which is exactly how the output reads.

4. Strip what does not belong in a chat. Your address, phone number and references have no business in a third-party tool, and neither do a current employer's confidential figures. Draft with placeholders, keep sensitive numbers as ranges, and add the real details in the document itself.

Drafting with AI: tips 5 to 10

5. One section per prompt. Whole-resume prompts produce mush, because the model averages 5 different jobs into one tone. Draft the experience section role by role, then skills, then the summary, each with its own context and its own constraints; the table below is the whole workflow in one view, and each row is a separate conversation.

The prompt that does the work

6. Draft bullets from your notes, never from the title. The prompt that works hands over your raw material and constrains the shape, like the card below. The prompt that fails is write me bullets for a marketing manager, which can only return the average of everyone else's resume.

7. Ask for 3 versions of everything. The first output is rarely the best and always the most generic. Three variants per bullet gives you parts to splice: the verb from one, the structure from another, your number in all of them.

8. Constrain the shape, not just the topic. Verb, task, number, under 20 words. Constraints are what separate a resume bullet from a paragraph of enthusiasm, and they are the difference between output you edit and output you rewrite.

9. Write the summary last, from the finished page. Paste your completed resume back in and ask for three sentences: title and years, two proof numbers, the fit line. Written first, the summary invents a resume that does not exist yet; written last, it compresses one that does. The shape and 30 finished models are in resume summaries by role and level.

10. Keep the master prompt. Save the setup that worked (role, reader, constraints) as one reusable block, alongside the raw-material file from tip 1. Those two documents are the actual asset; any chat session is disposable.

Editing the output: tips 11 to 16

Hearing the tells

11. Replace every adjective with a number. Generated drafts run on impactful, dynamic and proven. Each one is a slot where a figure from your notes belongs, and the swap is usually one edit: impactful campaign becomes campaign that booked 240 demos.

12. Read it aloud and cut what you would not say. The fastest AI detector is your own mouth. A line you would not say in an interview does not belong on the page you will be interviewed about, and the awkwardness you hear reading is the same one a screener hears skimming.

13. Re-verb the bullets yourself. Models overuse the same dozen verbs, and verb choice is a truth claim: led, supported and contributed to describe 3 different jobs. Purdue's Online Writing Lab keeps a categorized action verb list; one distinct verb per bullet, no repeats within a role, each one at the honest altitude.

14. Break the rhythm. AI bullets come out the same length with the same cadence, and screeners hear the drum before they read a word. Vary lengths; let one bullet be 8 words, and let the biggest number sit in the shortest line.

15. Audit every fact against the raw-material file. Dates drift, titles inflate and numbers round themselves up in generation, always in the flattering direction, which is exactly why the drift is hard to catch by feel. The audit is mechanical: every number and date on the page gets found in the file or fixed. It takes 3 minutes; an interview contradiction costs the offer.

16. Kill the tells. Spearheaded, results-driven, dynamic professional, passionate about excellence: the whole empty-verb, empty-adjective family. Any phrase you have read in 50 other resumes is doing you damage in the 51st.

Matching the job with AI: tips 17 to 20

17. Run the keyword gap. The comparison below is AI's highest-value resume trick, and it writes nothing at all. Add only the true matches afterward; which skills earn a line is still an editorial call.

18. Mirror the exact strings. Screening software matches literally, so the posting's Google Analytics 4 beats your GA. This mechanical layer decides more outcomes than phrasing: in Harvard Business School's Hidden Workers study, more than 90% of surveyed employers using recruiting software relied on it for the first cut.

19. Tailor per posting in minutes, not hours. With the master resume and master prompt saved, each application is one gap-run and 2 or 3 swapped bullets. That is the honest speed AI adds: not writing the resume faster once, but making the 15-minute version of tailoring real across a 30-application hunt.

The file still matters

20. Check the file, not just the words. Keyword-perfect text inside a two-column layout still dies at the parser; how tracking systems read the file is a formatting problem no model fixes for you, and it gets checked after every AI edit, not before.

Section-by-section prompts: tips 21 to 24

21. Summarize the finished page, not the ambition. The summary prompt runs on the completed resume, which is why it comes last in the workflow even though the section sits first on the page. Give it the page and the posting, ask for the three-sentence shape, and pick the version whose numbers you would defend first.

22. Mirror the skills section mechanically. From the gap run in tip 17, ask for your skills list reordered to the posting's priorities, in the posting's exact strings, capped at 12. This is the one section where accepting AI output nearly verbatim is safe, because it is your own list rearranged.

23. Draft the cover letter from one story. Hand it the resume, the posting and the single anecdote you want told, and ask for 3 short paragraphs around that story. Without the anecdote constraint you get a summary of your resume in letter form, which is the cover letter nobody needed.

24. Rehearse the awkward parts before anyone asks. Gaps, short stints and career changes: ask for the one-sentence explanation of each, stated plainly and without apology, then say it aloud until it sounds like yours. The line does not go on the resume; it goes in your mouth for the screening call the resume earns, which is a use of AI nobody talks about and everybody should.

Chat model or builder AI?

A general chat model gives you the strongest drafting and comparison engine with zero workflow: you carry text back and forth by hand. Builder-integrated AI gives you a weaker generator sitting exactly where the formatting, templates and export live, which converts the whole loop of draft, place, restyle, export into one screen. The honest split for most people is both: a chat model for the thinking work, tips 6, 17 and 24, and the builder's assistant for in-place phrasing while the layout stays locked. The one combination to avoid is neither, hand-typing into a raw document while also hand-carrying every edit through a chat window, which spends the tool era's time savings on clipboard work. Which builders bundle AI worth using, and which export the result as a free PDF, is exactly what the ranked and tested resume builders page scores.

A worked example: notes to bullet, start to finish

Here is the whole pipeline on one real-shaped case, a retail shift lead drafting one experience bullet. The notes are deliberately messy, because that is what real raw material looks like, and the point of the exercise is watching what each stage adds: the model contributes structure, the human contributes truth.

The first pass did real work: structure, verbs, compression. The edit did the irreplaceable work: pinning the numbers to what the notes actually support (6 hires, not the flattering 7), swapping greatly improving for the concrete outcome, and keeping the phrase now store policy that a model could not know. Two minutes per bullet, and the page stops sounding generated.

What never gets delegated: tips 25 and 26

25. Never let AI invent. No imagined metrics, no rounded-up titles, no skills you cannot demonstrate on a Tuesday. The resume is a claims document, and the interview is its audit.

26. Do the final pass with your hands. The last read is aloud, backwards for typos, and against the posting once more, the same finishing routine as the highest-payoff resume fixes. AI got you to a strong draft faster; the send decision, and everything the send implies, stays yours.