Why Job Application Autofill Fails on Greenhouse and Workday

By Haseeb Kamran, Founder of VeloApply, 8+ years in recruiting · Updated July 24, 2026 · 11 min read

You click the autofill button. Your name and email appear instantly. For a second it feels like the whole thing is solved.

Then you scroll down. The work authorization dropdown still says Select. The time zone picker is empty. The question asking why you want to work there is a blank box. The consent checkbox is untouched.

You are back to filling the form by hand, except now you also have to check the parts the tool did fill.

This is not a bug in one product. It is the normal outcome of how most autofill tools are built, and it happens on the platforms where you apply most often. Here is what is actually going on, and what a tool has to do differently.

What is job application autofill?

Job application autofill is software, usually a browser extension, that completes online application forms for you. It stores your details once. After that it fills them into forms on career sites so you are not retyping the same information on every application.

The idea is simple. The execution is where tools separate.

Every application has two kinds of fields, and they are nothing alike.

Predictable fields. First name, last name, email, phone, LinkedIn URL, resume upload. These appear on every application. They are ordinary text inputs. Their underlying names are fairly consistent across sites. A tool can keep a list of likely names, find a match, and type the value in.

Everything else. Work authorization. Visa sponsorship. Notice period. Time zone. Country and phone code. How did you hear about this job. Have you worked here before. Are you subject to a non-compete. Do you consent to interview recording. Why this company. Demographic questions.

On a typical Greenhouse or Workday application, that second group is fifteen to twenty five fields. It is the reason one application takes twenty minutes instead of two.

Why do autofill tools stop after a few fields?

Because they were built for the first group and never solved the second.

The common approach is keyword matching. The tool scans the page for input elements, reads their names and labels, and looks for a match in its list. See email, type the email. See phone, type the phone number.

This works beautifully on text inputs. It falls apart everywhere else, for one reason that is easy to miss.

Why does autofill fail on Greenhouse dropdowns?

Look at a dropdown on a career site. It behaves like the dropdowns you have used for years. Underneath, it is usually nothing like one.

A traditional dropdown is a select element. All of its options sit in the page from the moment it loads. Any script can read them. Setting the value takes one line of code.

Greenhouse, Workday, Lever and Ashby do not use those. They build custom components instead, following the combobox pattern described by the W3C. What looks like a dropdown is really a text input, a wrapper, and a menu that gets created only when the control opens.

That single design choice breaks the standard approach three separate ways.

1. The options are not there to read

A script looking for the answer choices to "What time zone do you operate in" finds nothing. The list has not been built yet. It does not exist in the page.

To pick the right option, a tool has to open the control first, wait for the menu to render, then read what appeared. Most tools never do this. So they skip the field, or they guess.

2. Typing the answer often makes it worse

Some tools type the answer in, since the control is technically a text input. On a searchable component, typing filters the menu. Type an answer that does not exactly match an option and the list empties out.

Worse, the typed text can sit there afterwards. The field looks filled. Nothing was selected. This is how a former employer's name ends up stranded inside a question about non-compete agreements, or the word Google gets left in a dropdown whose only choices are Job Board, Social Media and Other.

3. The browser marks the click as fake

This is the part almost nobody knows about, and it explains the most frustrating failure of all.

When a script clicks something, the browser flags that event. According to MDN Web Docs, an event's isTrusted property is true only when the browser generated it from a real user action. Events sent by a script, including through HTMLElement.click(), are marked false.

Components can check that flag. Many React components also expect a full sequence of pointer and mouse events, not a single click. A tool that fires one simple click watches nothing happen, with no error to explain why.

The result: the menu opens on screen, something appears to move through it, and the field still reads Select when it all finishes.

Which fields break most often?

The failures cluster around the same handful of controls. If an autofill extension has let you down, it was almost certainly on one of these.

Why is the consent question the riskiest field?

This one deserves its own section, because getting it wrong is worse than leaving it blank.

Greenhouse asks candidates about its interview recording policy. The two options are close to identical paragraphs. One starts "I consent to my interview recording being used for candidate evaluation." The other starts "I do not consent to my interview recording being used for candidate evaluation."

Every significant word is shared except one.

A tool that matches options by counting overlapping words finds both equally attractive. It has a real chance of selecting the opposite of what was intended. And you may not notice, because both answers look reasonable at a glance.

Matching here has to understand negation, not similarity. A tool should refuse to select a negated option when the intended answer was affirmative, no matter how well the wording overlaps.

Can an autofill tool erase what you already typed?

Yes, and this is the failure that should worry you most.

A tool that leaves a field blank has cost you a few seconds. A tool that erases something you already entered has done real damage, and you may not catch it before you submit.

Here is how it happens. The tool types into a field to trigger a search. No match comes back. It clears the field to tidy up after itself. Whatever was in there before is now gone. A phone number you entered by hand. A city the site prefilled from your account.

A tool built properly captures the original value before it touches anything, and puts it back on every failure path. It also treats a field with real data in it as already answered.

These safeguards never appear on a feature list. They are the difference between a tool you can trust and one you have to audit afterwards.

Does adding AI fix the problem?

