This article covers what to include, where to find it, and what happens after you submit.
What to include
Every report needs four things:
1. Where it happened
Firm/Company name — the firm/company you were working in
Page URL — the full URL from your browser's address bar at the time
Thread ID or Close ID — part of the full URL from your browser's address
Time it happened — date and approximate time, including timezone
2. What you were trying to do
One sentence describing your goal. Examples:
"Import a Gusto payroll PDF as journal entries"
"Categorize last month's Stripe transactions"
"Reconcile the operating account for April"
3. What went wrong
What you expected the AI to do
What it actually did
The steps to reproduce it, numbered (1, 2, 3…)
4. Evidence
A screenshot of the AI's response and any error message
A short Loom (under 2 minutes) if it's easier to show than describe
The AI's response, copy-pasted — we need the exact text, not a summary
How to grab the URL & Thread ID
Click into your browser's address bar and copy everything. It should look something like:
The whole URL is more useful than a shortened version — it tells us exactly which firm, page, and thread you were in.
What we'll fix vs what needs your judgment
Puzzle AI is a powerful assistant, but you're still the accountant. The AI works best when you guide it with your expertise.
We'll always fix:
Incorrect calculations
Errors, crashes, timeouts
Workflow gaps — things that should be possible but aren't (e.g. file types we don't yet ingest well)
We'll work with you on:
Cases where the AI needs more direction to do what you want — sharing your prompts helps us improve the defaults
Ambiguous accounting judgment calls — the AI can suggest, but the final call is yours
If you've tried rephrasing 2-3 different ways and still aren't getting what you need, that's worth a report. It tells us our defaults need work.
Common reasons we ask you for more info
These are the most frequent gaps:
No Thread ID or URL — we can't see what the AI saw or did without one of these. The URL alone is often enough.
No screenshot of the AI's response — descriptions help, but we need the exact output to debug.
No reproduction steps — if we can't make the issue happen on our end, we can't fix it.
Multiple issues in one report — please file one report per issue. They often route to different team members.
FAQ
Do I really need a screenshot if I describe it well? Yes. Even a thorough description plus a screenshot is faster to resolve than a description alone.
What if I can't reproduce it myself? Tell us that. Include everything you remember — what you were doing right before, what action triggered it, whether it's happened before. Intermittent bugs are still worth reporting.
Can I report something that's not technically broken but just slow or annoying? Yes. Performance issues and friction points are bugs to us. Report them the same way.
The AI gave me an answer but it's wrong — is that a bug? Maybe. If it's a calculation error, a wrong category that's clearly identifiable, or a broken link, yes. If it's an accounting judgment call where reasonable accountants might disagree, that's worth flagging too — we use those cases to tune the AI.
How long until I hear back? See "What happens after you submit" above. Critical issues get the fastest response.
Where do I send the report? Reply to this conversation, or message us at [email protected].