Skip to main content

Get Support
Have a question? Let's get you an answer!

How Do I...? Troubleshooting Code & Webhooks Developer Zone

Other Help Resources
Wait, there's more! Check out these Community sourced spaces for additional resources.

Zapier Help Center Hire an Expert

Learning Resources
All of our best learning resources can be found here.

Featured Articles Show & Tell Zapier's products AI Hub Zapier Tools Hub

Share your knowledge
Give back to the Community by sharing your automation wins and helping to answer questions.

Share your success See unanswered questions User Groups

Other Learning Resources
Some of our favorite additional ways to upskill your Zapier knowledge.

Zapier Learn Zapier Blog Zapier Webinars

Product Updates
Check out what is new and upcoming with Zapier with our regular product updates.

Product Updates

Early Access Program
Want to join Early Access? Learn more here.

When I use my ChatGPT ZAP to upload a TXT file for analysis and response, the document uploads properly and I am able to get responses. The problem is that the responses are incorrect. I have tried multiple different variations of how to phrase the questions, different version of ChatGPT, etc. All result in incorrect answers. 

However, when I login to ChatGPT directly and upload the same document and ask the same questions, I get the proper answers. Not sure why using the Zapier process yields incorrect results. I have another Zap I use to analyze PDF files and that ZAP works perfectly. 

Is the Zapier ChatGPT limited or work differently than chatting directly with ChatGPT? If so, is there a way to get responses as if you are chatting directly with ChatGPT using Zapier? Can I create a Python script or some other solution to get accurate results? 

I also noticed that when I am chatting directly, I can upload an Excel file for analysis but cannot do so using the Zapier integration. I am just trying to understand why there is such a difference in responses between Zapier and chatting directly with ChatGPT? 

Hi @smccarthy956 

For us to have more context, post screenshots with how your Zap steps are outlined and configured.

 

 

https://help.openai.com/en/articles/8983675-what-types-of-files-are-supported

Here are the steps:

 

@smccarthy956 

Is the mapped variable with the file url a publicly accessible file?

Help for using files in Zaps: https://help.zapier.com/hc/en-us/articles/8496288813453

 

Yes, the file uploads no problem and chatgpt can access it. I have tested it. 

@smccarthy956 

Things to try…

  • Copy/paste the returned file URL from Knack in Step 1 into a new/different private browser to see if it starts a direct download of the file. (If not, then it means the file url is not publicly accessible.)
  • Try using a different Model.
  • You can try adding a Delay step to make sure the file finishes uploading before asking questions about it.

 

 

Yes, the URL works if I copy and paste it into a different browser that is not logged in. 

@smccarthy956 

Things to try...

  • Try using a different Model.
  • You can try adding a Delay step to make sure the file finishes uploading before asking questions about it.
  • Try uploading the file directly using the New Files field instead of the previous ChatGPT Upload File Zap action
  • Make sure the question being asked in the Message field is relevant to the file being uploaded.

 

Have tried all of the above with the same results. Some answers are correct and some are just plain wrong. 

@smccarthy956 

Have you tried including additional guidance for the AI?

  • Include context before/after the mapped variable in the Message field
  • Include Assistant Instructions
  • Response Instructions

Make sure to check the field descriptions for guidance.

 

 

Going back to my original point, if I upload the same file to chatgpt’s interface, it answers everything perfectly even if I provide no context. I don’t understand how Zapier is getting different answers when this works perfectly using the standard Chatgpt interface. The data is not that complicated. 

@smccarthy956 

When you are using the ChatGPT web app it can use context across chats.

The way you have the ChatGPT Zap action step configured is running in a silo.

So it’s not a true 1:1 comparison and that may account for some of the variance you are seeing with the results.

So how do you get it to work like chatgpt if I was using the interface?

@smccarthy956 

Try using these fields and check the field descriptions.

 

Terms & Conditions