Skip to main content

I’m getting the error “Something went wrong: Unable to extract content from attachment(s)” when I’m running a field agent on an attachment field.

The attachments are all PDFs. I’ve personally reviewed them all (there are only 17 rows) and they all render just fine in multiple PDF viewers, including Airtable’s.

This is not a consistent error. Using some AI models I do not get the error; with other AI models I get the error on first run with one row, but running it again clears the error.

What gives? Anyone else hit this or have any insight?

-Natalka

Hey ​@nroshak,

This issue has been raised multiple times on the community. In fact, I stumbled upon similar issues. I don’t have a clear answer for you.. worst case scenario, you can always build a side automation on Zapier, Make, or n8n, and get the heavy lifting done with it (whilst connecting whichever model you prefer). I know this is a pain. Until we find what patterns are causing this issues, you might want to open a ticket with support (support@airtable.com).

Also, if these are documents you’d have no problem sharing, feel free to send them over to mike@automaticnation.com and I can check if the same issue is encountered on my side.

Completely different matter, but would love to have you join our Airtable Hackathon! Make sure to sign up!!

Mike, Consultant @ Automatic Nation 
YouTube Channel 

​@nroshak 

I would open a support ticket with support@airtable.com about this issue.

In the meantime, I would suggest using Make’s AI automations, as ​@Mike_AutomaticN mentioned above. Make’s built-in AI automations will enable you to do this successfully.

If you’ve never used Make before, I’ve assembled a bunch of Make training resources in this thread. For example, here is one of the ways that you could instantly trigger a Make automation from Airtable. 

Hope this helps!

If you have a budget and you’d like to hire the best Airtable consultant to help you with this or anything else that is Airtable-related, please feel free to contact me through my website: Airtable consultant — ScottWorld

Hi,
usually in my tests default AI model ‘GPT-4’ produces less errors than others (even GPT-5 sometimes less reliable). Of course, when you want to use less AI credits, sometimes it’s wise to choose ‘low cost’ AI model, when it’s no difference between results. But in many cases change of model really make things worse.

Terms & ConditionsAccessibility statement