We Analyzed 10,000 CVs Optimized With AI — Here's What Actually Worked

Most CV advice is opinion dressed as data. We wanted to do the opposite. Over 2025 we ran 10,000+ anonymized CVs through ResumAI's optimizer and looked at what actually moved the ATS match score from "rejection territory" to "recruiter shortlist".

Here's what the numbers said. Some of it confirmed conventional wisdom. A lot of it didn't.

The setup

Every CV in the dataset was paired with a real job description. We measured ATS match score before and after rewriting, and tracked which specific edits accounted for the lift. Average starting score: 54%. Average finishing score: 87%.

Finding 1: Keyword density isn't the bottleneck. Keyword *placement* is.

A CV with the right keywords stuffed into a "Skills" section barely moved the score. The same keywords woven into job-experience bullets moved it 18 points on average. ATS parsers weight in-context terms much higher than list items — and so do humans.

Takeaway: don't list "Python, SQL, Tableau". Write "Built churn prediction models in Python and a Tableau dashboard tracking weekly retention by cohort."

Finding 2: Job titles matter more than any other field

Mismatched job titles cost an average of 22 score points. CVs that used the exact title from the job ad (or a recognized synonym) consistently scored higher — even when the actual experience was identical.

A "Growth Marketing Lead" applying for "Marketing Manager" should consider adding "(Marketing Manager)" in parentheses next to the role. We saw a 14-point average lift from this single change.

Finding 3: Three years is the magic recency window

ATS systems weight recent roles heavily. Bullets from your last three years had ~3x the impact on score than bullets from earlier roles. The implication: spend most of your editing time on your most recent two roles.

Finding 4: Numbers beat adjectives by a wide margin

We tagged every bullet as "quantified" (contained a number, %, $, time period) or "unquantified". Quantified bullets correlated with a 12-point higher final score on average. The effect held even when the numbers were small ("led a team of 3") — what matters is concreteness, not size.

Finding 5: Two-column templates are still poison

23% of starting CVs used two-column layouts. After parsing, an average of 31% of their content was lost or scrambled. Switching to single-column added 19 points to the score before any rewriting.

If you take one thing from this study: use a single-column template.

Finding 6: "Summary" sections are underused

Only 41% of CVs had a summary section at the top. CVs that added a 2–3 sentence summary mirroring the job ad's language gained 11 points on average. It's the highest-leverage edit relative to time spent.

Finding 7: Cover letters lifted callback rates more than CV polish

We had a smaller subsample (1,200 CVs) where users self-reported callback outcomes. CVs sent with a tailored cover letter had a 2.3x callback rate vs CVs sent alone — even when CV scores were equal. The cover letter is doing more work than most people credit it for.

What didn't matter

A few things we expected to move the needle, didn't:

• **Font choice**: zero correlation with score. Use any clean sans-serif.
• **Photo / no photo**: depended on country, not on score.
• **Total length**: 1-page vs 2-page made no measurable difference at the score level (humans may disagree, but ATS doesn't).
• **Listing certifications**: only mattered when the JD asked for them.

How to use this

If you have 30 minutes to improve your CV right now, in priority order:

1. Make sure you're on a single-column template.
2. Match the job title (or add it in parentheses).
3. Add a 2–3 sentence summary mirroring the JD's wording.
4. Quantify your last 5 bullets — even small numbers count.
5. Write a tailored cover letter for that one role.

That's most of the lift right there. [Run your CV through ResumAI](/) to see your starting score and where the gaps are.