People talk about "AI search" as one thing. It isn't. The engines your customers use build their answers in different ways, from different sources, and they change every time a model is updated.
This comparison uses the GEO Compass benchmark: more than 900,000 AI answers and 6.6 million classified source citations, collected between August 2025 and September 2026 across 11 industries and 24 markets. Every figure is pooled across studies and anonymised. You can reproduce any of them in the explorer.
How many sources each engine cites
The number of distinct websites behind an answer varies more than twofold between engines.
GEO Compass benchmark, all industries and markets, Aug 2025 – Sep 2026.
Fewer sources means each citation carries more weight. Being one of four sources in a Gemini answer is a stronger position than being one of ten in AI Mode, and it is harder to win.
What kind of sources each engine prefers
All engines lean on brand and company websites, but the balance differs:
| Engine | Brand & corporate | Earned media | Social & UGC | Academic & research |
|---|---|---|---|---|
| ChatGPT | 26% | 24% | 5% | 8% |
| Google AI Overviews | 32% | 15% | 12% | 8% |
| Google AI Mode | 36% | 17% | 12% | 11% |
| Perplexity | 27% | 23% | 8% | 7% |
| Gemini | 40% | 18% | 4% | 13% |
| Claude | 38% | 25% | 6% | 7% |
| Microsoft Copilot | 25% | 18% | 4% | 16% |
Share of all cited sources, by type. Remaining shares are reference, government, institutions, marketplaces and other sites.
Three patterns stand out:
- Google's AI surfaces cite social and community content about twice as often as ChatGPT, Gemini or Claude.
- Gemini and Claude rely most on brand sites. Clear, factual, well-structured owned pages matter most here.
- ChatGPT, Perplexity and Claude give earned media its largest share. Coverage in the publications they trust is a direct route into their answers.
Each engine has favourite platforms
The clearest differences are in which platforms each engine reaches for.
GEO Compass benchmark. Video is effectively a Google-only visibility lever.
- YouTube is a Google lever. If Google's AI Overviews and AI Mode matter to your audience, video explainers, reviews and demos are part of GEO.
- LinkedIn appears in 16% of Perplexity answers and 14% of AI Mode answers, and only 4% of ChatGPT answers. Executive and company content on LinkedIn reaches these engines.
- Wikipedia is cited in 24% of OpenAI API answers, against 9% in the ChatGPT app. The same company's models behave differently in different products.
- Reddit appears in about 12% of ChatGPT and Google AI answers, and almost never in Claude's.
How often each engine names brands
In questions that don't name a brand, engines differ in how often they name the brand the question set was built around, and in how often they cite that brand's own site.
| Engine | Focus brand named | Own site cited |
|---|---|---|
| Google AI Mode | 34% | 15% |
| Gemini | 35% | 10% |
| Claude | 36% | 9% |
| ChatGPT | 27% | 13% |
| Google AI Overviews | 27% | 13% |
| Microsoft Copilot | 23% | 14% |
| Perplexity | 18% | 8% |
Unbranded questions only. "Own site cited" is a share of answers that use sources.
Perplexity names brands least often in category questions, and its answers are shorter. ChatGPT and Google's AI surfaces sit in the middle. Gemini and Claude name brands more often but cite their sites less, which means they are drawing the description from somewhere else.
Model updates move citations overnight
Engines aren't stable. Within the same studies, the switch from Claude Sonnet 4.6 to Sonnet 5 in August 2026 cut earned media's share of citations from 42% to 30%. ChatGPT's move to live web search over 2025–26 changed what it cites more than any optimisation a brand could make.
The Trends page tracks model versions. The lesson for GEO is to track visibility by engine and model version, and to re-check your position after every major release.
What this means for your GEO plan
- Start with the engines your audience uses. In most Asia-Pacific markets that means Google AI Overviews and ChatGPT first, then Gemini and Perplexity. See the market pages for local engine coverage.
- Match the source type to the engine. Video for Google's surfaces; earned media for ChatGPT, Perplexity and Claude; clean, factual owned pages for Gemini and Claude; LinkedIn for Perplexity and AI Mode.
- Measure each engine separately. An average across engines hides the gaps that matter.
Explore any engine in your category with the GEO Compass explorer.
Frequently asked questions
- Which AI engine cites the most sources?
- In the GEO Compass benchmark, Google AI Mode cites the most: about 10.4 distinct websites per answer that uses sources. ChatGPT, Google AI Overviews and Perplexity cite about 8.7 to 8.9. Gemini and Microsoft Copilot cite fewer than five.
- Which AI engine cites YouTube most?
- Google's AI surfaces. YouTube appears in 38% of Google AI Overviews answers with sources and 37% of AI Mode answers, against 4% for ChatGPT, 5% for Gemini and under 1% for Claude.
- Does ChatGPT cite Reddit?
- Yes. Reddit appears in about 12% of ChatGPT answers with sources in the benchmark, similar to Google AI Overviews and AI Mode. Claude almost never cites Reddit.
- Should brands optimise for each AI engine separately?
- Yes. The engines differ in how many sources they use, which platforms they favour and how often they name brands. Model updates within the same engine can also change cited sources, so tracking should be by engine and model version.
Figures marked as GEO Compass data come from the anonymised GEO Compass benchmark of AI answers collected by We AI & Data Advisors. Related: AI search in Australia · AI search in Singapore · AI search in India · AI search in Europe