Complete a 31-step web form: ChatGPT agent vs Perplexity Comet

Posted

Aug 2, 2025 by Rohit Mittal (@rohitdotmittal)

·

Last checked

Oct 16, 2026

·

Method:

How we verify

Evidence: Described by posterThe person describes what happened in words, without screenshots or video.

Outcome

On a 31-step form with buttons, dropdowns and fields, Comet failed after 3–5 steps, while ChatGPT agent completed all 31 and corrected its own mistakes by checking against the steps.

Agent

ChatGPT agent, Perplexity Comet

Strengthened by us to avoid the problem in the source. Not tested by the poster. It doesn't change what happened in the source.

Here's a task I'd like your help with. Read it all the way through before you start.

Goal: Fill out this long multi-step web form, checking each step against my list.

Check with me first: Do I have the list of steps and all the answers ready?

What I'll give you:
- The link to the form
- My step-by-step list and the answer for each field

Rules:
- Work through the steps in order. After each step, compare what's on screen with my list before you move on.
- If a click doesn't work (small buttons, dropdowns), try once more, scroll or zoom, then tell me the step number instead of guessing.
- If something doesn't match my list, fix it before moving on.
- If a page asks for a login or captcha, pause and let me handle it.
- Stop before the final submit so I can review.

Output: The form completed up to submission, plus a list of any step you had to correct or couldn't do.

You'll know it worked when: every step matches my list and the form is ready to submit.

Fill out this form for me.

Here's a task I'd like your help with. Read it all the way through before you start.

Goal: Fill out a long multi-step web form, checking each step against my list.

Quick check first: tell me if anything below is missing or unclear.

Inputs (fill in or confirm):
- Form link
- Step list with answers
- Fields that must match exactly (IDs, dates)

Rules:
- Work in order; after each step, compare entered values with the list.
- Report progress every 5 steps.
- Don't skip optional fields that are in my list.
- Save drafts where possible.
- Take a screenshot after each completed step.
- Tick off each step in my list as you go.
- Screenshot each page before moving on.
- Note any step that differed from my list.

What went wrong before, and how to avoid it:
- Comet failed after a few steps in the source test: if you lose your place, re-read the page and resume at the logged step.

Checkpoint: Stop before final submit.

If you get stuck: If the form errors, screenshot and stop. If a login, CAPTCHA or two-step code appears, stop and ask me to handle it; never guess or store passwords.

Verify before you finish: Final review against all steps.

Output: Step log and status.

Done when: all steps are filled correctly and ready to submit.

Source

https://x.com/rohitdotmittal/status/1951450652867174754