How We Tested Our Job Autofill Chrome Extension

We built this extension, so read this page as the maker's account, not an independent review. It's also why you won't find star ratings or customer quotes here. The US Federal Trade Commission's rule on consumer reviews bars a business from presenting a review site or group it controls as independent (16 CFR 465.6), and Google's review snippet guidelines make a business ineligible for review stars when it controls the reviews about itself. What I can give you is the method and the counts, including the ones that don't flatter us.

The short answer#

Jobs AI Autofill fills job application forms in Chrome from the profile and resume you saved in your InterviewChamp account. It drafts answers to written questions, lists what it left for you, and stops. You read the form and press Submit.

Our largest test graded fills on saved copies of 214 real application forms, for up to six made-up applicants each. Of the 18,870 boxes we could grade, 11,152 held an accepted answer, 5,013 were rightly left blank because the applicant's saved details didn't answer them, 2,690 were left blank when an answer did exist (each one listed for the applicant), and 15 were wrong. That test used a build from 3 October 2026. A second, smaller run on this month's code comes further down.

How the extension fills a form#

1. Save your details once. Your profile (contact details, links, work eligibility, optional self-ID answers) and your resume live in your InterviewChamp account.
2. Open an application. On 46 hiring systems, including Greenhouse, Lever, Ashby, Workday and Workable, it opens by itself. On other application pages you click its icon in Chrome's toolbar.
3. Let it read the form. It sends the form's questions and options, the page's web address and the job post text to our server, which works out which saved answer belongs in which box.
4. Let it fill what it can. That covers text boxes, dropdowns, checkboxes, your resume upload and the work history on your resume. A written question gets a draft you can edit.
5. Answer what it left for you. Anything it has no answer for goes on a list. Your answer is saved for the next form unless you untick the box.
6. Turn the page. On a form with several pages, you tap Next step; it presses the form's own Next button and fills the new page. If you turn on Click Next for me, it does that by itself after a short pause with a Stop button, and it still stops before Submit or a review page.
7. Press Submit yourself. It never does.

One behaviour catches people out. It ticks required agreement boxes, the kind that say your answers are true, though never a marketing opt-in. So read those boxes before you send.

How we tested it#

No real employer was involved at any point.

• Saved copies, not live forms. We saved complete copies of real application pages, then replayed each one on our own PC at its original web address. A network fence kept every request on the PC. In the newer run the pages tried to send data out 206 times; the fence kept each one on the PC, and none reached an employer.
• Made-up applicants. Six invented people: a US citizen who answers every question, a protected veteran with a disability and a security clearance but no degree, an H-1B holder who needs sponsorship, an F-1 OPT student who declines every self-ID question, a green card holder who fills only required boxes, and a London-based applicant who needs sponsorship. None of them is a real person.
• Nothing submitted. The test tool never presses Next or Submit, and it never starts the extension's "Apply for me" path.
• An answer key. For every question and applicant, the key lists the acceptable answers, or says the box should stay blank for the person. A second AI check read each entry. Where it disagreed with the key, we left that box out of the grading (942 boxes) instead of settling it in our favour.
• A key that only grows. A corrected entry has to name the entry it replaces and say why. After a review on 7 October 2026 we corrected 9 entries that were themselves wrong, for example a street-address box where the whole address is also fine. The counts here use the key as it stood on 22 October.
• The real extension. The lab build is made with the same build command as the Chrome Web Store package, with two lab changes: it talks to our server running on the same PC, and a site permission it normally asks you for is switched on ahead of time.

One more difference matters. The step that writes answers to open questions ran on a different AI model from the one live users get today, so this page doesn't grade written answers (263 boxes in the largest test).

What it got right, missed and got wrong#

The largest test ran overnight on 3 October 2026. It tried 309 forms saved on 1 and 2 October, from 22 hiring systems and four companies' own career sites, with six applicants each: 1,854 fills. 574 of them never got going. Our lab server stopped answering for the last part of the run, which accounts for 496; most of the rest failed while opening the saved page or its form. That left 1,280 graded fills on 214 forms from 18 hiring systems.

"Right" counts a box that held an accepted answer or stayed blank when it should have. "Ready to send" means nothing wrong and every required box done, so the applicant had nothing left to type.

