How we test
Every fill rate on this site comes from one test, run the same way on every application system. Here is the test, the arithmetic, and what the numbers can't tell you.
The test, in six rules
- A real, public job posting. We open a live posting on each system and use the employer's own application form, exactly as an applicant would.
- A fictional applicant. The profile belongs to a made-up person: a London-based software engineer with a work history, education, skills, two languages and a CV file. On systems that need a candidate account, we create one for that fictional person.
- One press of Autofill. No second tries, no fixing things by hand first.
- The first page of the form. Where we measured a different page, that system's page says so.
- AI switched off. Only the profile fills the form, so the number shows what Autofill does on its own. AI answers are checked separately, as one of the 16 features in the score.
- Nothing is ever submitted. We read what's in each field and leave. No application is sent, on any system, at any point in the test.
How the fill rate is worked out
Fill rate = (fields filled correctly − fields filled wrong) ÷ fields the profile could have answered
- Filled correctly adds one.
- Left empty adds nothing, but still counts in the total.
- Filled wrong takes one away. A wrong answer is worse than a blank one, so it costs more.
- The result is rounded to the nearest whole per cent.
Some fields aren't counted either way:
- Consent boxes, signatures and sensitive questions. The test runs with default settings, which leave these for you, so they aren't counted as misses.
- Questions the profile has no answer for. If the fictional applicant has nothing to say, there's nothing to fill.
An example. A form's first page has 20 fields the profile could answer. Autofill fills 17 correctly, gets 1 wrong and leaves 2 empty. The fill rate is (17 − 1) ÷ 20 = 80%.
Two numbers in one cell, separated by a slash, mean two different postings on the same system, measured separately.
How the score is worked out
The score counts features, not fields. We check 16 things on every system:
| Feature | What we check |
|---|---|
| Text fields | Text and paragraph boxes are found and filled |
| Labels | Each question is matched to the right field |
| Dropdowns (standard) | Ordinary dropdown menus |
| Dropdowns (custom) | Custom-built dropdowns, set only when the site confirms the choice |
| Radio buttons | One-choice groups |
| Checkboxes | Single boxes and groups |
| AI answers | Questions the profile can't answer are drafted by AI and written in |
| Dates | Month and year pickers for work and education |
| Repeated entries | Several jobs and several degrees, each in its own block |
| CV upload | Your CV file is attached |
| Cover letter upload | Your letter is attached |
| Consent | Consent boxes are left and marked for you with the setting off; a single routine box is ticked and marked with it on |
| Sensitive questions | Left alone with the setting off; filled from your saved answers, and never sent to AI, with it on |
| Rich text boxes | Editors that aren't ordinary text boxes |
| Job description | The posting's description is captured from the apply page |
| The site's own CV import | Left alone during the test and checked last; we check the upload and parsing, never sending |
Each feature earns one point if it works, half a point if it works but fails in some cases, and nothing otherwise. The score is always out of 16.
That means a feature we haven't checked yet, or one the system doesn't have, scores the same as one that doesn't work. The results table uses these words:
- Fills
- Works.
- Partly
- Works partly, or fails in some cases.
- By hand for now
- Doesn't work yet. You type these.
- By hand
- Doesn't work, and we've decided to leave it to you: a fix would cost more than typing it yourself.
- Not checked yet
- We haven't tested this part here.
- Not on the tested form
- The form we tested didn't ask for it.
What “not measured” means
Some systems have no fill rate. That happens when:
- we couldn't reach a form, because it needs a candidate account we didn't have, or the posting showed no form;
- or we reached the form but couldn't complete a measurement on it.
"Not measured" means we don't know. It doesn't mean Trabalero fails there, and it doesn't mean it works. Workday and Paycor are listed with feature checks but no fill rate yet.
What the numbers can't tell you
We'd rather you read this before you trust a number:
- One posting per system, sometimes two or three. Employers set up the same system differently. Your form may ask questions ours didn't.
- One fictional profile. A different address, degree or job history can match a form differently.
- The first page only, unless the system's page says otherwise. Later pages can behave differently.
- A snapshot. Systems change their forms without notice. A number is true on the date it was measured.
- English-language forms only. We don't test forms in other languages yet.
- Filling, not hiring. A fill rate says how much typing Trabalero saved on our test. It says nothing about whether an application succeeds.
- Our test, not your data. The numbers come from our fictional applicant, never from anyone's real applications.
When we tested
The current figures were measured in September 2026. Each system's row in the table, and each system's own page, shows the date it was last tested.
| System | Fill rate | Last tested |
|---|---|---|
| Workday | not measured yet | 16 Sep |
| Greenhouse | 82% / 100% | 16 Sep |
| Lever | 90% | 22 Sep |
| Ashby | 75% / 100% | not settled yet |
| SmartRecruiters | 100% | 22 Sep |
| Workable | 88% | 16 Sep |
| Oracle Taleo | 100% / 60% | not settled yet |
| Oracle Recruiting Cloud (Fusion) | 91% | 21 Sep |
How often we re-test
We re-test the systems people use most at least once a quarter, and we re-test a system when reports say its form has changed. A date moves only when we've actually re-tested. If a number goes down, we publish it anyway, and What's new says what changed.
Report a broken form
If Autofill filled something wrong, or missed something it should have filled:
In the extension
Press Report a problem in the side panel and choose "A field was filled wrong or left empty". You'll see exactly what the report includes before you send it.
By email
Write to [OWNER: support email] with the site's name, the page of the form and what happened. Please don't send your CV or your answers; we don't need them.
Reports decide what we re-test first.
Try Autofill on your next application
The job application autofill that leaves the submit button to you.
Add to Chrome — free betaFirefox is coming.