Directing the Agent
The agent runs your SEO on its own, but a sharp question gets you a sharper answer. Learn how to direct the agent in chat — specific metrics, date ranges, action requests, and chained questions on your real data.
Why Prompts Matter
The agent runs your SEO continuously on its own — but when you step in to chat, the prompt you write shapes the answer you get back. A vague prompt gets a generic answer. A specific, well-structured SEO prompt gets data-backed insights with actionable recommendations on your real Search Console numbers. This page shows the prompting pattern that works, with copy-paste examples you can run right now.
Because the agent works from real tools — GSC queries, site crawl data, content generation — your prompts steer which tools it reaches for and how it structures its analysis of your data.
Anatomy of a Good SEO Prompt
The best prompts for directing the agent include four elements:
- Specific metric — what data do you want? (clicks, impressions, CTR, position)
- Time range — when? (last 28 days, this month vs last month)
- Filter or threshold — what subset? (position 5-15, 100+ impressions, mobile only)
- Action request — what should the agent do with the data? (suggest, generate, compare, explain)
"How are my keywords doing?"
"Show my keywords by clicks for the last 28 days (only 100+ impressions). Which ones have CTR below 2%? Suggest title improvements for those."
Prompt Patterns That Work
Analysis Prompts
When you want the agent to find insights:
“Compare my top 30 keywords this month vs last month. Highlight any that dropped more than 20% in clicks and explain possible reasons.”
Period comparison with threshold filter and analysis request.
“Break down my organic traffic by device type for the last 3 months. Is mobile growing or declining relative to desktop?”
Dimension-specific analysis with trend question.
Action Prompts
When you want the agent to produce something:
“Find my 3 best striking-distance keywords and generate a content brief for each one”
Combines data query with content generation — the agent chains GSC query → brief_generator.
“Write a 2000-word article targeting 'technical SEO checklist' based on my site's existing content and GSC data”
Direct content generation grounded in your data.
Chaining Prompts
The agent remembers your conversation (up to 20 messages). Use follow-ups to drill deeper:
- “Show me my top declining pages”
- “For the #1 declining page, what keywords is it losing?”
- “Generate a content refresh brief for that page”
- “Write the updated article using my writing style”
Chaining is where the agent shines. Each follow-up question builds on the previous data, going from raw analysis to published content in a single conversation thread.
Common Mistakes
| Mistake | Why It's Bad | Better Approach |
|---|---|---|
| Too vague — "Do SEO for me" | Agent can't choose tools or scope | Start with a specific question about one aspect |
| No date range | Agent defaults may not match your intent | Always specify: "last 28 days", "this month vs last" |
| Asking for "everything" | Response gets too broad and shallow | Ask for top 20 with specific sort order |
| Not following up | Missing the deeper insights | Drill into specific items from the first response |
| Ignoring tool results | You miss what data the advice is based on | Read the tool call results — they're shown for a reason |
Prompt Library
Copy and paste these directly into the agent chat:
“Give me an SEO health check: top 10 keywords, top 5 declining pages, and your #1 recommendation”
“Find keywords where I rank 4-10 with 500+ impressions. These are my quick wins — prioritize by impact.”
“Compare branded vs non-branded traffic for the last 3 months. Am I too dependent on brand terms?”
“Find pages with high impressions but CTR below 2%. Suggest new titles and meta descriptions for each.”
“Check my site for cannibalization — are any pages competing for the same keywords?”
Try these prompts
© 2026 Agentic SEO. All rights reserved.