AI Overviews: Strong Evidence for Snippet Usage Instead of Full Text

For months, there has been an ongoing debate about whether Google Gemini in AI Overviews and AI Mode has access to the full text of web pages or only evaluates short excerpts, so-called snippets. This is exactly what I have suspected in earlier posts – and now this suspicion is confirmed.

A recent statement from Logan Kilpatrick, Lead Product Manager at Google for the Gemini API, provides clear evidence that snippets are used by default.

1. The Statement on X (Twitter)

On August 18, 2025, Kilpatrick published the decisive clue on X:

search grounding only accesses snippets of the page, it cannot see the full page without URL context.

The statement confirms the earlier assumption and is contextualized by a
follow-up tweet
that explains the difference compared to the new URL Context feature.

This makes it clear:

• Search Grounding processes only excerpts of a page and does not see the full text.
• The new URL Context feature allows full page content to be included – but only if the URL is explicitly provided.

This statement describes the design of Grounding within the Gemini API and is therefore strong evidence for the operating principle that also underlies AI Overviews.

2. What This Means for AI Overviews and AI Mode

The official documentation for AI Overviews and AI Mode repeatedly refers to “grounding via search results.” This strongly suggests that both systems also work with snippet-based contexts.

This matches practical observations:

• Answers often appear superficial.
• Details are missing even though they exist in the full text.
• Source citations almost always refer to small excerpts and rarely to the overall context.

In contrast, the Gemini API demonstrates with URL Context what true full-text access looks like – but this is not the default mode of AI Overviews.

3. Why This Matters

For SEO and AI visibility, the consequences are clear:

• Content must be structured so that it remains understandable and useful in small excerpts.
• Structure is more important than length. Long continuous text is of little use if it is not broken into snippet-friendly sections.
• AI Overviews behave more like an indexing system that reproduces excerpts rather than like a browser that fully processes entire texts.

4. Broader Context

There is also strong evidence that ChatGPT, when accessing the web, uses snippet- or chunk-based retrieval rather than the full text of a page. This assessment is based on observed answer patterns and publicly known retrieval mechanisms.

This confirms a pattern that applies across multiple systems: LLMs do not use full pages for search grounding but instead rely on fragments from search indexes.

5. Conclusion

As of now, there is strong evidence that AI Overviews and AI Mode access snippets by default and not full texts.

For publishers, this means:

• Content should be snippet-friendly.
• Clarity, conciseness, and structure are more important than ever.
• Those who understand the snippet logic can systematically optimize visibility in AI Overviews.

Sources & References

1. Logan Kilpatrick on X (August 18, 2025):
   Tweet
2. Google AI Developer Docs:
   Grounding with Google Search
3. Google AI Developer Docs:
   URL Context
4. Vertex AI Docs (additional source):
   
   Grounding with Google Search (Vertex AI)