A guide to building a news-scanning AI agent

Day 1: Prepare your inputs

Define your audience and niches 

Write a one-paragraph description of your target reader. Osman targets CMOs interested in content marketing and AI. Then list 3–5 topic keywords the agent should scan for — these become the search queries for Agent 1.

Document your style guide 

Create a document covering: brand voice, social media tone, formatting rules for LinkedIn, formatting rules for X, phrases or styles to use, and phrases to avoid. This gets uploaded directly to Zapier.

Set up Airtable 

Create a new base with columns for: title, source URL, summary, context (“Why should [audience] care?”), date found, last modified (auto), status (New / For Posting / Posted / Skipped), LinkedIn draft, and X draft.

Day 2–3: Build Agent 1 — the news scanner

Open Zapier Agents and create a new agent. Write instructions in natural language — ​“you can interact with it in natural language like you would interact with ChatGPT,” Osman says. Configure the agent to:

1. Run every morning at a scheduled time
2. Search Google News using a connected search API for your niche topics
3. Filter stories based on your relevance criteria
4. For each relevant story: summarize key points, explain why it matters, answer ​“Why should a [target role] care?”
5. Log results to Airtable — check source URLs first to avoid duplicates

Test it: Run manually. Check Airtable. Are the stories relevant? Are the summaries useful? Refine the criteria until they are.

Day 3–4: Build Agent 2 — the social drafter

Create a second Zapier Agent. Set it to trigger a few minutes after Agent 1 so the analysis is complete. Configure it to:

1. Pull fresh news from Airtable (filter by last modified date)
2. Apply your uploaded style guide
3. Draft one LinkedIn post and one X post per story
4. Save drafts back to Airtable
5. Send posts to a dedicated Slack channel for review

Test it: Run manually. Check Slack. Do the posts match your voice? Do they follow platform best practices? Adjust instructions.

Day 5–7: Refine and go live

Osman’s lessons from building her agent:

• 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 change. One small tweak can break a later step.
• Use existing integrations first. Zapier’s built-in connections to Airtable, Slack, and APIs reduce complexity.
• Think long term. Airtable tracks content history so the agent improves over time.

Run both agents for 3–5 days. Review every post. Make small adjustments daily until output consistently matches your expectations.

What ​“done” looks like

Every morning you wake up to a Slack channel with drafted social posts: researched, analyzed, and written in your voice. You spend 5–10 minutes reviewing instead of an hour scanning news and writing from scratch. ​“I’m capitalizing on timely things without spending my time on them, while still having something that uses my logic,” Osman says.