I Let AI Fill Every Form for a Month. Here's What Surprised Me.

The rules of the experiment

I review Chrome extensions for a living. Most of them end up uninstalled within 48 hours. So when I decided to test an AI form filler for a full month, I gave myself strict rules.

1. Every form gets the AI treatment first. No manual filling unless the AI fails completely.
2. Log every form. Category, number of fields, what worked, what broke.
3. No cherry-picking. Government sites, sketchy checkout pages, that one dentist office with the form from 2009. All of them.
4. Track time saved by comparing against my average manual filling speed (measured in week one).

The extension I used was Filliny. I picked it because it uses actual AI to read forms instead of matching HTML field names like browser autofill does. That distinction matters. A lot.

Here is what happened.

Week 1: calibration and early skepticism

The first three days felt weird. I kept reaching for my keyboard to type my email address and then catching myself. Muscle memory is a strong thing.

I set up two profiles: one personal (home address, Gmail, casual tone) and one for work (office address, company email, professional tone). That took about 4 minutes total. Then I started logging.

Week 1 numbers:

• Forms filled: 87
• Total fields: 1,438
• Fields the AI got right on first try: 1,391 (96.7%)
• Manual corrections needed: 47 fields across 22 forms
• Complete failures (AI could not fill anything): 3 forms

The 3 failures were all on the same type of site: custom-built web apps that render form fields inside iframes with cross-origin restrictions. One was a government tax portal. Another was an older health insurance enrollment page. The third was a bank loan application.

96.7% is solid for week one, but not mindblowing. I was still double-checking everything.

Week 2: the trust shift

Something changed around day 10. I stopped checking every single field before submitting. I would glance at the form, confirm the name and address looked right, and hit submit.

This is the part that surprised me. Not the accuracy numbers. The behavior change.

By week 2, filling a form went from a task I thought about to something that happened in the background. Click the Filliny button, scan the result, submit. Three seconds. The mental load of forms basically disappeared.

Week 2 numbers:

• Forms filled: 118
• Total fields: 1,926
• First-try accuracy: 97.8%
• Complete failures: 1 (a Flash-based form, yes, those still exist)

The accuracy bump from 96.7% to 97.8% was subtle but I felt it. Fewer corrections meant less friction. Less friction meant I stopped treating forms as obstacles.

By day 10 I stopped verifying every field. That trust shift was more valuable than the time saved.

The form categories that matter most

By the end of the month I had logged 412 forms across 9 categories. Here is how each one performed.

Shopping and checkout (164 forms)

The bread and butter. Shipping address, billing address, contact info. 99.1% accuracy. Filliny handled Shopify stores, WooCommerce sites, BigCommerce, and custom-built checkouts with equal competence. The AI reads labels like "Ship to" and "Where should we deliver?" the same way you would.

One quirk: some stores split the address into separate fields for apartment/suite number. Browser autofill always misses this. Filliny got it right about 90% of the time.

Account signups (89 forms)

Name, email, phone, sometimes a company field. 98.6% accuracy. The only consistent miss was when sites asked for a username instead of a name. Filliny would try to use my full name as the username, which obviously does not work on sites that ban spaces.

Travel bookings (42 forms)

Passenger details, passport info, frequent flyer numbers, emergency contacts. These are long forms. 97.2% accuracy. The misses were mostly on dropdown menus for nationality and country codes where the site used nonstandard option values.

Government and official forms (28 forms)

DMV appointments, permit applications, voter registration updates. 94.3% accuracy. The lowest category. Government sites are uniquely terrible. Many use custom form widgets, CAPTCHA breaks between sections, or multi-page wizards that reset if you navigate backward. Filliny handled the individual pages fine but could not push through CAPTCHA gates.

Job applications (31 forms)

Workday, Greenhouse, Lever, Taleo, iCIMS. 96.8% accuracy. The AI handled the big applicant tracking systems well. It figured out that "Preferred Name" is different from "Legal Name" and that "LinkedIn URL" wants a link, not just your name. Open-ended fields like "Why do you want this role?" got filled with text matching the tone I set in my profile.

