Practical guides for keyword research, competitor analysis, local AI tuning, plugin development, and reporting — from teams shipping real SEO work every day.
1. Keyword research workflow
Start with 5-10 seed keywords, expand to 200+ via AI Agent suggestions.
Filter by difficulty (under 30) + volume (over 500) for quick wins.
Group by intent: informational, commercial, transactional, navigational.
Connect GSC for real performance data layered on top of estimates.
2. Competitor analysis rhythm
Pick 3-5 competitors per project — not 50.
Run a full analyze_competitor monthly, not daily.
Focus on the keyword gap — what they rank for that you don't.
Track their content velocity, not just their rankings.
3. Local AI model selection
4GB RAM: Gemma 2B or Qwen3 1.7B. Fast, good enough for tagging.
8GB RAM: Qwen3 4B or Mistral 7B (quantized). The sweet spot.
16GB+ RAM: Qwen3 8B or Qwen2.5 7B. Best Chinese SEO content quality.
20GB+ + GPU: Phi-4 14B. Reasoning-heavy tasks.
4. AI Agent patterns
One task per prompt: "Analyze this site" beats "do everything".
Watch the live status bar — cancel early if a tool goes sideways.
Pin iterative rounds: 1-2 for quick questions, 3 for full audits.
Save successful prompts as templates for the team.
5. Python plugin development
Read the official example plugins in docs/examples/ first.
Sandbox blocks network/files/DB by default — request permissions in plugin.yaml.
Always handle SIGALRM for the 60s timeout.
Version your plugins in your own git, never trust untrusted scripts.