Partly. It also introduces a second set of problems worth understanding before you judge any tool that advertises AI.

The model has to see the real options

Ask an AI "How did you hear about this job?" with no list of choices and it answers something sensible like Google. If the actual options are LinkedIn, Job Board, Referral and Other, that answer matches nothing. The field stays empty.

The model needs the options that are genuinely on the page. Which means the options have to be collected first. Which brings us back to opening the control. AI does not remove that step. It sits on top of it.

An over-cautious model fills nothing

There is an obvious instinct when building this. Tell the model to leave a field blank whenever it is not certain, so it never invents anything.

Push that too hard and it returns empty answers for almost everything. Including questions it could have answered from the candidate's own profile. Country follows from a stated location. Time zone follows from a city. Work authorization in your own country is usually straightforward.

A model told repeatedly to be careful will decline all of it. The user sees an empty form and wonders what the AI was for.

What works is answering by default, with named exceptions. Fill anything that follows safely from the profile. Leave blank the categories that genuinely need the candidate's own knowledge, such as existing restrictive agreements or anything about criminal history.

Matching by meaning, not by letters

Even a correct answer fails if the matching is naive. The model answers Yes. The option on the page reads "Acknowledge / Confirm." The profile says Black and the choices include "Black or African American" next to "American Indian or Alaska Native."

A tool matching on loose substrings can land on the wrong choice. Matching has to work on whole words, treat affirmative phrasings as equivalent, and never let a negated option satisfy an affirmative answer.

What does a complete autofill tool actually do?

Here is the sequence that works. It is worth knowing even if you never build anything, because it gives you a way to judge what you install.

1. Fill the predictable fields directly. Name, email, phone and links need no AI. They should fill instantly and cost nothing.
2. Find what is genuinely unanswered. Read the value from where the component renders it, not from the surrounding text. On many sites that surrounding text includes the question, which makes every field look filled.
3. Open each control and read the real options. The only reliable source of truth for a dropdown is the dropdown itself, once open.
4. Send the question, the options and the profile together. The AI needs your background, the job context, and the exact choices available.
5. Match by meaning. Handle synonyms. Handle negation. Prefer whole-word matches. Refuse to guess when nothing fits.
6. Select the way a person would. A full event sequence, and where that is rejected, a genuinely trusted click.
7. Verify it stuck. Read the control back. A tool that assumes success reports a filled form that is still empty.
8. Close the menu before moving on. An open menu covers the next field and blocks everything behind it.
9. Handle one control at a time. Fire at every dropdown at once and each one closes the last one's menu.
10. Highlight everything it filled. You need to see at a glance what came from the tool.

This is the approach behind the VeloApply Chrome extension, and it is why it completes the sections other tools skip on Greenhouse and Workday applications.

Basic autofill compared to complete autofill

Why does remembering your answers matter?

Across a job search, the same questions repeat constantly. Are you authorized to work. Do you need sponsorship. What is your notice period. How did you hear about this role.

You answer them on application one. Then again on applications two through forty.

A tool that treats every application as a fresh problem asks the AI the same questions every time. That is slower than it needs to be. It can also produce slightly different answers to the same question, which does not look good if a recruiter compares two of your applications.

Memory changes the shape of the whole thing. The first application does the thinking. Later ones reuse what you already approved and fill instantly.

The more valuable half is corrections. When you edit an answer, that edit is the strongest possible signal about how you want the question answered. Your wording should be what gets used next time, not the model's original attempt.

Done right, the tool gets faster because it got more accurate.

What should you always check before you submit?

No autofill tool should submit an application for you. Any tool that offers to should make you suspicious. Mass submission is how people end up with hundreds of applications and no interviews, and how wrong answers reach employers under your name.

Read these four things before you press submit:

• Anything legal or eligibility related. Work authorization, sponsorship, non-compete agreements, background checks. These carry real consequences.
• Free text answers. A generated answer about why you want the role should sound like you and be true. Edit it. Your edit also teaches the tool.
• Anything about money. Never let a default answer set your salary expectation.
• Demographic questions. These are voluntary. If you would rather not answer, select the decline option instead of leaving a guess in place.

Review takes about a minute. Filling the form by hand takes fifteen or twenty. That trade is the whole point.

How do you test an autofill tool before trusting it?

Open a real Greenhouse or Workday application with a long screening section. Run each tool you are considering. Then check seven things:

• Did the dropdowns actually get selected, or do they still read Select?
• Did the consent and acknowledgement fields get handled, including checkboxes?
• On a consent question with a yes and a no option, did it pick the right one?
• Did anything you had already entered get erased?
• Was any text left stranded inside a field that was supposed to be a selection?
• Are the filled answers marked so you can review them quickly?
• Does it stop and let you submit, or does it try to submit for you?

Those seven questions separate tools that save real time from tools that fill four fields and leave you the rest.

The short version

Autofill on job applications is not hard because forms are complicated. It is hard because the fields that take the longest are custom components. They hide their options, ignore simulated input, and punish naive matching.

Filling a name is trivial. Selecting the correct option from a menu that does not exist until you open it, on a question the tool has never seen before, is the actual problem.

Any tool can do the first part. Judge them on the second.

Frequently asked questions