Insurance quotes (18 forms)

These are the long ones. 30 to 50 fields per form. Vehicle info, driving history, coverage preferences. 95.4% accuracy. Filliny filled personal info flawlessly but needed manual input for vehicle-specific details like VIN numbers and current mileage, which makes sense since those are not in my profile.

Surveys and feedback (24 forms)

Post-purchase surveys, NPS ratings, product feedback forms. 97.9% accuracy. The AI filled in contact fields and wrote reasonable responses for text areas. I still picked ratings and multiple-choice answers myself.

Medical and healthcare (8 forms)

New patient intake, pharmacy accounts, appointment scheduling. 93.8% accuracy. Lowest after government sites. Health portals often run inside embedded iframes that limit what extensions can access. The forms Filliny could reach, it filled correctly. But some were simply walled off.

Everything else (8 forms)

Event registrations, newsletter signups, donation forms, library card applications. 98.2% accuracy. These tend to be simple forms with well-labeled fields. Easy wins for the AI.

The full month in numbers

After 30 days, here is the complete picture:

• Total forms filled: 412
• Total fields across all forms: 6,847
• Fields filled correctly on first try: 6,641 (97.0% overall accuracy)
• Manual corrections: 206 fields (3.0%)
• Complete form failures: 7 out of 412 (1.7%)
• Estimated time saved: 11 hours 20 minutes
• Average time per form (manual): 2 minutes 14 seconds
• Average time per form (AI-filled): 8 seconds

412 forms. 6,847 fields. 97% accuracy. 11 hours saved. One Chrome extension.

Five things that genuinely surprised me

I went in expecting the accuracy numbers to be the story. They were not. These five things caught me off guard.

1. I filled more forms, not fewer

This was the biggest surprise. Before the experiment, I actively avoided forms. I would skip surveys, abandon carts that required account creation, and ignore websites that asked for too much information up front.

Once forms stopped being annoying, I stopped avoiding them. I signed up for services I had been putting off. I compared insurance quotes on four different sites instead of settling for the first one. I actually filled out that gym membership form that had been sitting in my browser tabs for two weeks.

My form count went from roughly 8 per week in the first week to 16 per week by the end. Removing friction did not just save time on existing tasks. It removed the invisible barrier that was stopping me from doing things.

2. The AI handles open-ended fields better than I expected

Fields like "Delivery instructions" or "Tell us about yourself" are where browser autofill gives up entirely. Filliny actually writes contextual responses based on your profile and tone settings.

My delivery instruction field consistently generated something like: "Please leave at front door. Ring doorbell. Thank you." Not poetry, but exactly what I would have typed. And for job application cover letter snippets, it pulled from the professional tone in my work profile and generated something I only needed to tweak slightly.

3. Government sites are a special kind of broken

I already mentioned the 94.3% accuracy on government forms. But the story is more nuanced than that number suggests.

When Filliny could access the form fields, accuracy was actually above 98%. The failures were almost always access issues: embedded iframes, custom widgets that do not use standard HTML inputs, or CAPTCHA walls between form sections. These are problems no browser extension can solve. The sites are just poorly built.

For the record, Chrome autofill was worse on every single government site I tested. Most of the time it filled nothing at all.

4. Multiple profiles are underrated

I started with two profiles (personal and work) and ended with three. The third was a shopping-specific profile with my PO box address, a dedicated shopping email address, and a casual tone.

Switching between them takes one click. And it eliminated a problem I did not realize I had: accidentally using my work email for personal shopping and my home address for business orders. Those small mistakes happen more than people admit.

5. I started trusting the AI more than myself

By week three, I realized the AI made fewer typos than I do. My address has a tricky apartment number format that I mistype about once every ten times. Filliny got it right every single time because it pulls from my profile, not from my tired fingers at 11pm.

