Skip to content

AI Search Optimization (GEO)

From RAG to RichesMark Williams-CookFresh

Overview

This SOP covers the process for optimizing content visibility in AI search platforms (ChatGPT, AI Mode, Perplexity) based on Mark Williams-Cook's "From RAG to Riches" presentation.

Key Insight: GEO is mainly SEO - what we do is similar, but HOW we do it has evolved.

Prerequisites

  • [ ] Traditional keyword list
  • [ ] ChatGPT API access (for prompt enrichment)
  • [ ] Also Asked account (API preferred)
  • [ ] Access to Gemini Grounding API or Dan Petravic's models
  • [ ] Google Search Console access

The GEO Process

Procedure

Step 1: Start with Traditional Keywords

Begin with your existing keyword list. These become the foundation for AI search optimization.

Important

Traditional keyword lists alone are useless for AI search - people have full conversations, not single queries.

Step 2: Apply Persona Layer

Use ChatGPT API to enrich keywords with persona data:

Prompt template:
"If I was [PERSONA DESCRIPTION] and I was trying to find [KEYWORD], what might I ask you?"

Example personas:

  • Middle-aged vegan starting running
  • Small business owner looking for CRM
  • First-time homebuyer in [city]

This reflects how people frame conversations, drawing from training data (forums, subreddits, etc.)

Step 3: Expand with Also Asked Data

Use Also Asked to find "closest intent proximity" - the follow-up questions in a conversation.

Why Also Asked?

  • API available for scale
  • Captures real user question patterns
  • Shows where conversations naturally go

Step 4: Filter for Grounded Queries

Not all queries trigger web searches. LLMs score each query for "grounding probability."

Grounding thresholds (ChatGPT free): ~0.65

How to check:

  1. Browser DevTools → Network tab
  2. Find search_prob in conversation response
  3. Score above threshold = grounded

Or use tools:

In-Model vs Grounded

  • In-model answers: Can only influence through training data (slow)
  • Grounded answers: Can influence through web ranking (fast, 3-4 days)

Step 5: Extract Background Web Searches

Find what searches LLMs actually perform:

ChatGPT method:

  1. DevTools → Network tab
  2. Look for search_model_queries
  3. Lists actual web searches being performed

Gemini method:

  • Use Grounding API directly
  • Pass query, get list of triggered searches

Example: Query: "Who won Euro 2024?" Background searches: euro 2024 winner, euro 2024 final score, euro 2024 champion team

Step 6: Traditional SEO for Those Queries

Now you have a list of actual search queries to rank for. Apply traditional SEO:

  • Content creation/optimization
  • On-page SEO
  • Link building

Step 7: Multi-Site Visibility

Key Shift

You don't need to rank your site - you need to be MENTIONED across top 10-50 sites.

Tactics:

  • Get reviews on review sites
  • Appear in roundups and listicles
  • Digital PR for brand mentions
  • Forum participation (authentically)

Verification Checklist

  • [ ] Personas applied to keyword list
  • [ ] Also Asked data collected
  • [ ] Grounding probability checked for all queries
  • [ ] Background searches extracted
  • [ ] Content optimized for grounded queries
  • [ ] Brand mentioned across multiple authoritative sites

Tools Reference

ToolPurposeLink
queryfan.comFull GEO workflow automationComing soon
Also AskedPAA data collectionalsoasked.com
Dan Petravic's modelsGrounding predictionHugging Face
Gemini APIGrounding API accessGoogle AI Studio

Key Metrics

  • Citation frequency in AI answers
  • Grounding rate of target queries
  • Brand mentions across top-ranking sites
  • AI referral traffic (via UTM or logs)

Troubleshooting

Query never gets grounded

The query may be "solved information" - try adding recency signals or focusing on topics that deserve freshness.

Can't find search_prob in DevTools

Make sure you're looking at the correct API response. It may be nested in the conversation object.

Content not appearing in AI answers

Check if you're in the top 10-50 for the background search queries. Consider building mentions on authoritative sites that are appearing.

See Also

Knowledge extracted from Search 'n Stuff Conference talks