How to build sentiment analysis in Bubble

The simplest way to build sentiment analysis in Bubble is to send the user’s text to an external AI API (like OpenAI, Google Cloud Natural Language, or HuggingFace) using the API Connector, then store the returned sentiment score/label in your Bubble database. Bubble itself has no built‑in sentiment analysis, so you always rely on an external API.

Overall Flow

You let the user submit text → trigger a backend workflow → send that text to an API → get back the sentiment → save it to a field like Sentiment (text) or Sentiment Score (number).

• Front-end: User types something in an Input → Clicks a Button.
• Backend workflow: Receives the text, calls the AI API, stores the response.
• Database: Your data type (for example “Message”) contains a field for the sentiment.

Setting up the API Connector

Example using OpenAI’s sentiment-style call via their /chat/completions endpoint. This works reliably in Bubble today.

{
  "model": "gpt-4o-mini",
  "messages": [
    {
      "role": "system",
      "content": "Return only one word: Positive, Negative, or Neutral."
    },
    {
      "role": "user",
      "content": "<TEXT_TO_ANALYZE>"
    }
  ]
}

In the API Connector:

• Method: POST
• URL: https://api.openai.com/v1/chat/completions
• Header: Authorization: Bearer YOUR_API_KEY
• Header: Content-Type: application/json
• Body type: JSON
• Body: the JSON above, but replace <TEXT_TO_ANALYZE> with a dynamic key you expose.

Saving the Sentiment in Bubble

In your backend workflow step “Call OpenAI”, Bubble will return something like response.choices:first.message.content. You save this to the Thing you’re analyzing (e.g., Message’s Sentiment field).

• Action: “Make changes to a thing”
• Thing: The message being processed
• Sentiment = Result of API call’s response.choices:first.message.content

Practical Tips

• Use backend workflows so no API keys are exposed on the page.
• Store both raw sentiment and the full API response if you want to debug later.
• If you want a numeric score, adjust the system prompt to “Return a number between -1 (negative) and 1 (positive).”