Why our form filler stops before SubmitYour click, on purpose

Picture a visa application filled perfectly by software, except for one dropdown on page 7 where "Yes" was chosen instead of "No". If a person presses Submit, there's a moment where they might notice. If the software presses Submit, that moment never happens.

That moment is why AppliFlo for Chrome fills every page of an online form and then stops. It never submits, and it never pays. Here's the reasoning, and what it means if you fill forms for other people.

Why is the last click on a form a legal act?

The Submit button on most applications is attached to a declaration: that your answers are true, complete and yours. Many also charge a fee or record your consent. Pressing it is a statement made by the applicant, with consequences if it's wrong. Software can prepare that statement, but the person accountable for it should make it.

Look at what the forms actually say. Australia's Form 80, Part S, opens with "WARNING: Giving false or misleading information is a serious offence" and asks the applicant to declare that the information "is complete, correct and up-to-date in every detail" (Form 80 PDF (opens in a new tab)). Form 888, a supporting witness statement for partner visas, warns that the Migration Act provides penalties for false or misleading statements of "10 years' imprisonment or a fine of AUD313,000, or both" (Form 888 PDF (opens in a new tab)).

Online portals carry the same kind of declaration on the last page. Job applications ask you to confirm your answers are accurate. Payment pages take money. Consent checkboxes record that you agreed.

The last click is where three things happen at once:

This is general information, not legal or migration advice. A registered migration agent or lawyer can advise on your case.

What does auto-submit get wrong?

Auto-submit turns small, fixable fill errors into lodged errors that are slower and harder to fix. It removes the one review a person would naturally do, it sends generic answers where specific ones were needed, and it leaves no clear moment where someone approved what was sent. The time it saves is a few seconds; the risk it adds is the whole application.

What tends to go wrong:

• Silent misreads. A dropdown that matched "Korea, Democratic People's Republic of" instead of "Korea, Republic of". Nobody looked.
• Branching surprises. An answer on page 2 unlocks a question on page 6 the tool has no saved answer for, so it guesses or leaves a default.
• Generic answers. Auto-apply tools reuse text for questions that needed something specific to the role.
• Fees and duplicates. A retried submission can mean a duplicate application, or a duplicate charge.
• No approval trail. Afterwards, nobody can say who checked the application before it went.

Fixing a mistake before lodgement is a click. Fixing it after can mean a correction form, a phone queue or a refusal. In Australia, for example, there is a separate form for notifying the department of incorrect answers after the fact.

What does AppliFlo automate, and what does it hand back?

We automate the slow, repetitive parts: reading each page, filling fields from saved details or a client record, attaching uploads, and pointing to the exact field behind each site error. We hand back the decisions: approving sensitive details on each new site, reviewing each page, and the final submit and payment. Typing is ours; judgment is yours.

This split follows the same idea as flag, don't guess: software should be confident where it can be and visible where it can't.

How do you design a form filler that stops before Submit?

Design the stop as a feature, not a missing one. The filler needs to recognise when it has reached the final step, stop there every time, and hand control back clearly so the person knows the next click is theirs. A stop that's easy to miss, or easy to override, isn't really a stop.

The principles we follow:

1. Stop on the final step, always. There's no setting to switch it off, no "trusted site" exception.
2. Make the hand-off obvious. When the extension stops, it says so, and it's clear the next click belongs to you.
3. Never touch payment. No card entry, no "Pay now", no saved payment methods.
4. Show what was filled. You should be able to see what the extension filled on each page before you decide.
5. Ask before sensitive data, per site. Tax file numbers and bank details wait for your yes.
6. Keep values away from AI. AI sees field labels only.

The result is a number we're happy to publish: AppliFlo has submitted zero forms on anyone's behalf.

What does this mean for agents filling forms for clients?

For agents, the human stop is also the client-approval step. Fill the portal from the client's record, stop before lodgement, review with the client or under your own professional process, and only then submit. That gives you a clear point where the client confirmed the answers, and a cleaner record if anything is questioned later.

In Australia, registered migration agents work under a Code of Conduct. Among other duties, it requires the agent to advise the client in writing when an application is lodged and to give the client a copy (s 39), and to keep a client file for 7 years after the last action on it (s 56) (Code of Conduct, compilation 11 April 2024 (opens in a new tab)). A tool that lodges on its own doesn't remove those duties. A deliberate human stop makes them easier to meet.

Practical habits for a practice:

• Fill from one client record, not from old applications.
• Review the final page against the client's documents.
• Confirm the declaration with the client before lodging.
• Save the final review page and the lodgement receipt to the client file.
• Decide, in your process, who presses Submit.

If you run a practice, our migration agent paperwork playbook covers the whole case, from appointment to lodgement. If you're applying for jobs yourself, read how to autofill job applications without auto-applying.

What about AI agents that browse and submit for you?

AI agents that can operate a browser make auto-submit easy to build, which makes the question more pressing, not less. An agent can fill a form well and still misread one question. The same rule applies: let the agent prepare, let a person approve, and keep the final click with whoever makes the declaration.

A sensible rule set for any agent that touches applications:

• Prepare, don't lodge. The agent fills and stops; a person reviews and submits.
• Labels, not values. If AI needs to understand a field, it should see the label, not the applicant's passport number.
• Ask before sensitive data. Tax and bank details need an explicit yes for that site.
• Leave a trail. Record what was filled and who approved the final step.

These rules cost a few seconds per application. They also mean that when something is questioned months later, a person can say "I checked that page and I submitted it", which is exactly what the declaration says.

The short version

Software is good at typing. People are responsible for declarations. AppliFlo for Chrome, coming soon to the Chrome Web Store, does the typing on every page and stops where the responsibility starts. Our security page covers how your details are protected along the way.