ATS Optimization•10 min read•

ATS-Friendly Resume Checklist: 12 Rules, Each Checked Against a Real Text-Extraction Test

Twelve resume rules, each tied to rows in a published CSV from 20 layouts run through the text extraction behind our ATS checker, plus what did not matter.

By CVPosh Team

TL;DR

  • Most "ATS-friendly" checklists repeat each other. This one has twelve rules, and each points to rows in a CSV from a test we published: one synthetic resume, 20 layouts (14 PDF, 6 DOCX), run through the same text extraction as CVPosh's free ATS checker.
  • The rules that mattered most are about where text lives: contact details in the body as plain text, headings as real words, and a PDF whose text you can select. In the data, those failures cost 20 points or lost whole fields.
  • Rules we did not find evidence for are not on the list. This tests text extraction, not Workday, Greenhouse, Lever or any employer system, and it is one resume. Treat it as a checklist of things to inspect, not as pass rates.

CVPosh is our product. The checker is free and needs no sign-up. Evidence below is cited as results.csv line numbers (line 1 is the header) in the published CSV; pdfparse and poppler are two PDF text extractors, mammoth reads DOCX.

Update, 2026-11-18: the checker's DOCX list-bullet handling was fixed (PR #123) after this test, and we re-measured the six DOCX files. Word list bullets now count, so the DOCX scores below changed from 90 to 100 and from 70 to 80; the header and footer files still lose their contact details. The original run is kept as published in results.csv.

Contact details

1. Put your name and contact line in the page body, not a Word header or footer

A DOCX with the contact line in the header lost the name, email, phone and link (33 of 37 items found, line 32). In the footer version, email, phone and link were lost (34 of 37, line 33). mammoth returns the body only. The same details in the body were all found (line 30). Each file scored 70 in the original run, a 20-point drop from 90 because of the missing email (15) and phone (5). After the checker started counting Word list bullets (PR #123) the same files score 80 against 100 for the body version: the 20-point drop is unchanged, and the contact fields are still missing.

2. Write email, phone and URL as visible text, never as an image or "Email me" labels

A contact line drawn as an image lost all three fields (34 of 37, lines 14-15). A line of link labels ("Email me", "Call me") lost them too (lines 24-25): the real address existed only as a link target, which neither PDF extractor returned. Both scored 80. We ran no OCR, so an image is simply absent from the text.

3. Make sure your LinkedIn URL does not wrap

In the two-column PDF and the icon PDF, the LinkedIn URL broke at its hyphen: pdfparse returned jordan- and sample on separate lines, and poppler dropped the hyphen (jordansample). The exact URL was missing in lines 4-5 and 12-13. It was intact where it fitted on a line (line 2). After exporting, search the extracted text for your full URL.

Headings

4. Use the section names a reader expects

With "About me", "Where I've worked", "Studies" and "Toolbox", every word was extracted (37 of 37) but only 1 of 4 sections was detected, by both extractors (lines 18-19), and the score fell from 100 to 80. Summary, Experience, Education and Skills are the words the checker looks for. We can't say what an employer system does with unusual names, only that a checklist-style parser has nothing to match.

5. Keep headings as real text

When headings were images, the text layer had no headings: 1 of 4 sections, score 80 (lines 26-27). Everything else was intact (37 of 37), which is why this failure looks fine on screen.

6. Skip letter-spaced headings

Letter-spaced capitals came out of pdfparse as E X P E R I E N C E: 1 of 4 sections, score 80 (line 16). poppler returned EXPERIENCE: 4 of 4, score 100 (line 17). The same file passed or failed depending on the extractor, so there is no safe way to know which your reader resembles.

Layout and order

7. Prefer one column; if you use two, read the extracted order

Both column layouts kept every word, but both extractors returned order 8 of 10 landmarks for the sidebar PDF (lines 4-5), with skills and education before the summary and experience. The DOCX two-column table did the same (line 35). A plain table that held dates in one column and roles in the other extracted in order in both formats (lines 6-7, 34).

8. Avoid drawn shapes and absolutely positioned blocks

Bullets drawn as small positioned dots left all nine bullets after the three job headings in pdfparse (order 7 of 10, line 10) and no marker character in either extractor. A one-page resume built from positioned blocks put skills first and the name last in pdfparse (order 7 of 10, line 22). poppler re-sorted both into correct order (lines 11 and 23).

Bullets, files and checking

9. Use text bullets (•, ➤, -), not drawn dots

A ➤ bullet came out as a character and all nine bullets were intact (lines 8-9). The checker once counted that as zero bullets and scored it 90. Since PR #118 it detects the glyph: re-scoring the same extracted text gives 100 (results-after-bullet-fix.csv, lines 8-9). The drawn-dot PDF still scores 90 there, because no marker was extracted.

10. In Word, keep real list bullets; do not retype them to chase a score

Word list bullets (line 30) and typed "•" characters (line 31) both extracted all nine bullet sentences (9 of 9). In the original run the checker scored them 90 and 100: mammoth.extractRawText returns list items without their marker, so the bullet rule saw none. PR #123 fixed this by marking each Word list item before scoring (parseDocxWithListBullets). Re-scoring the same files, both now count 9 bullets and score 100 (results-docx-list-bullets.csv, lines 2-3), and the live checker returned the same on 2026-11-18. So retyping bullets was never needed for parsing, and since #123 it no longer changes our score either.