Here's how I read it:

• It never skipped a box quietly. Every graded box it left blank went on the applicant's list. On four hiring systems, 108 more boxes were never picked up by the extension at all, so they aren't in the table.
• Expect to type something on most forms. Only 327 of 1,280 graded fills (25.5%) were ready to send untouched. Most forms asked at least one required question it left for the person.
• Most misses are location and employer-specific questions. Sorted by the words in each question, about two in three missed boxes asked where you live, whether you'd relocate or which office you want (899), or asked something only that employer asks (850).
• Workable had four graded forms. That's too few to say much about it.

The 15 wrong answers#

Fifteen wrong answers in 18,870 graded boxes is about 1 in 1,250. Here's every one:

• 6: a personal website typed into a box our test read as labelled only with the employer's name. That was two Lever forms, three applicants each, and the key said to leave it.
• 4: optional location or work-style questions it answered when the key said to leave them for the person: two about relocating, one about preferred working model, and one asking which state or region the London applicant lives in.
• 2: AI-written answers to experience questions the key said to leave for the person.
• 1: an internship period picked when the key said to leave it.
• 1: a pronoun question answered "Prefer not to say" for the applicant who declines self-ID questions, where the key said to leave it blank.
• 1: country left as United States for the London applicant.

None of the 15 was on a work-permit or sponsorship question, or on a gender, race, ethnicity, veteran or disability question. We haven't graded the newer code against the same key yet, so I can't tell you which of these are fixed.

Work permit and sponsorship questions#

These are the boxes I'd worry about most, because a wrong answer here can cost you the job (our guide Is auto-apply safe? explains why). 128 of the 214 graded forms asked whether the applicant needs visa sponsorship or is allowed to work in the country. Across the six applicants those questions came to 1,146 boxes: 794 right, 0 wrong, 227 left blank and listed for the person, and 125 we didn't grade. That zero is the number I watch. The newer run below didn't get a wrong answer on these either, but it left a required one blank on half its Greenhouse forms.

Gender, race, disability and veteran questions follow a stricter rule. When the extension recognises one, it fills it only from an answer you saved and never asks the AI. If you didn't save an answer, the box stays blank. In the largest test these questions came out 1,499 right and 0 wrong. But 237 stayed blank and went on the list even though the applicant had an answer saved, and a question worded in an unusual way can slip past the check, so look at these before you send.

A second run on this month's code#

On 21 October 2026 we ran the newest code on 30 forms saved on 13 October: 10 each from Greenhouse, Lever and Ashby, all filled for the US-citizen applicant. These forms don't have answer-key entries yet. So this run counts what it filled, then reads each box back to check the page holds exactly what the extension typed.

• Resume: attached on 30 of 30 forms.
• Read-back: of the 284 filled boxes, 253 read back exactly as typed, 30 were resume uploads (a page can't show those back as text), 1 dropdown couldn't be read, and 0 differed.
• AI-written: 9 of the 284.
• Left for you: questions only that employer asks (46), location questions (41), self-ID questions (29), work-permit and sponsorship questions (9, 7 of them required), and smaller groups such as file uploads, education and agreements.

Two findings we don't like. On 5 of the 10 Greenhouse forms, a required work-permit or sponsorship dropdown stayed blank and went on the list, because the extension couldn't read that dropdown's options. Counting voluntary self-ID dropdowns too (gender, Hispanic or Latino, veteran status), 7 of the 10 Greenhouse forms had at least one left blank even though this applicant had saved an answer. On Lever, the veteran-status question stayed blank on 5 of the 10 forms. Nothing was filled wrong, but you'd have to answer these yourself. The October 3 run got most questions like these right on older forms. We haven't found the cause yet, so on Greenhouse and Lever, check those questions every time.

Is the tested code the same as the store version?#

Mostly, and here's exactly where it differs. The Chrome Web Store version is 0.1.17, built on 20 October 2026; Chrome's update service was handing it out when we checked on 22 October. The second run's build is from the next day. Between the two, the extension's own changes are fixes for one hiring system that isn't in this test, a new question flow that's switched off, three parts of Apply for me (a form recorder, a form reader for our cloud browser, and a check it runs before pressing Submit, which also touched code the normal fill shares), a separate build for our own cloud browser that never goes to the store, and code moved between files without changing what it does. We expect the two to fill Greenhouse, Lever and Ashby forms the same way, but we didn't run the store build itself.

