Chapter 2
How a CV becomes a record, and what breaks
A parser reads your file and fills in a database record: contact details, jobs, dates, education and skills. Whatever the parser misses is missing from every search a recruiter runs later.
What the record holds
Textkernel documents one of the most detailed outputs. For each job it records the employer and title as written and in a standard form, start and end dates, and whether the year, month or day was actually found. For each skill it records an id, months of experience, when it was last used and where in the CV it was found. It also returns "resume quality" findings about problems in the document PS13. Daxtra's parser likewise records skills with months of experience and when they were last used, and detects a UK locale for addresses, phone numbers and currency PS34.
That is why dates matter so much: months of experience per skill are worked out from the dates of the job the skill appears in. A job with no readable dates gives its skills no months Our view, from how these fields are described.
The parser is not where your CV is kept. Textkernel says its parsing service stores no CVs and parses in memory PS6; the record that results is kept in the agency's own system, which is why an old parse can follow you for years (chapter 1).
What vendors say breaks parsing
The clearest UK guidance we found is Mercury's page for its Daxtra parser. It asks for: a name plus a phone number or email; no headers or tables; no graphics, photos or WordArt; no image files; no text boxes, shapes or slanted text; simple fonts such as Arial, Helvetica, Times New Roman or Courier New; files up to 8 MB M1.
| Layout feature | What the evidence says | What to do |
|---|---|---|
| Columns | Bullhorn's 2026 parser update claims better handling of column layouts B1, which suggests columns were a weak point before, and CVs parsed before the update keep their old parse B1 Our reading. | One column. Put dates and locations on the same line as the job, not in a side column. |
| Tables | "No tables" M1. | Use plain lines and bullets, even for a skills list. |
| Headers and footers | "No headers" M1. A name plus phone or email is required M1. | Put your name, phone, email and town in the body at the top, not only in the page header. |
| Text boxes and shapes | "No text boxes, shapes or slanted text" M1. | Remove them; retype the text as normal paragraphs. |
| Images, photos, scans | No image files, graphics or photos M1. Textkernel through Mercury accepts image files M2; Bullhorn reads image-based and scanned CVs only on its updated parser, and older setups may need manual entry FP4. Workday advises avoiding images W1. | Send a file with real text in it, never a scan or a picture of a CV. Skill ratings drawn as bars or stars carry no text a parser can read Our view. |
| Section headings | Not documented: no vendor we read publishes the headings it recognises. | Use plain headings a person would expect: Profile, Experience (or Employment history), Skills, Education, Qualifications, Certifications Our view. |
| Dates | Textkernel reads ambiguous dates such as 5/1/09 using the country it detects from your address PS14, needs the file's last-modified date to interpret dates PS7, and records whether a month was found PS13. | Give a month and year for every job, in one format throughout ("Mar 2024 - Present" or "03/2024 - 09/2026"). Avoid day-first numbers that could be read either way. |
| Fonts and symbols | Simple fonts M1. | A standard font; plain bullets. |
File types
- Bullhorn parses PDF, .doc, .docx, .rtf, .txt and HTML FP4.
- Mercury's Daxtra parser accepts PDF, doc and docx, HTML, RTF, txt, ODT and Pages M1.
- Greenhouse accepts .doc, .docx, .pdf, .rtf and .txt FP1; Workable accepts .pdf, .doc, .docx, .rtf, .html and .odt up to 5 MB FP6; Ashby needs a file of 16 MB or less for full-text search and autofill FP13.
A Word file (.docx) or a PDF made from a word processor (so the text can be selected) is accepted everywhere we checked Our view, from the lists above. Keep it well under 5 MB.
How accurate are parsers?
Nobody knows independently. We found no published, independent head-to-head accuracy test of the main parsers; round-ups say the published figures are the vendors' own PS30. Affinda claims 92% overall PS11 and Daxtra "~90%" PS15, both vendor marketing with no published method.
No "ATS score" is the recruiter's score. Any tool that gives you a match percentage, including Fitread, is running its own scorer, not the agency's. Fitread's match rate is our own calculation and says how it is worked out; it is not what any agency or employer sees.
A test you can do in a minute
Open your CV, select all the text, copy it and paste it into a plain text editor. If the order is jumbled, words run together, or your contact details or dates are missing, a parser is likely to struggle too Our view. Fitread's "Text we could read" check reports whether text could be extracted; it cannot prove that every parser will read the file correctly (chapter 9).