Improving ChatGPT Automation for Notion Content: Accuracy & Character Limit Challenges

Hi @Shiwani 

Good question.

We’d need to see the OpenAI prompt being used.

Hi @Shiwani! 👋 

When you say you’re “using a tweet as a prompt”, I’m a little confused. Is the trigger for the Zap Twitter or Notion? Or is it that the tweets are being added into Notion by another Zap and you’re building a new Zap that’s triggering from Notion and then wanting to use ChatGPT in a Zap to rewrite it?

If you could send over a screenshot of the Action section of the ChatGPT action in the Zap that would help to allow us to see if there’s any settings or tweaks you could make to the set up of that action to help improve it’s accuracy. Please remove/hide any private information from the screenshot before sharing (like names, email addresses etc.).

With the need to “restrict the content generation”, if you’re wanting to ensure what’s output by ChatGPT is under a certain character limit, perhaps you could add that requirement into the prompt you’re using? 

I look forward to hearing from you on this!

Also, is there any possibillity to get support from zapier team directly apart from asking here in the community? 

Thanks for sending over that helpful screenshot, @Shiwani! 🙂

Ah ok, so the prompt is just the tweet itself. Usually for the Prompt field in a Send Prompt (OpenAI) action I’d expect to see some instructions given for what you’d like OpenAI to do. Updating the Prompt field so that it includes instructions for what you’d like OpenAI to do with that tweet should help to improve it’s accuracy here. 

You’d also want the prompt to include instructions to have only a specific number of characters returned. For example you could try using a prompt like: Rewrite this text in no more than 100 characters: “[Tweet field goes here]” 

You can learn more about writing prompts here: How to write an effective GPT-3 or GPT-4 prompt 
 

Also, is there any possibillity to get support from zapier team directly apart from asking here in the community? 

There’s a number of different support options depending on the Zapier plan you have. You can read more about this here: Zapier Support 

Hope that helps. Please do keep us in the loop on how you get on with this! 

Thanks for giving that a try, and for sharing those helpful screenshots here, @Shiwani. Sorry to hear it didn’t improve the results.

Hmm, it looks as though the text in the ChatGPT_Content field has been cut short, well before it reaches 180 characters. Is that what you mean by it not giving better results?

If so, there was a similar issue of the text produced by ChatGPT appearing truncated, that was discussed here: 

So I’m wondering if that’s what is causing the text to be cut off here as well.

To test that theory, can you try either doing a live test of the Zap or removing the value in the Max Tokens field (as mentioned on that other topic), to see if that allows you to see the full text produced by OpenAI?

Let us know how it goes, keen to ensure you’re all set here! 🙂

Thanks for reaching out here, @BakerAdan! 👋

In order to restrict the output to a specific length, in the prompt itself you’d need to reference the maximum number of characters you want it to return. That way ChatGPT will know how many characters it needs to limit it’s response to.

Improving the accuracy is going to take a fair amount of trial and error. The blog post I shared on How to write an effective GPT-3 or GPT-4 prompt contains a number of tips on prompt writing that can help to improve the accuracy. I’d recommend giving some of those tips a try to see what works best for your use case here. Please keep us updated on how you get on with this!