Why the Same URL Is Invisible in ChatGPT and Dominant in Grok

TL;DR

The most expensive mistake in GEO in 2026 is treating all AI engines as if they were the same channel. They are not. According to GEO Metrics' analysis of ~400,000 responses across 7 engines over 14 months, two engines answering the same question share on average only 10% of their sources. A URL can be cited in 60% of Grok's responses and in 0% of ChatGPT's — for the same prompt, on the same day. The source strategy that works in one engine is, in many cases, completely irrelevant for the next.

There is a conceptual mistake we make every time we talk about "appearing in AIs" as if it were a single thing.

It is not a single thing. It is seven different things — and in many cases, opposite ones.

GEO Metrics Research Lab #03 data leaves no room for interpretation: two AI engines answering the same question share on average only 1 source domain out of every 10. The typical pair shares none.

That means a URL that leads citations in Grok can be completely invisible in ChatGPT. Not because the content is bad — but because ChatGPT and Grok draw from nearly independent source universes.

The Number That Changes Everything: Jaccard 10%

To quantify the overlap between each engine's source universes, we calculated the Jaccard similarity — the percentage of shared domains over the total unique domains cited — between each pair of engines for the same query.

The average: ~10%.

Not 40%. Not 60%. 10%. For the same question, any two engines share on average 1 source out of every 10. The pair with the highest overlap in the study — AI Mode and AI Overviews, the Google family — reaches 24.6%. ChatGPT and Grok, with the most opposite diets, share 6.1%.

The implication is direct: if your GEO strategy optimizes for a single engine, you are playing in 10-25% of the ecosystem — and ignoring the remaining 75-90%.

Why This Happens: Source Diets Are Incompatible

The reason is not random. Each engine has a different information retrieval architecture, and that architecture defines which source types it prioritizes.

Grok and the Google engines (AI Mode, AI Overviews) are social-first. Grok cites social networks in 80.9% of its responses and YouTube in 66.5%. AI Mode and AI Overviews replicate that profile. For these three engines, social presence and video content are the most powerful source signals.

ChatGPT is editorial. Wikipedia appears in 15.8% of its responses. News media in 13.6%. These are the highest editorial proportions in the study — and ChatGPT is practically the only engine that consistently covers that ground.

Perplexity distributes. Balanced across video, community, professional networks and editorial sources. The most diversified engine — and the most predictable in volume (standard deviation of just 2.2 sources per response).

Copilot goes niche. Specialized comparison portals through the Bing index. For Copilot, what matters is not Wikipedia or YouTube — it is appearing in the right comparison site for the category.

DeepSeek cites nothing. Zero sources in 100% of its responses across the 14 months of analysis. Its visibility depends exclusively on the training corpus — no PR that impacts in real time, no link equity that enters.

The Concrete Case: What Works in Grok Doesn't Work in ChatGPT

Imagine a technology brand with a solid YouTube content strategy — tutorials, reviews, video comparisons with good distribution and volume. That strategy is perfect for Grok and the Google engines.

Now imagine that same brand has no Wikipedia entry, doesn't appear in editorial media and has not published studies or proprietary data that could be referenced by high-authority outlets. For ChatGPT — which bases 15.8% of its citations on Wikipedia and 13.6% on news media — that brand practically doesn't exist as a verifiable source.

The result: the same URL, with the same content, can be the dominant source in Grok and completely invisible in ChatGPT. Nothing is broken. It is the system working exactly as designed — with radically different source criteria.

The only way to know is to measure both universes separately.

The Three Strategies That Emerge From the Data

Strategy 1 — To Dominate Grok, AI Mode and AI Overviews: Social Presence and YouTube

These three engines draw primarily from social networks and video. The source strategy for them runs through:

• YouTube content with titles that answer high-intent questions — because YouTube is the most cited domain by all three engines

• Active presence on X (Twitter) with indexable text content

• LinkedIn content with proprietary data not on the website

A brand with no active, indexable social presence is giving up three of the seven engines in the study — including the two Google engines with the highest user volume.

Strategy 2 — To Dominate ChatGPT: Editorial Authority

ChatGPT feeds on editorial sources. The source strategy for it runs through:

• Wikipedia — the highest-ROI investment for ChatGPT. It is the only citation asset with cross-engine traction across the entire study

• Presence in news media and high-authority sector publications

• Proprietary data published in sources ChatGPT considers references — not just on the own blog

A brand that only optimizes its own domain without working the external editorial layer is ignoring the source signals ChatGPT weights most.

Strategy 3 — To Dominate Copilot: Niche Comparison Sites

Copilot has the most specific source diet in the study. The strategy runs through:

• Being well positioned in the 2-3 most relevant niche comparison sites for the category

• Maintaining updated profiles on G2, Capterra or the sector equivalent

• Optimizing the Bing Places profile and ensuring the Bing index has the correct data

For B2B brands operating in Microsoft environments, Copilot has very high penetration — and its source criteria are completely different from the rest.

The Consequence for Monitoring

If source universes are different by engine, monitoring also needs to be by engine.

An aggregated Share of Voice — "my brand appears in X% of AI responses" — can hide completely opposite realities: leading in Grok and AI Mode while being invisible in ChatGPT and Copilot. Or vice versa.

GEO Metrics measures Share of Voice, position and cited domains separately for each of the 9 main models — ChatGPT, Gemini, Perplexity, Claude, Copilot, DeepSeek, AI Mode, AI Overviews and Grok — with daily monitoring and per-prompt breakdown.

That is what allows seeing not just whether you appear, but in which engines you are cited as evidence and with what type of sources — which is the question that defines GEO strategy in 2026.

Data: GEO Metrics Research Lab #03 · ~400,000 responses · 7 engines · 14 months · March 2025 – May 2026

Measure your visibility by engine in GEO Metrics → trygeometrics.com