11. Export a text-based PDF and try to select the text

The full-page-image PDF yielded no text at all: 0 of 37, 0 of 4 sections, score 30 (lines 28-29). Open your PDF and try to select a word. If you cannot, nothing else on this list matters.

12. Read the extracted text, not just the score

The checker scored the two-column PDF and the positioned-blocks PDF 100 although their order changed (lines 4 and 22): it has no reading-order rule. The two PDF extractors also disagreed on 4 of the 14 PDFs (02, 05, 08, 11: item count, order or sections). Paste your PDF text into an editor, or run the checker, and read what comes out.

The rules in one table

| # | Rule | Evidence (results.csv lines) | What happened | | --- | ---------------------- | ------------------------------ | -------------------------------------------------------------------------------------- | | 1 | Contact in body | 32, 33 vs 30 | Header/footer content lost, score 70 (80 after #123), 20 points below the body version | | 2 | Contact as text | 14-15, 24-25 | Email, phone, link lost, score 80 | | 3 | URL on one line | 4-5, 12-13 | Link item missing after hyphen wrap | | 4 | Standard section names | 18-19 | 1 of 4 sections, score 80 | | 5 | Headings as text | 26-27 | 1 of 4 sections, score 80 | | 6 | No letter-spacing | 16 vs 17 | 1 of 4 vs 4 of 4 by extractor | | 7 | One column | 4-5, 35 vs 6-7, 34 | Order 8 of 10 vs 10 of 10 | | 8 | No drawn shapes | 10, 22 vs 11, 23 | Order 7 of 10 vs 10 of 10 | | 9 | Text bullets | 8-9, 10-11 | 0 bullets counted, then 9 after #118; drawn dots still 0 | | 10 | Word list bullets | 30 vs 31 | 90 vs 100 originally; both 100 with 9 bullets after #123 | | 11 | Text-based PDF | 28-29 | 0 of 37, score 30 | | 12 | Read the text | 4, 22; 4 PDFs differ | Score 100 with changed order |

What did not matter

A Palatino font with ligatures extracted normally (lines 20-21). Icons next to contact details did not remove the values, although pdfparse added seven private-use characters (U+F393 to U+F6E0 range) to the text; the link loss in that file was the wrapped URL (rule 3). Tables for dates and roles extracted in order. A ➤ glyph was extracted as a character.

Reproduce one claim

curl -s https://cvposh.com/research/resume-layout-test/results.csv | python3 -c "
import csv,sys
for n,r in enumerate(csv.DictReader(sys.stdin),start=2):
    if n in (32,33): print(n,r['variant'],r['extractor'],r['email'],r['phone'],r['link'],r['score'])"

This prints the two DOCX rows where the contact fields are false.

Frequently asked questions

Does an ATS reject resumes with columns?

We did not test any employer ATS, so we can't say. In our extraction every word of the two-column files survived, but the reading order changed in one extractor.

Is DOCX or PDF safer?

Neither, in this test. DOCX had no ordering problems except the two-column table, but it was the only format where header and footer content disappeared (we did not test PDF headers separately).

Does a high checker score mean my resume parses cleanly?

No. The checker starts at 100 and subtracts points per rule. Two layouts with changed reading order scored 100, and, in the original run, a Word list-bullet file lost 10 points although all nine bullets were extracted (fixed in PR #123).

Should I retype my Word bullets as "•"?

Not for parsing: both versions extracted every bullet sentence, and since PR #123 our checker counts real Word list bullets the same as typed ones (both score 100).

Is this representative of real resumes?

No. It is one synthetic resume laid out by us in Chrome and a DOCX library; Word, Google Docs, Canva or LaTeX output may behave differently.

Methodology and data

  • Run: 2026-11-17, one fictional resume ("Jordan Sample", 228 words), 14 PDFs from Google Chrome 157 and 6 DOCX from docx 8.6.0. Only the layout changed.
  • Extraction: pdf-parse 1.1.1 and mammoth 1.11.0, as in the checker; PDFs also through poppler pdftotext 26.07.0. 37 content items counted per file by normalised substring; order is a 10-landmark test.
  • Scores: results.csv uses the checker scorer at commit 184d4b4. On 2026-11-18 we re-scored the same extracted text with the scorer after PR #118 (commit c6e34f9) into results-after-bullet-fix.csv; only the ➤ PDF changed (90 to 100). rescore.mjs reproduces it and fails if the old scorer does not reproduce results.csv. The DOCX files were re-scored again on 2026-11-18, after the checker's DOCX list-bullet handling (PR #123) was deployed, into results-docx-list-bullets.csv (numbers confirmed by uploading the six files to the live checker): the four DOCX with Word list bullets gain 10 points (90 to 100, 70 to 80), and the typed-bullet file stays 100. The header and footer files still lose their contact details; only their bullet penalty changed.
  • Limits: one resume, one run, no employer ATS, no OCR. Chrome's native list markers were not extracted, so the baseline draws "•" as text. This is a list of things to check, not rates.
  • Not claimed: length, fonts beyond one serif, file names, keywords, and date formats were not tested, so no rule covers them.

Tags:

ATS resumeresume checklistresume formattingresume parsingATS checker

Ready to build your resume?

Build and edit your resume in the browser with CVPosh. No account needed to start.

Build Your Resume