Why Ranking in Google Doesn’t Guarantee Visibility in AI Overviews or Gemini – Article and Video 65

Why Ranking in Google Doesn’t Guarantee Visibility in AI Overviews or Gemini

Author: Kevin C. Roy · Published: 2026-09-17

Watch the Breakdown

What Changed?

SEO has traditionally operated around a fairly understandable objective:

Rank a page highly enough in Google that users find it.

AI search changes that model.

Researchers from NJIT, Nanyang Technological University and Indiana University analyzed 11,500 queries across:

• Traditional Google Search
• Google AI Overviews
• Gemini 2.5 Flash with Google Search grounding

The research was published in the proceedings of ACM SIGIR 2026.

The study included nine query categories encompassing real-user searches, retail searches, product comparisons, product questions, local searches, debate queries, complex questions and natural-language versus keyword-style queries.

The most important finding for marketers was not simply that rankings changed.

The actual sources being retrieved changed.

The Proof: Google’s Search Products Retrieve Different Sources

The researchers calculated Jaccard similarity between the URL sets retrieved by each system.

Jaccard similarity measures the overlap between two sets relative to all unique items appearing in either set.

A score of 1.0 would represent identical source sets.

These scores were nowhere close.

For Google Search versus AI Overviews, the average was 0.18.

In practical language, there was relatively little overlap between the pages surfaced by traditional Search and those used by AI Overviews.

That challenges one of the most common assumptions about AI search: AI Overviews are simply summarizing Google’s highest-ranking organic pages.

The research suggests something substantially more complicated is happening.

The Commercial Searches Were Especially Interesting

The divergence wasn’t evenly distributed across every query category.

Product and localized query groups produced some of the lowest similarity levels.

For example, the study reported the following Jaccard similarities between traditional Google results and AI Overviews:

This matters to businesses because these aren’t obscure academic queries.

They resemble commercially meaningful searches:

• “best accountant near me”
• “best CRM for a small business”
• “Brand A vs. Brand B”
• “best roofing company in Boston”
• “which software is better for…”

Those are precisely the searches where becoming part of an AI-generated recommendation can influence revenue.

This Is What This Means

There are increasingly at least three distinct visibility problems marketers need to solve.

1. Search Ranking

Can Google understand the page and rank it competitively?

Traditional SEO remains important here.

That includes:

• crawlability
• technical SEO
• relevance
• authority
• internal linking
• content quality
• links
• site architecture

2. AI Retrieval

When an AI system needs evidence to answer a question, does it retrieve your information?

This is not necessarily identical to traditional ranking.

Pages increasingly need to make individual facts, answers, entities and relationships easy to identify and extract.

3. AI Citation

Even if information is retrieved, does the system choose your website as supporting evidence for the final response?

That creates another set of questions:

• Is the claim clearly supported?
• Is the source authoritative?
• Is the information easy to extract?
• Is the author or company identifiable?
• Do independent sources corroborate the information?
• Is the business/entity consistently represented across the web?
• Does the page provide something worth citing?

This is why GreenBanana treats SEO visibility and AI citation visibility as related but separate measurements.

The Rank → Retrieve → Cite Framework

Businesses and agencies can use a simple three-stage framework.

Stage 1: RANK

Build conventional search eligibility.

Make sure the site has:

• strong technical fundamentals
• dedicated pages for important topics
• logical internal linking
• authoritative content
• links and mentions
• solid page experience
• clear service and location relevance

Do not abandon traditional SEO.

AI visibility is not evidence that SEO suddenly stopped mattering.

Stage 2: RETRIEVE

Create information AI systems can easily identify.

That means moving beyond one generic “services” page.

Develop pages answering the actual questions surrounding the service, product or entity.

Examples include:

• comparison pages
• service-specific pages
• location pages
• pricing explanations
• qualification criteria
• FAQ content
• implementation guides
• use cases
• definitions
• original research or data

The objective is to create clear retrieval targets.

Stage 3: CITE

Give the system reasons to select your information as evidence.

Possible supporting signals include:

• clear expert authorship
• factual specificity
• original research or data
• references and sourcing
• schema where appropriate
• consistent organization and entity information
• authoritative external mentions
• strong topical coverage
• third-party corroboration

None of these should be marketed as a guaranteed “AI ranking factor.”

Generative search systems are complex, and many individual signals remain difficult to isolate.

Instead, think of them as components of a stronger citation environment.

What Most Companies Are Getting Wrong

The mistake is measuring AI visibility as an extension of rank tracking.

A company sees:

Google position: #3

and assumes:

AI visibility: excellent

That conclusion is no longer safe.

The correct dashboard needs separate measurements for:

1. Traditional organic ranking
2. AI Overview inclusion
3. AI citation/source visibility
4. Gemini visibility
5. ChatGPT visibility
6. Other commercially relevant answer engines

The study itself reinforces this point because even AI Overviews and Gemini — two Google systems — had an average source similarity of only 0.11.

Visibility in one AI system does not guarantee visibility in another.

Important Limitation

The study’s primary dataset was collected on December 7–8, 2025.

Google continuously changes Search, AI Overviews and Gemini.

Therefore, the exact similarity percentages should not be interpreted as permanent characteristics of Google’s systems.

The stronger conclusion is directional:

Generative search retrieval and traditional search retrieval were substantially different across a large controlled benchmark.

That finding is far more important than any individual percentage.

Key Takeaway

Traditional SEO is not disappearing.

Its role is expanding.

Businesses now need to solve three connected problems:

Rank. Retrieve. Cite.

Winning Google rankings creates visibility and helps establish relevance and authority.

But AI search introduces another layer of retrieval and evidence selection.

The companies that recognize that distinction early will stop asking only:

“Where do we rank?”

and start asking:

“When AI answers our customer’s question, why would it choose us as the source?”