Okay, I need to vent about this a bit because I was genuinely excited for it. I'm trying to organize my team's external contacts—vendors, support contacts, other devs we collaborate with—into a Notion database. I thought the new AI autofill for properties would be a huge time-saver, pulling in company info, titles, etc., from just an email or name.
But in practice? It's wildly inconsistent, to the point of being unusable for anything serious.
Here's what I mean. I created a "Contacts" database with properties like "Company," "Role," "Website." I'd type in a LinkedIn profile URL (or even just a name and company), hit autofill, and get... weirdness. For one DevOps tooling vendor contact, it filled the "Company" field with their personal GitHub handle instead of the actual company name. Another time, for a known SaaS platform, it populated the "Website" property with a Wikipedia page for the *city* the company is headquartered in, not their actual site.
It feels like it's pulling from some generic, unstructured web scrape without any real verification. In our world, if a CI/CD pipeline is that unpredictable, we'd rip it out immediately.
I'm curious if anyone else has tried using this for similar structured data? Have you found a workaround, or is it just not ready for this kind of use case? I love the idea of AI automating the grunt work, but the output has to be reliable. Right now, I'm spending more time correcting its guesses than I would just filling the fields manually.
Maybe I'm using it wrong? My setup is pretty basic:
```plaintext
Properties:
- Name (Title)
- Email (Email)
- Company (Text) -> AI autofill
- Role (Text) -> AI autofill
- Website (URL) -> AI autofill
```
The promise is there, but the execution feels like a beta feature.
Learning by breaking