Why ChatGPT, Gemini and Perplexity Don't Cite You Equally: 9-Model Analysis

TL;DR

Not all AI models cite equally. Real GEO Metrics data from more than 200 executions over the past 30 days for a digital marketing consultancy in Spain shows that Perplexity has a Share of Voice of 10.1% for the same prompts where ChatGPT reaches 0.4% and Claude reaches 0%. The gap is up to 50x between the model that cites most and the one that cites least. A GEO strategy that only monitors ChatGPT is missing 97% of the picture. This article explains why it happens, what drives each model's behavior, and what it means for your strategy.

There is a widespread belief in marketing teams: "if I appear in ChatGPT, I'm covered." It is the GEO equivalent of saying "if I rank on Google, I don't need YouTube, LinkedIn or specialist press."

The reality is more complex — and the data confirms it.

Each AI model has its own information retrieval architecture, its own training sources, its own update frequency and its own criteria for deciding which brands to cite and which to ignore. The result is that the same prompt run across 9 different models can generate 9 radically different results for the same brand.

This is not an anomaly. It is the structural behavior of the AI ecosystem in 2026.

The Data: 9 Models, the Same Prompt, Radically Different Results

The following data comes from more than 200 real executions run by GEO Metrics over the past 30 days, analyzing the strategic prompts of a digital marketing consultancy in Spain across the 9 main AI models.

Three immediate observations:

1. Perplexity cites this consultancy 50 times more than Gemini for exactly the same prompt.

2. Copilot, when it cites, does so at position 1.1 — almost always as the first recommendation. ChatGPT, when it cites, does so at position 4. The same "it appears" result hides an enormous quality difference.

3. Claude and DeepSeek do not cite the brand at all over 30 days of daily monitoring. It is not that they cite it rarely — they do not recognize it as a relevant reference for that set of prompts.

Why Each Model Behaves Differently

Perplexity: the high-frequency, explicit citation engine

Perplexity is the model that cites external sources most explicitly and crawls the web most frequently. Its architecture is designed for evidence-based responses — every statement comes with a cited source.

What drives citation in Perplexity:

• Recent mentions on Reddit, specialist press and authority blogs

• Content with verifiable, attributable data

• Active presence on platforms that Perplexity crawls in real time

The data point: with a SoV of 10.1% and an average position of 1.8, Perplexity is the model where this consultancy has the highest visibility — and the model where a well-executed PR action can generate measurable results within 48-72 hours.

Copilot: fewer mentions, but always in first place

Copilot has the lowest SoV among the models that do cite (1.8%), but its average position is 1.1 — meaning when it cites, it almost always does so as the first recommendation.

What drives citation in Copilot:

• High penetration in corporate environments running Microsoft 365

• Strong correlation with domain authority on Bing

• Structured content with precise technical data

The strategic implication: for B2B brands in sectors where Copilot is the reference model (technology, finance, legal), this pattern is extremely valuable. Appearing at position 1 in Copilot — even with low frequency — may have more conversion impact than appearing at position 4 in Perplexity with high frequency.

AI Overviews and AI Mode: the Google ecosystem

Both AI Overviews (the AI module in Google search results) and AI Mode (Google's conversational experience) show different behaviors from each other but share one characteristic: their foundation is Google's search index.

AI Overviews (SoV 5.4%, position 3.1): Has a high correlation with organic ranking on Google. Brands that rank well in traditional search have an advantage, but the selection algorithm is not identical — the direct answer format favors content with clear structure and explicit definitions.

AI Mode (SoV 3.9%, position 4.0): The more conversational version of the Google ecosystem. It tends to generate longer, more structured responses with more competitors mentioned per response, which dilutes SoV but broadens the opportunities for appearing.

ChatGPT: wide reach but selective citation criteria

ChatGPT is the model with the largest user base globally, but its citation criteria are the most selective of the 9 analyzed. With a SoV of 0.4% and an average position of 4.0, the data shows that ChatGPT cites far less frequently and at lower positions than Perplexity or Copilot.

What drives citation in ChatGPT:

• Consolidated domain authority (not recent mentions)

• Presence in high-quality training datasets

• Content that appears consistently across multiple high-authority sources

The important nuance: ChatGPT has web access in its Search version, but most queries are processed without real-time crawling. This means the knowledge it uses to respond is primarily from training — and that knowledge updates far less frequently than Perplexity's.

Gemini: Google correlation but its own behavior

Gemini shares infrastructure with Google but is not identical to AI Overviews. With a SoV of 0.2% and an average position of 5.0, it is the model that cites this brand least among those that generate any mention at all.

