Hey team,
I have a GPT node. I’m wanting it to return a json object and I have that ticked:
But the output of the node isn’t itemising the different json keys like I’d expect it to
Try to add JS node after OpenAI node and ask AI Copilot to parse JSON string, like this: “parse JSON from response of previous node”.
Then click on “Replace button” and “Save & Run”
Yeah and that’s basically what I’ve done. But then I don’t get what the point of the response format checkbox is in that case:
It’s not like it’s even injecting anything into the query asking for a json formatted response because in the description it says I still need to request that in the prompt. So I can’t tell what difference that response format option does?
This is less a Latenode question than an OpenAI issue. Check out https://platform.openai.com/docs/guides/structured-outputs#json-mode
Ahh I see.
Would be handy though if that box was ticked then Latenode assumed the response was going to be JSON and parsed it in the node. Can’t imagine any instance where you’d ask it for a JSON response but not want to parse it unless I guess you were wanting to just inject JSON into a table or something but then you could take it from a higher tree level.
Hi @dhtbrowne !
To ensure that GPT always returns valid and readable JSON, I recommend using a combination of GPT and JSON parse.
Uploading: brave_oR2wkZvdjt.png…
Additionally, a crucial point is that, besides asking GPT to return data in JSON format, you should provide it with an example of the exact JSON structure you expect.
Uploading: brave_oR2wkZvdjt.png(1)…
This approach works flawlessly, and I use it in numerous automation scenarios.
Yes, our developers did a great job! This has become much more convenient
