How NTUC automated social listening with Zapier, OpenAI, and ExportComments

TABLE OF CONTENTS

The Challenge: A time consuming key process to understand ground sentiment

The Impact: (Almost) Instant data-to-insight

The Solution: How we did it

1. Submitting URLs via a Lightweight Interface

2. Exporting Comments Automatically

3) Analysing and Reporting

The Next Build: Levelling up the process

The work.flowers Approach: Why it works

Learn More

How NTUC automated social listening with Zapier, OpenAI, and ExportComments

How NTUC cut hours of manual work by automating social media comment exports, AI analysis, and reporting

Dennis Chiuten

Oct 22, 2025

The Challenge: A time consuming key process to understand ground sentiment

The National Trades Union Congress (NTUC) is Singapore’s national confederation of trade unions and a key social partner in shaping labour and workforce policies. Through its network of unions, cooperatives, and social enterprises, NTUC works to improve the lives of workers and strengthen the country’s economic resilience.

When NTUC wanted to understand how Singaporeans feel about hot‑button issues, they were looking to build a seamless, AI-enabled approach.

They used monitoring tools to find public social posts about relevant topics with high engagement, exported the comments, and used a well‑crafted ChatGPT prompt to classify sentiment and themes.

They were already using AI for what it does best: processing, structuring, and analysing large amounts of unstructured text. But the workflow still required time-consuming manual steps:

Export an Excel file for each social media post using ExportComments

Upload the file to ChatGPT, paste in the analysis prompt, and run

Copy and paste the output somewhere useful for sharing

This worked well enough but simply didn’t scale.

The Impact: (Almost) Instant data-to-insight

Recognising the opportunity to streamline further, NTUC partnered with us to automate the entire process.

They outlined three clear requirements:

Remove the manual exports and uploads

Produce an output that’s easy to read and share with no copy-and-paste

Keep NTUC’s existing analysis logic and prompt, which was already providing high-quality results

With that, we were ready to get to work!

By building an end-to-end automation pipeline around NTUC’s core use of AI, we would be able to save the team hours of manual exporting and copy‑paste and replace that effort with a consistent, repeatable process.

The new workflow enables faster time‑to‑insight, including batch analysis across multiple posts, and delivers the final output in minutes.

The Solution: How we did it

We redesigned the workflow around Zapier so the NTUC team could stay focused on insights rather than file handling.

In doing so, we took advantage of a recent update to Zapier’s pricing model that made Zapier Tables and Zapier Interfaces free with any paid plan - perfect for building end-to-end automation.

Zapier Tables is a built‑in data store for your automations. You can use it to capture submissions, track state across multi‑step workflows, and join data from multiple Zaps without reaching for an external database. It’s designed for reliability at automation scale, with schemas, permissions, and native Zap actions.

Zapier Interfaces lets you build simple web front ends—forms, dashboards, and internal tools—that sit on top of your Zaps and Tables. Teams can submit data, trigger flows, and review outputs from a clean UI without touching the Zap editor.

1. Submitting URLs via a Lightweight Interface

The workflow kicks off when a member of the NTUC team submit the URL for a social media post they want to analyse via a simple form in a Zapier Interfaces web app. Once they click Submit, the automation is triggered.

2. Exporting Comments Automatically

A Zap creates an export job for each URL using ExportComments’ native Zapier integration. Because a single post can have hundreds or thousands of comments, one Zap starts the export while another “listens” for job completion. When ready, it stores the result in a Zapier Table.

Before sending the data to GPT-5 for analysis, a quick JavaScript Code Step cleans and flattens the comment structure, removing unnecessary fields and un-nesting replies.

This blend of low-code automation, lightweight code, and LLM-based text analysis unlocks a higher level of automation where logic, data handling, and AI reasoning work together.

A bit of code to make the output more LLM-friendly

3) Analysing and Reporting

The processed comments are passed to an OpenAI step, using NTUC’s existing analysis prompt. GPT-5 summarises comments, groups them by theme, and calculates the proportion of each sentiment category.

Finally, the resulting report is stored in Google Docs, accessible directly through the Interfaces app. The whole process takes just a few minutes and no more than two clicks.

The Next Build: Levelling up the process

Once the initial automation went live, NTUC wanted to take it a step further: analysing multiple posts in one submission. With a few tweaks, we extended the design to support batch processing for 3–5 URLs at a time:

The form accepts multiple URLs at once

Zapier Tables tracks each submission and its related URLs

The automation runs exports in parallel and triggers analysis only after all are complete

GPT-5 produces a combined report covering cross-post themes and sentiment

Batch analysis further reduced processing time per post and helped the team compare themes across related posts.

This experience mirrors a trend we often see with clients early in their automation journey: once you begin to see what’s possible, it becomes second nature to imagine even more ways to streamline your work and save your team time and effort.

The final multi-post workflow in all its glory

The work.flowers Approach: Why it works

When designing a solution for NTUC, we had three key principles to our approach:

Meet teams where they are: We kept NTUC’s proven prompt intact and automated around it.

Design for human review: Google Docs remain a familiar, effective format for collaboration.

Build from small, composable parts: Interfaces for input, Tables for state, Zaps for orchestration; fast to build, easy to extend later.

These principles are at the core of how work.flowers helps teams to operationalise AI: automating the movement of data while keeping human analysis and experience at the centre.