You tailored the CV, attached the file, watched the progress bar finish, and the next screen still asks for every employer, every start and end date, every job title you had just listed. The reflex is to assume the site is broken. It usually is not. Uploading a document and filling a form are two different operations serving two different audiences, and once you see the difference, a twenty-minute retyping session turns into a six-minute review.
What actually happens when you upload
The form does not read your CV the way a person does. It runs the file through a parser: a program that pulls out a stream of raw text, then guesses which fragments belong in which database column. It hunts for anchors. Something that looks like a job title, next to something that looks like an employer, next to something that looks like a date range. When those anchors appear in a predictable order, autofill works. When the order is disturbed, the parser still produces an answer, and that answer can be badly wrong.
The layouts that break the mapping
Two-column CVs. Text extraction is broadly linear. A sidebar sitting beside your experience can be pulled out interleaved, one line from the sidebar, one line from the main column. The parser then finds "Python" between your job title and your employer and has no idea what to do with any of it.
Tables, including invisible ones. Plenty of templates lay out dates and roles in a borderless table. Cells get read in an unexpected order, and the link between a date cell and the role beside it disappears.
Headers and footers. If your name, phone number and email live in the document header, many extractors skip them entirely, so you get empty contact boxes on a CV where the contact details are impossible to miss visually.
Icons and graphics. An envelope icon instead of the word "Email" gives the parser nothing to anchor on. Skill bars, rating dots and logos contain no text at all, so a page that looks information-dense can be almost empty to a machine.
Unusual date formats. Parsers handle a handful of patterns reliably. "03/2019 - 07/2022" is fine. "Sept '19 to date", "Spring 2021", "2019-" and roman numerals are not. The failure is often silent: rather than leave the field blank, the system commits a plausible but wrong year.
PDFs that are really images. Some design tools export flattened files, and any scan is an image by definition. There is no selectable text, so either every field comes back empty or the system runs character recognition and invents its own spelling of your employer. The test takes two seconds: open the PDF and try to select a line of text with your cursor. If you cannot, no parser can either.
The form asks anyway, even when parsing worked
Perfect parsing would still not remove the form, and it helps to know why before you resent it.
Structured fields are the only part that is searchable
Recruiters do not read applications in the order they arrive. They filter and sort: location, current job title, tenure, notice period, work authorisation, graduation date, salary expectation. Those operations run on database columns, and an attachment is not a column. Even where full-text search over documents exists, it returns a pile of files rather than a sortable list. If your eight years of experience only exist inside the PDF, you do not appear when someone filters for more than five.
Some fields exist for legal or process reasons
Right-to-work declarations, reference consent, data retention consent, equal-opportunity monitoring stored separately from your application, salary disclosure where local rules require it. None of that can be inferred from a CV, and in most cases a recruiter is not allowed to infer it. The box exists so there is a record of you answering it.
Knockout questions need an explicit answer
"Do you hold a valid driving licence?" "Can you be on site three days a week?" A line buried in a side panel of your CV is not an answer of record. The system needs an unambiguous yes or no to route your file, and a blank is treated as a no far more often than as a maybe.
Which version wins when the two disagree
Almost always the structured fields. The recruiter's first view of you is a row in a list built from those fields: name, current title, current employer, tenure, location. The CV gets opened later, if the row survives the first pass.
This is why leaving a mangled autofill in place is the expensive mistake. If the parser turned "Senior Analyst, Acme, 2019-2023" into a job title of "2019" at an employer called "Senior", you are a candidate with no readable history, and nobody opens the attachment to check whether the machine got it wrong. A broken form is worse than a blank one: blank gets rejected as incomplete, which sometimes prompts a follow-up, while wrong gets rejected as unqualified, silently.
The review routine
- Upload first, even when the form looks short. Correcting pre-filled text is faster and produces fewer typos than typing from scratch.
- Read every autofilled field top to bottom before touching anything. Do not fix as you go on the first pass. You want the shape of the damage first, because one misplaced block often explains four odd fields.
- Fix dates and employment gaps first. Dates drive the filters, and a parser that shifts one end date can invent a two-year gap you never had, or overlap two roles. Month and year for every entry, no blanks.
- Check the skills or keyword field. It is often pre-populated from whatever the parser found in your skills section, so it can arrive full of fragments, section headings or a stray employer name. Trim it to real skills, and add the ones the advert names, in the advert's words.
- Confirm your most recent role is actually flagged as current. Many systems have a "current position" tick box or sort entries by date, and a mis-parsed date can push a role from four years ago to the top. The row a recruiter sees is built from whatever the system thinks is current.
- Check that nothing landed in the wrong box. Phone number in the postcode field, full address inside the city field, a line of your summary inside "reason for leaving", employer name inside job title.
- Reread after saving. Some forms silently truncate at a character limit or reformat dates, so the version you approved is not always the version stored.
Should you upload at all when it is optional
Yes. A human reads the CV at shortlist stage, and it carries what the boxes cannot: emphasis, sequence, the context behind a career move. It is also the file a recruiter forwards to the hiring manager, so skipping it makes you the one candidate nobody can pass along.
But optional never means the form matters less. The fields decide whether a person ever reaches the document.
When the form and the CV contradict each other
Make them agree, literally. The same employer name in both, not "Acme" in one and "Acme Group Holdings Ltd" in the other. The same month-level dates. The same job titles. If your internal title was "Growth Lead" and your CV says "Marketing Manager", pick the one you can defend, use it in both, and put any explanation in the body of the CV rather than the title line.
Mismatches read as carelessness during screening, and become a real problem later, when a background check compares your stated dates against what a previous employer confirms.
Then treat a badly parsed autofill as free diagnostics on the document itself. If this parser could not find your last employer, the keyword screen at your next application will not find it either, and neither will the search a recruiter runs across their database six months from now. Fix the file, not just the boxes.
How to make the next upload parse cleanly
- One column for experience and education. Keep any sidebar for decoration you can afford to lose.
- Real selectable text. Export to PDF from your word processor rather than flattening to an image, and never submit a scan.
- Standard section headings: Work Experience, Education, Skills. Not "My Journey" or "Where I Have Been".
- The same order for every role: title, employer, location, dates. Consistency lets a parser learn your pattern from the first entry.
- Plain dates. Numeric month and year, or full month name and year, joined by a hyphen. Drop apostrophe years and season names.
- Contact details in the body of the first page, not in a header or footer.
- No text inside shapes, text boxes or graphics, since it often does not survive extraction.
- If the form accepts both, a .docx from a simple template usually parses at least as well as a PDF.
Re-applying and saved profiles
Most large employers run a candidate portal where your profile persists between applications. That is why the second application at the same company takes a fraction of the time: the history is stored and only the role-specific answers need attention.
Two things follow. Create the account properly rather than using a guest checkout style flow, because a stored profile is what makes future applications cheap. And remember that a stored profile preserves last year's bad parse forever, since every later application there inherits it. Spend the extra minutes once, on the profile itself.
When you re-apply, still open each field. Your latest role, notice period and salary expectation are the fields that go stale, and the fields the filters run on.