After the recent update to AI by Zapier, including the UI changes, I’m no longer able to perform OCR on images or PDFs.
Previously, I could extract text from images and PDFs using AI by Zapier alone. Could you please let me know whether this is still possible, and if so, how I can set it up?
Best answer by aiappsapi
One thing to watch with the Google Drive workaround: the standard share link points to Drive's HTML viewer page, not the file itself, so vision models usually cannot read it. After the file sharing preferences step, build the URL manually as drive.google.com/uc?export=download&id=FILE_ID using the file ID from the Upload File step's output. That returns the raw file bytes, which is what the AI step actually needs from a "publicly accessible URL."
Also worth adding a cleanup step at the end of the Zap that removes the sharing permission or deletes the temp file, since anything uploaded this way is readable by anyone who has the link.
+11I’m having the same thing!
I sent a message to support and they said it’s a “known bug” but no ETA on a fix.
As a workaround I’m using OpenRouter, sending the image/PDF to a model that supports file input to get back the information needed.
Thank you for your detailed response!
That’s helpful to know. I’ll try using OpenRouter as a workaround for now.
I hope they fix it soon!
+11Hi
Sorry to hear that you’re both running into this issue. I couldn’t locate the bug report the Support team added Andrew to, so
Keep us posted on how you get on. Hopefully the OpenRouter app helps to keep your Zap running in the meantime.
+11You're welcome
I heard back from the support team, and unfortunately, they confirmed that OCR won’t work with the method I’m currently using.
Here’s their response:
Looking at your Zap, the AI step is currently passing a Gmail attachment object directly into the prompt. The current version of AI by Zapier does not support processing file or attachment objects in this way. For image analysis, the image must be a publicly accessible URL. Gmail attachments are not publicly accessible URLs, so they cannot be passed directly into the AI step for analysis.
For PDFs, the "Upload a file" knowledge source in AI by Zapier does support text extraction, but this is a static pre-uploaded file, not a dynamic per-run input. It cannot process a different PDF on each Zap run.
For dynamic per-run OCR of attachments, the recommended approach is to use a dedicated document parsing integration. Two options available on Zapier:
PDF.co has specific OCR and document parsing actions that can process files per run.
If you are open to using your own API key, the Bring Your Own Key option in AI by Zapier lets you connect directly to OpenAI or Anthropic. Vision-capable models can analyze images via URL, though you would still need to make the file publicly accessible first.
+11Thanks for sharing this
For example you could use an Upload File (Google Drive) action to upload the file to Google Drive, then use an Add File Sharing Preference (Google Drive) action to get a publicly accessible link, then pass that to the AI by Zapier action. Then if needed, add a Remove File Permission (Google Drive) action after the AI by Zapier action to stop it from being publicly accessible anymore.
Do you think that could work?
One thing to watch with the Google Drive workaround: the standard share link points to Drive's HTML viewer page, not the file itself, so vision models usually cannot read it. After the file sharing preferences step, build the URL manually as drive.google.com/uc?export=download&id=FILE_ID using the file ID from the Upload File step's output. That returns the raw file bytes, which is what the AI step actually needs from a "publicly accessible URL."
Also worth adding a cleanup step at the end of the Zap that removes the sharing permission or deletes the temp file, since anything uploaded this way is readable by anyone who has the link.
Thanks for the advice!
I have a much better understanding of how it works now. I’ll take another look at the approach with actual operations in mind.
+11Keep us posted on how it goes
We're celebrating the Zapier builders making AI transformation happen. Winners get a published story on zapier.com, a trophy, $5,000 cash, and a feature at ZapConnect. Nominate yourself or a peer by July 24th, 2026.
No account yet? Create an account
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.