That said, I still use test mode on forms with financial consequences. The preview-before-submit feature is there for a reason.

Where it failed (and why I am okay with it)

No honest review skips the bad parts. Here is every type of failure I logged across 30 days.

• Cross-origin iframes (government and health portals): 5 complete failures. No extension can access these.
• Flash-based form (yes, in 2026): 1 failure. I was amazed this site still existed.
• Heavily custom React component with non-standard event handling: 1 partial failure. AI filled the fields but the site did not register the changes until I clicked into each one.
• Username fields (filled with full name instead): happened about 15 times. Quick manual fix each time.
• Ambiguous dropdowns ("Country" menus with nonstandard values): about 20 misses across travel and insurance sites.
• Payment fields: never filled, by design. Filliny skips credit card numbers and CVVs entirely. I consider this a feature, not a failure.

Total failure rate: 1.7% of forms completely unfillable. Another 3% of fields needed manual correction. For context, my manual typing error rate before the experiment was around 4%. So the AI was literally more accurate than me.

Security notes for the skeptics

Handing your personal information to a Chrome extension sounds risky. I get it. I spent the first three days reading their security documentation before committing to the full month.

• AES-256 encryption on all stored profile data
• No passwords or credit card numbers stored. Period.
• Data never sold, shared, or used for AI training
• Test mode previews fills before anything touches the form
• Full data deletion available anytime

The fact that it deliberately refuses to touch payment fields earns genuine trust from me. A tool that knows its boundaries is more trustworthy than one that promises to handle everything.

What it costs vs. what you get back

The free tier gives you 5 fills. Five real forms on real websites. For someone who fills out one or two forms a week, that might last.

I went through my 5 free fills on day one. Not even close to enough for a proper test, which is how I ended up on the Pro plan.

Here is my math after 30 days:

• Cost: $12 for one month of Pro
• Time saved: 11 hours 20 minutes
• Value of that time at $30/hour: $340
• ROI: roughly 28x
• Annual plan: $99/year ($8.25/month). Even better math.

Even if you fill half the forms I do, the numbers work. And if they do not, there is a 30-day money-back guarantee. I did not need to use it.

Week 3 and 4: the habit was set

By the second half of the month, I stopped thinking of Filliny as a tool I was testing. It was just part of how I used the internet. The same way you do not think about your password manager anymore. It is just there.

Week 3 accuracy hit 97.4%. Week 4 was 97.6%. Marginal improvement as I got better at setting up my profiles for edge cases. The tool did not get smarter. I got better at feeding it the right data.

One thing I appreciated: the vision mode that lets you see exactly which fields the AI identified before it fills anything. I used this sparingly by the end, but early on it helped me understand how the AI reads forms. Watching it correctly identify a "Apt/Suite" field that Chrome autofill ignores was oddly satisfying.

Would I keep using it after the experiment?

Yes. I switched to the annual plan the day after the experiment ended.

Look, I test a lot of extensions. Most of them are clever solutions looking for problems. This one solved a problem I had gotten so used to that I forgot it was a problem.

Typing your information into online forms is not a task anyone enjoys or benefits from. It is pure friction. And removing friction from 405 out of 412 forms in a month is about as close to eliminating it as I have seen.

The 7 forms that failed were all edge cases with technical limitations no extension can fix. I can live with that.

How to run your own experiment

You do not need 30 days. A week gives you enough data to decide if it is worth it. Here is the quick version:

1. Install Filliny from the Chrome Web Store. Takes 30 seconds.
2. Create one profile with your basic info. Two minutes.
3. Use it on every form you encounter for the next 7 days. No exceptions.
4. Count how many corrections you make. If it is under 5% of fields, you have your answer.
5. Notice the behavior change. That is the real value. Not the seconds saved per form, but the friction removed from your entire online experience.

The 5 free fills will get you through the first day or two. After that, you will know if it is worth $12 a month. In my case, it was not even a close call.

412 forms later, I am not going back to typing.