Steps to build a news-scanning AI agent
This resource is based on She built two Zapier agents. One found her a new client, featuring Maddy Osman of The Blogsmith, published on the AI Lab by ActiveCampaign.
Get the checklist
Before you start
Make sure you:
- Set up a Zapier account: paid plan required for Agents access and multi-step workflows
- Create an Airtable base: free tier works; you’ll need a table with columns for title, source URL, summary, context, date, status, and draft fields
- Connect a search API: Zapier uses a connected search API for live Google News results (not cached content)
- Prepare your brand style guide: document tone, formatting rules, and platform-specific preferences for LinkedIn and X
- List 3–5 niche topics: the industry keywords and subjects the agent should scan for
The workflow
Phase 1: Build agent 1 — find and analyze news
- Create a new Zapier Agent: go to zapier.com/agents; name it something like “[Brand] News Scanner”
- Set a daily morning trigger: configure the agent to start at the same time every morning
- Configure Google News search: connect a search API so the agent pulls live, accurate results
- Define relevance criteria: write natural language instructions telling the agent what makes a story worth sharing
- Add analysis instructions: for each story: summarize key points, explain why the development matters, answer “Why should a [target customer] care?”
- Connect Airtable as storage: map fields: title, source URL, summary, context, date found
- Enable deduplication: the Airtable log acts as a memory bank; the agent checks source URLs to avoid resharing
Phase 2: Build agent 2 — draft social media posts
- Create a second Zapier Agent: name it something like “[Brand] Social Drafter”
- Set the trigger a few minutes after Agent 1: ensures the analysis is complete before drafting begins
- Connect Airtable for retrieval: pull news items tagged for daily posting using the “last modified date” field
- Upload your style guide to Zapier: include social media guidelines, tone, formatting, and platform-specific rules
- Write platform-specific drafting instructions: one post for LinkedIn and one for X, following best practices for each
- Save completed posts to Airtable: record finished drafts in LinkedIn Draft and X Draft columns
- Send to Slack for review: configure a dedicated Slack channel for the human-in-the-loop approval step
Phase 3: Test and refine
- Run agent 1 manually: check Airtable output for relevant stories and accurate summaries
- Run agent 2 manually: check the Slack posts for tone, formatting, and platform fit
- Keep instructions explicit but concise: don’t assume the AI remembers details unless stated clearly
- Limit unnecessary detail: too much information can confuse the AI or cause it to skip steps
- Test after every single change: one small tweak can break a later step in the workflow
- Run the full pipeline for 3–5 days: review every post and refine instructions based on what you see
Quick reference
- Total time: ~1 week for the first version (4–6 hours build + iteration)
- Tools needed: Zapier (paid plan with Agents), Airtable, Slack, search API
- Key output: A two-agent system that scans news, analyzes relevance, drafts platform-specific social posts, and sends them to Slack for approval every morning
Ready for the full story?
Read She built two Zapier agents. One found her a new client, featuring Maddy Osman of the Blogsmith, published on the AI Lab by ActiveCampaign.
Related
More data from the AI Lab.