Most marketing teams treat AI visibility as a single question: is my brand visible in ChatGPT? But when the same prompts are tested across ChatGPT, Gemini, and Perplexity, each model can rely on a very different set of sources, page types, and recommendation signals.

This research looks at how those differences shape AI visibility, citations, prominence, and first-party attribution.

12,678 AI responses analyzed

1,368 Prompts tested

36 Brands tracked

20,005 Unique domains cited

HOW WE DID THIS

Methodology

The study combines two waves of Clamint's ongoing GEO research across ChatGPT, Gemini, and Perplexity.

The latest measurement includes 36 brands, 1,368 unique prompts, 12,678 AI responses, and 20,005 unique cited domains.

The deeper source-category, page-type, and intent analysis uses 78,964 individual citations from the first 27-brand cohort. Sources were grouped by domain type, page type, and prompt intent to understand where AI systems retrieve information and how those patterns change across platforms.

FINDING 1

Source overlap between models is almost zero

56% of ChatGPT and Gemini answer pairs generated for the same prompt share no common source. Across the dataset, average overlap between model source lists remains between roughly 4% and 12%.

Average source overlap between model pairs, same prompt

Gemini ×
Perplexity

11.9%

Gemini × ChatGPT

5.2%

ChatGPT ×
Perplexity

5.2%

0% 4% 8% 12% 14%

Overlap measured as Jaccard similarity: the size of the intersection divided by the size of the union of each model's cited domains.

Gemini and Perplexity show the highest overlap at 11.9%. ChatGPT sits much further apart, sharing only around 5.2% of its source set with either platform.

FINDING 2

ChatGPT reads narrow. Gemini and Perplexity read wide.

ChatGPT cites an average of 4.25 unique domains per response. Gemini reaches 12.89 and Perplexity 14.09, meaning the latter two models draw from a much wider source ecosystem.

Average unique domains cited per response, by model

4.25

ChatGPT

12.89

Gemini

14.09

Perplexity

ChatGPT tends to concentrate on a smaller group of sources. Gemini and Perplexity distribute their citations across a much broader collection of publishers, communities, marketplaces, and independent content.

FINDING 3

Community content is now a distribution channel

Community sources play a very different role across AI platforms. Reddit, YouTube, and forums appear far more frequently in Gemini and Perplexity responses than in ChatGPT.

Share of responses citing at least one community source

ChatGPT

2.5%

Gemini

44%

Perplexity

52%

Community sources include Reddit, YouTube, and forum content.

For decision-stage and comparison-oriented queries, the role of community content becomes even stronger.

FINDING 4

Being visible is not the same as being recommended

Visibility tells us how often a brand appears somewhere in an answer. Prominence tells us whether that brand appears early enough to have a meaningful chance of influencing the user.

Share of mentions appearing in the first 25% of the answer

ChatGPT

52.29%

Gemini

44.85%

Perplexity

62.08%

This is why a single visibility percentage is not enough. A brand may appear often but remain buried deep in answers, while another appears less frequently but receives much stronger placement.

FINDING 5

Visibility is not stable from one day to the next

13% to 15% Change observed between consecutive visibility measurements across repeated scans.

Even when the same prompts and source environment are used, AI systems do not always surface the same brands. A single measurement therefore represents a snapshot rather than a permanent position.

FINDING 6

Getting into the source set pays off, a lot

80% to 95% Observed recommendation range once a brand's own website enters the model's cited source set for a relevant prompt.

Source inclusion was one of the strongest relationships observed in the analysis. Technical accessibility and retrievable, question-focused content therefore become foundational parts of AI visibility.

DEEP DIVE

Where the citations actually come from

Citations are distributed across a broad source ecosystem rather than being concentrated on a small group of dominant domains.

Share of citations by source category

Blogs & long-tail

51.3%

Marketplace & e-commerce

18.2%

Brand-owned sites

2.7%

DEEP DIVE

Deep, specific pages get cited. Homepages do not.

AI platforms disproportionately cite pages that answer a specific question or provide detailed information.

Share of cited URLs by page type

Deep content

61.6%

Blog & guide

16.5%

Homepage

9.5%

Buying intent changes the mix, a little

Commercial prompts make greater use of marketplaces and comparison sources than informational prompts.

Source usage by prompt intent

Source type Informational Commercial

Marketplace & e-commerce 14.5% 20.8%

Review & comparison 1.5% 4.0%

Brand's own site 20.4% 23.7%

DEEP DIVE

First-party visibility depends on the brand, not the industry

First-party attribution varies dramatically between brands operating in similar categories.

Lowest 0%

Average 22.3%

Highest 65.8%

Strong first-party performers tend to have deeper libraries of useful content, technically accessible pages, and material that directly answers relevant user questions.

WHAT TO DO ABOUT IT

Five ways to act on this data

Build a model-specific content strategy

Treat ChatGPT, Gemini, and Perplexity as separate discovery surfaces instead of assuming one GEO strategy will affect all three equally.

Publish deep, question-first content

Build dedicated pages around the questions, comparisons, use cases, and decisions users actually ask AI systems about.

Manage marketplace and comparison-site presence

Third-party commercial sources form an important part of the information AI platforms retrieve.

Invest in community and forum reputation

Understand how your brand is represented across forums, reviews, YouTube, Reddit, and other community environments.

Strengthen first-party visibility deliberately

Improve crawlability, technical accessibility, structured data, and direct-answer content around the queries that matter most.