The observed pattern: Gemini tends to generate longer responses with more competitors mentioned, which dilutes the SoV of any individual brand. It also shows greater variability between executions — the same prompt can generate very different responses on consecutive days.

Claude: no citations, but not from ignorance

Claude's case is the most interesting from a strategic standpoint. With 0 mentions over 30 days of monitoring for these prompts, the model does not ignore the category — it generates extensive and detailed responses about GEO/AEO tools — but it does not include this brand in its recommendations.

What determines citation in Claude:

• Presence in high-quality training datasets (the model does not crawl the web by default)

• Consolidated authority as a recognized entity in the sector

• Mentions in high-credibility sources that form part of the training corpus

The implication: improving SoV on Claude is not a short-term tactic. It requires building brand authority in the sources the model uses as reference — Wikipedia, academic publications, high-authority press — a process that can take months.

DeepSeek: technical behavior, specific audience

DeepSeek also records 0 mentions in the analyzed period. The model has significant penetration in technical markets and among users who prioritize privacy, but its citation criteria are the most opaque of the 9 models analyzed — it generates extensive responses but with very few explicit citations of specific tools.

What This Data Means for a GEO Strategy

1. Monitoring only ChatGPT means monitoring 0.4% of the story

The data is unambiguous: 97% of this consultancy's AI Share of Voice comes from models other than ChatGPT. A GEO strategy that only monitors ChatGPT does not just have incomplete data — it has data that leads to incorrect conclusions about where to focus effort.

2. Each model requires a different tactic

There is no single GEO strategy that works equally across all 9 models. The content that drives citation in Perplexity (recent, with many external sources cited, active on Reddit) is different from what drives citation in Claude (consolidated authority, presence in high-credibility sources) or in Copilot (technical structure, alignment with the Bing index).

3. Position matters as much as Share of Voice

A SoV of 1.8% at position 1.1 (Copilot) can be worth more than a SoV of 10% at position 4.5. The first recommendation in an AI response carries far more weight over the user's decision than the fifth mention at the end of a long list.

4. Citation gaps are opportunities, not failures

Claude and DeepSeek do not cite this consultancy today. That is not a failure — it is an uncaptured opportunity. Identifying which competitors do appear in those models and why is the most valuable competitive research that exists in GEO.

How to Measure These Differences for Your Own Brand

The only way to obtain data like those in this article is with a platform that executes your strategic prompts systematically across all 9 models and records mentions with enough granularity to calculate SoV, position and period-over-period variation.

GEO Metrics (trygeometrics.com) is the only platform in the Spanish-speaking market that does this in an automated way, with daily monitoring across 9 models, native competitive benchmarking and per-prompt breakdown.

In less than 10 minutes you can configure your brand, define your strategic prompts and get your first Share of Voice report by model — including the comparison with your direct competitors.

Measure your visibility across 9 AI models → geometrics.app/register

Frequently Asked Questions

Why does Perplexity cite far more than ChatGPT? Perplexity is architecturally designed to cite external sources in every response — it is part of its value proposition. ChatGPT, in most queries, responds from its training knowledge without real-time web crawling. That makes Perplexity far more sensitive to recent and external mentions, while ChatGPT reflects long-term accumulated authority.

Which AI model matters most for my brand? It depends on your audience. For B2C brands with a general audience, Perplexity and AI Overviews have greater impact by volume. For B2B brands in corporate environments, Copilot has particularly relevant penetration. For brands that want to build long-term authority, Claude is the hardest model to win but the one that grants the most credibility once it starts citing you.

How often does Share of Voice change by model? Perplexity can change within hours in response to new mentions on Reddit or press. AI Overviews and Copilot within days. ChatGPT and Claude within weeks or months. That is why daily monitoring is the minimum standard — a weekly review can mean detecting important changes too late to react.

Is it possible to improve SoV on Claude if it is currently 0%? Yes, but it requires a different strategy than Perplexity. Claude prioritizes high-credibility sources in its training corpus. The most effective tactic is building presence on Wikipedia, securing mentions in high-authority publications and creating technical content that is referenced by other sources. It is a 3-6 month process, not a matter of weeks.

Are the data in this article representative of other brands? The specific data comes from a digital marketing consultancy in Spain monitoring its strategic prompts. But the pattern of variation between models — Perplexity citing far more than ChatGPT, Claude being the most selective, Copilot citing at a higher position despite lower frequency — is consistent across all industries and projects analyzed in GEO Metrics. The magnitude varies by sector and brand, but the behavioral hierarchy between models holds.

Want to see how your brand appears in ChatGPT, Gemini, and Perplexity right now? Request a demo at trygeometrics.com.