Open AI response

[sportinvest.org]

Your Workflow URL

https://connect.pabbly.com/workflow/mapping/IjU3NjYwNTY1MDYzMTA0MzI1MjZhNTUzNDUxMzAi_pc

Your Task History ID

IjU3NjUwNTZiMDYzMTA0Mzc1MjZhNTUzZDUxMzQ1MTY0NTQzODBmMzYi_pc

Consent to Access & Modify

I authorize Pabbly Support to log in to my account and make changes to the specified workflow for troubleshooting.

Hello,

I’m running into an inconsistency with the OpenAI integration in Pabbly Connect.

Sometimes, when OpenAI returns a JSON response like this:


{
"title_short": "De Bruyne torna a brillare: 3 gol e fantamedia 8",
"title_social": "De Bruyne affila le armi",
"caption": "Nella serata in cui il Napoli cade contro il Milan, Kevin De Bruyne si distingue...",
"wikimedia_keyword": "Kevin De Bruyne Napoli",
"hashtags in base al post": ["#sportinvest", "#calcio", "#fantacalcio"]
}

Pabbly automatically parses the JSON and exposes each property (title_short, caption, etc.) as individual fields in the next steps — which is perfect.

However, other times, for the same type of prompt, Pabbly doesn’t parse it and instead returns the entire JSON as a single string field (the whole object under choices0messagecontent).

The issue is that I cannot predict when Pabbly will parse it correctly, and when it won’t. This makes it difficult to consistently use the OpenAI output in later steps (for example, posting captions on Facebook/Instagram).

My questions are:

1. Is this a known limitation or a bug in the OpenAI app inside Pabbly Connect?
2. Is there a way to force Pabbly to always treat the response as JSON (not plain text), so fields are exposed consistently?
3. If not, what’s the recommended workaround inside Pabbly (without relying only on custom Code steps)?

Thanks in advance for your help!

 

[anushkasahu]

Hello @sportinvest.org,

This is not a bug. The behavior occurs because certain double quotation marks in the response break the JSON format, causing the output to be treated as a single string field instead of being parsed further.

To avoid this, we recommend adding a Text Formatter step before proceeding, where you can replace problematic double quotation marks with either single quotation marks or a placeholder (e.g., {{blank}}). This ensures the response remains valid JSON and can be parsed consistently across steps.