metehan.ai Research & SOPs

Appearance

AI Overview Optimization ​

Fresh

This content reflects Metehan Alp's AI Overview optimization approach as of 2025-2026.

AI Overviews (formerly Search Generative Experience) are Google's AI-generated answer panels that appear at the top of search results. Optimizing for them requires understanding how Google selects and processes source content.

Google Cloud Vertex AI Search Approach ​

Metehan's methodology uses Google Cloud Vertex AI Search to reverse-engineer how Google's AI Overviews select source content. By creating isolated search indexes and testing content against them, you can identify what structural and semantic factors drive inclusion.

Vertex AI Search uses the same underlying technology as Google's search AI features. By building a controlled index, you can:

  • Test content structure changes without waiting for Google to re-crawl
  • Isolate variables (structure, keywords, schema) to identify what moves the needle
  • Score content against the same semantic models Google uses
  • Iterate rapidly on content optimization

Isolated Index Creation Process ​

Step 1: Create a Vertex AI Search Data Store ​

Set up a data store in Google Cloud Console with your target content. This becomes your controlled test environment.

Step 2: Ingest Your Content ​

Upload your pages as structured documents. Include:

  • Page title
  • Full body content
  • URL
  • Metadata (author, date, topic)
  • Schema markup data

Step 3: Ingest Competitor Content ​

Add competitor pages for the same queries to the index. This lets you compare your content's relevance score against theirs.

Step 4: Run Test Queries ​

Query the index with your target search terms. Vertex AI Search returns relevance scores and source selection data.

Step 5: Analyze and Iterate ​

Compare your content's scores against competitors. Modify content structure and re-ingest to test improvements.

Content Structure Recommendations ​

AI Overviews favor content with specific structural patterns:

Heading Hierarchy ​

  • H1: One per page, matches the primary query intent
  • H2s: Each addresses a distinct sub-topic or sub-query
  • H3s: Break complex H2 sections into scannable sub-sections
  • Headings should be questions or declarative statements that match search patterns

Section Length ​

  • Target approximately 500 tokens per section (roughly 375 words)
  • This aligns with the chunk size Google AI Mode uses for processing
  • Sections shorter than 200 tokens may lack sufficient context
  • Sections longer than 800 tokens may get partially processed

Answer-First Structure ​

  • Lead each section with a direct answer in the first 1-2 sentences
  • Follow with supporting evidence, examples, and detail
  • This mirrors how AI Overviews extract source passages

Lists and Tables ​

  • Structured data (ordered lists, comparison tables, data tables) gets preferentially selected for AI Overview inclusion
  • Use tables for comparisons, specifications, and multi-variable data
  • Use ordered lists for processes, rankings, and sequential steps

Keyword Targeting ​

Long-Tail Informational Queries ​

AI Overviews appear most frequently for:

  • "How to" queries
  • "What is" queries
  • Comparison queries ("X vs Y")
  • "Best" queries with qualifiers
  • Multi-part questions

Targeting Strategy ​

  1. Identify queries where AI Overviews currently appear (use Semrush or Ahrefs AI Overview tracking)
  2. Prioritize queries where no strong source currently dominates
  3. Focus on queries with 3+ sub-questions (more content to include)
  4. Target queries where your existing content ranks in the top 20 (easier to push into AI Overview selection)

Cosine Similarity for Semantic Alignment ​

Cosine similarity measures how semantically close your content is to the ideal answer for a query. Higher cosine similarity means your content better matches what the AI system expects to find.

How to Use It ​

  1. Embed your content -- Convert your page content into a vector embedding using the same model family Google uses (Gecko, or similar)
  2. Embed the query -- Convert the target search query into a vector
  3. Calculate similarity -- Compute the cosine similarity between the two vectors
  4. Benchmark -- Compare your score against competitors' scores for the same query
  5. Optimize -- Adjust your content's language to increase semantic alignment

Practical Application ​

Metehan's Cosine Similarity tool (available as a deployed app) automates this process. Input your content URL and target query to get a semantic alignment score.

Semantic Alignment Is Not Keyword Stuffing

Increasing cosine similarity means aligning your content's meaning with the query's intent. This is about covering the right topics, using the right entities, and addressing the right sub-questions. Adding more keyword mentions without adding topical depth will not improve your score.

Key Strategies ​

1. Answer the Question Immediately ​

The first 100 words of any page targeting AI Overviews must directly answer the primary query. Do not use introductions, preambles, or throat-clearing.

2. Cover the Topic Comprehensively ​

AI Overviews pull from sources that cover the full scope of a query. Partial coverage loses to comprehensive coverage.

3. Use Structured Formatting ​

Tables, lists, and clear heading hierarchies increase the probability of content selection. Unstructured prose walls are harder for AI systems to extract from.

4. Build Supporting Content ​

A single page rarely wins AI Overviews for competitive queries. Build a topic cluster with supporting pages that address sub-queries, and link them together.

5. Update Frequently ​

AI Overviews favor fresh content. Pages with recent modification dates outperform stale content for the same query.

6. Leverage Schema Markup ​

Article, FAQ, and HowTo schema provide structured signals that help AI systems understand your content's purpose and structure.

Further Reading ​