Every installed copy also asks our server which answer goes where. The second run used our server code from the same day. One change to the server's autofill code has landed since then; it changes how private answers are encrypted, not what gets filled.

What we changed on purpose for the lab:

• the server ran on our PC, not online;
• a site permission was switched on ahead of time;
• written answers came from a different AI model.

The largest test is older. Its build is from 3 October 2026, the candidate for version 0.1.14, and about 50 of the extension's files have changed between it and 0.1.17. Treat its numbers as a baseline, not a picture of today's version.

Does it ever press Submit?#

Not the fill you start. It never presses Submit, and a test in our code runs on every change and fails if any submit button receives an event during a fill. The one button it does press is a form's own Next: when you tap Next step, or after a short pause with a Stop button if you turned on Click Next for me.

There's one exception, and you can't reach it today. The same extension carries "Apply for me", which can press Submit on jobs approved in advance. Our server offers it only to our own team and a few test accounts; for everyone else, the Apply for me page says Coming soon. If that changes, this page and our store listing change first.

What it stores, and for how long#

All of this comes from our privacy policy, last updated 16 October 2026.

• In your browser: your sign-in, your settings, a short record of the forms it filled, and the names of up to five sites where it couldn't fill a form (our feedback form asks about them if you uninstall). It doesn't read your browsing history.
• In your account: your profile for as long as you keep the account, and your resumes until you delete them. Answers you type on a form are saved so the next form can reuse them, and you can see, change or delete them.
• Private answers: your date of birth, current salary, Social Security number, typed signature and conviction answer are optional. They're encrypted apart from the rest of your profile, used only for a question that asks exactly that, and never sent to an AI provider. A conviction question is filled only when your answer is No.
• What AI providers receive during a fill: the form's questions, your saved profile and resume (contact details included, never your demographic answers), and earlier answers to similar questions (never demographic, salary-history or work-permit answers).
• Logs and deletion: server logs are kept 30 days. If you delete your account, we delete its data within 30 days, and backups roll over every 30 days.

We don't sell this data or use it for advertising.

What it costs#

Filling in your saved details is free, with no limit. When the AI writes an answer on a form, that application uses one AI credit, however many pages the form has. The free plan comes with AI credits every day, shared with tailored resumes and cover letters, and paid plans have no limit. Today's prices are on our pricing page.

Who it's for, and who it isn't for#

It's for you if you apply on the hiring systems above, you're sending more than a few applications a week, and you'll still read each form before it goes. It saves the typing, not the judgement.

It isn't for you if:

• You apply mostly through Workday. Our store listing names Workday, but neither test here included a Workday form, so this page can't tell you how well it does there.
• You want applications sent while you sleep. It stops before Submit by design.
• You don't want your resume and the form's questions going to an AI provider. A fill sends them, as listed above.
• You don't use Chrome. It needs Chrome 116 or newer.

My own advice for the first week: read every box on every form, including the ones it filled. You'll learn quickly where it's reliable for your answers and where you still need to look.

You can install it from the Chrome Web Store listing or read more on the job application autofill page.

Key terms#

Hiring system

The software an employer uses to post a job and collect applications, also called an applicant tracking system or ATS. Greenhouse, Lever and Ashby are three of them.

Saved copy

A complete recording of a real application page, replayed on our PC so the extension can fill it without anything reaching the employer.

Answer key

Our list of acceptable answers for each question and each made-up applicant, or a note that the box should stay blank for the person.

Ready to send

A filled form with nothing wrong and every required box done, so the applicant had nothing left to type.

Agreement box

A checkbox that says your answers are true or that you accept the employer's terms. The extension ticks the required ones, so read them.

Related guides#

• Is auto-apply safe?: account restrictions, wrong work-permit answers, your data and scam services, plus questions to ask any tool.
• H-1B visa sponsorship for CS new grads: what to know before you answer a sponsorship question.
• CS new-grad resume tactics for ATS: getting your resume through the same hiring systems.
• Ghost jobs by the numbers: how long postings stay open, so you know which forms are worth filling.

Originally published on InterviewChamp.AI