Date extraction is the single most common thing Runwita gets wrong, and the consequences cascade: a wrong meeting date means every relative due-date in the actions (“by Friday”) resolves to the wrong actual date. So this is worth getting right.Documentation Index
Fetch the complete documentation index at: https://docs.runwita.com/llms.txt
Use this file to discover all available pages before exploring further.
How Runwita decides the date
There are three sources, in priority order:- A regex sniffer that scans the meeting title plus the first 30 lines of the transcript for obvious date patterns:
2026-04-27,20260427,27 April 2026,April 27, 2026. If it finds one, that wins. - The AI’s extracted date, from the JSON the model returns. The AI is explicitly told to look for dates in the transcript body, filename, or title, and to not default to today.
- A fallback you set on the Process with AI screen. This is the date you pick before clicking the button. Used as the anchor for resolving relative due-dates (“by Friday”) and used as the engagement date if neither of the above produce one.
Common failure modes
”All my engagements show today’s date”
The most common report. Cause: neither the sniffer nor the AI found a date in the source, and the Process with AI fallback was left at today’s date (the default). Fix: on the Process with AI screen, always set the meeting date before clicking the button. Even if you think the AI will figure it out, the explicit date you set is the safety net. It’s a 5-second action that prevents this entire class of bug. If you’ve already saved a bunch of engagements with today’s date by mistake, you can edit each engagement’s date on its detail page. The journey timeline re-orders.”The AI returned today’s date instead of the real one”
Smaller models (gpt-5-nano in particular) sometimes copy the dateAnchor we send (the value you set on Process with AI) verbatim into the output’sdate field. We have an explicit guard for this, if the AI returns a date that exactly matches the dateAnchor, we discard it as an “anchor echo” and fall back to the regex sniffer.
If the sniffer doesn’t catch it either, the engagement date will fall back to your Process-with-AI input. So as long as you set the date on Process with AI correctly, this failure mode is harmless.
”The transcript clearly has a date but the AI missed it”
The regex sniffer should catch obvious formats. If you find a date in the transcript that the sniffer didn’t catch, it’s probably in an unusual format. Patterns the sniffer recognises:- ISO:
2026-04-27 - Compact:
20260427(filename style) - Day-month-year:
27 April 2026,27 Apr 2026 - Month-day-year:
April 27, 2026,Apr 27, 2026
27/04/2026, 04-27-26, Q2 2026), the sniffer skips it and the AI may or may not pick it up. Set the date manually on Process with AI in this case.
If you want the sniffer to catch a new format, tell us at hello@runwita.com, it’s a small code change and we’ll add it.
Editing the date after save
Click into the engagement on its journey page. The date field is editable inline, click it, change it, click out. The engagement re-orders on the journey timeline immediately. Note that changing the engagement date doesn’t recompute due dates on the actions. If the meeting was actually on April 27 but you saved it as April 29, and the AI extracted “by Friday” as a due date, that “Friday” was resolved using April 29 as the anchor. The action’s due date will be wrong, you’ll need to edit each one manually, OR re-run the extraction by clicking ↺ Reprocess with AI with the correct date set on Process with AI. This is why getting the date right before extraction matters more than fixing it after.What’s next
Wrong journey match
When the AI picks the wrong journey for a saved engagement.