Skip to main content
AI Strategy

How to Build a Resume With AI — a 5-Phase Playbook

👩‍💻
Pramida Tumma
Co-Founder & CTO
April 14, 2026
8 min read
AIResumeCareerHiringATS

In my last job search, I paid for a professional resume service. I walked away with a clean PDF that I was never quite satisfied with.

In my most recent round, I made 67 versions of my own resume using AI. I spent zero rupees. And for the first time in my career, I ended up with a resume I actually wanted to hand over.

Here is what I learned along the way — the dos, the don'ts, and the part that most people miss.

Why Resumes Got Harder, Not Easier

Writing a resume became a two-sided AI problem

The popular story is that AI has made resumes easy: paste your career, ask for a draft, ship it. The real story is that both sides of the table are now running the same category of tool.

On the candidate side, most of us are using ChatGPT, Claude, or Gemini to draft, polish, and tune every bullet. On the recruiter side, companies are running ATS plus LLM-assisted screening that decides whether a human ever opens the file.

Which means writing a resume is no longer about impressing a recruiter. It is about satisfying three readers, in order — and each of them pattern-matches differently.

The Three Readers

The recruiter's AI filter:keywords, format, role match
The recruiter:skimming in ten seconds, looking for a reason to say yes or no
The hiring manager:reading carefully, looking for the person behind the role

Phase 1 — Clarity, Before AI Touches the File

The step most people skip

Start by writing down everything. Every project, every role, every number, every achievement you might possibly want on your resume. Do not worry about length, format, or polish. Write as if you are telling a friend what you have actually done in your career. If it takes five pages, let it take five pages. This is your raw material, not your resume.

While you are doing this, write down the three things you want any reader to remember: the five most important projects, the domains you have worked in, and the named achievements that belong on page one. Do this before you open any AI tool.

Then, before you ask for a draft, ask the model to research. Something like: "What does a strong resume for a senior engineering leadership role look like in 2026? What formats are preferred in Europe versus the US versus India?" Let it build context before it writes a single bullet. I was unsure whether to include a photo — the research answer was specific: in European countries photos are not appreciated and are sometimes actively discouraged. That would have taken me an hour of searching.

🧭

AI is very good at generating text. It is not good at knowing what you should be leading with. Start with a blank prompt and the model will confidently write a generic resume — then you spend hours fighting it back to the one you wanted.

Phase 2 — Draft, Two Ways

The mistake is never switching from the first mode to the second

Two legitimate drafting modes

Full write
Paste your career notes, ask for a complete resume in your target format in one pass. Best when you have no draft at all and want something to react to.
Refine
Bring your existing resume and rewrite a single bullet or section at a time. Best when you have a draft you mostly like and want to sharpen specific parts.

After the first full draft, switch. Bullet by bullet, section by section. That is where the quality comes from.

One practical note on format: I used Claude to generate the final resume as HTML and saved it as a PDF. The typography came out clean, the layout held up across devices, and I did not need a separate design tool in the loop.

Phase 3 — Verify Every Line

Hallucination on a resume is disqualifying

I have personally seen drafts invent job titles I never held, inflate team sizes, attach awards I did not win, and place me on projects I was not part of. Every single line needs a verification pass.

The verification bar

  • If a number appears in a bullet, you should be able to defend it in the interview
  • If a technology appears in your skills list, you should be able to use it in a live exercise
  • A resume that cannot survive an interview is worse than a resume that never got one

Phase 4 — ATS Alignment, Without Overdoing It

Aim for 70–80% with specifics you can defend

Paste the actual job description into the model. Ask it to identify the keywords your resume is missing, and ask for a match percentage.

Here is the part most advice skips: do not add keywords you cannot back up. Use this step to surface gaps in how you are describing work you actually did, not to invent work you did not do.

🎯

A high match percentage on a resume that collapses in the first technical conversation is the worst kind of optimization. I have seen people push to 95% by stuffing keywords and then stumble on the first question.

Phase 5 — The Human-Voice Rewrite

The step that separates a good AI-assisted resume from a great one

When you are done with all the AI work, go back and rewrite in your own words the parts you actually want a human to read: the two or three bullets that summarize the project you care most about, the one line in the executive summary that says who you are, the single phrase that tells a hiring manager what they will get if they hire you.

✍️

"We don't see things as they are; we see them as we are." — Anaïs Nin. AI reads your career as keyword density and role fit. A human reader looks for voice, judgment, and taste. Both readings matter. Only one of them is automated.

The Dos and Don'ts

Sixty-seven versions, compressed

Start with what you want highlighted
Decide before you open any AI tool.
Verify every fact, on every draft, every time
No exceptions — this is the disqualifying failure mode.
Use AI for grammar, format, ATS alignment, and first drafts
These are the tasks it is genuinely good at.
Rewrite the most important lines in your own words
The lines you most want a human to read.
Keep a single source-of-truth document
Every project, number, and achievement. Feed that into the model each time — not your last draft.
Do not paste the AI draft directly as your final resume
The seams show.
Do not chase 100% ATS match
70–80% with authenticity beats 95% with invented skills.
Do not let AI choose what to highlight
That is your decision.
Do not ship the same generic resume for every role
Fifteen minutes of tuning per job description is where AI is actually useful.
Do not trust a bullet that sounds too polished
If it reads like a consulting case study, it probably does not sound like you.

Conclusion

"The humans hiring and the humans being hired both need to show up with something AI cannot generate for them."

For the resume, that something is taste: what you chose to lead with, which projects you put on page one, which numbers you claim and which ones you leave off. AI helps you structure. Your judgment is what ships.

The shift is not that AI has made the task easier. The shift is that AI has raised the ceiling of what you can do yourself — if you put in the clarity work first, verify every line, and rewrite the parts that matter in your own words.

That is the whole playbook. Sixty-seven versions was how long it took me to trust it.

About the Author

👩‍💻
Pramida Tumma
Co-Founder & CTO

Technology leader with expertise in AI/ML implementation and enterprise software development

View Medium Profile
Working on this?

We do this for a living.

If this article describes a decision you are making right now, bring it to a Blueprint Session — a working session, not a sales call.