What Content Each AI Model Prefers: Intent Preferences Analysis Across 6 Providers

You produce content — articles, guides, comparisons, reviews. But if someone asks: "Which AI model will this content drive visibility in?" — most teams cannot answer. That is a problem.

According to the AthenaHQ State of AI Search 2026 (analysis of 8 million AI responses in Q1 2026), different AI models exhibit fundamentally different intent preference patterns. Some models generate responses predominantly for informational queries, others for comparative ones, and others cover all types evenly. This means a brand that produces only informational content will be well visible in Copilot and nearly invisible in AI Overview — and vice versa.

This article provides concrete data across 6 AI providers and a practical framework: what content to create to be noticed where your audience actually is.

Copilot: The "Wikipedia Model" with an Extreme Informational Bias

Microsoft Copilot is the most predictable provider from an intent distribution perspective. Its pattern repeats across virtually every vertical: Informational dominates overwhelmingly, and this is the most pronounced bias among all 6 models.

AthenaHQ data by vertical for Copilot:

• Real Estate: Informational — 69.91%. The highest value across all models and verticals in the study.
• Healthcare: Informational — 51.94%. Second intent: Optimization/Improvement (9.54%).
• Government: Informational — 48.59%. Optimization/Improvement — 11.40% (also a record for this intent type).

What this means in practice: when a user asks Copilot about real estate, healthcare, or government services, nearly 7 in 10 responses are built around informational content. Copilot explains what a mortgage is, how insurance works, what a particular term means.

A second distinctive feature of Copilot is an anomalously high Optimization/Improvement intent share compared to other models. These are queries like "how to improve," "how to optimize," "what can be done better." For Copilot this is 9-11% in several verticals — in other providers this intent typically stays below 5%.

Google AI Overview: The Only Comparative-Biased Model

Google AI Overview is the polar opposite of Copilot. It is the only provider where Comparative/Selection intent regularly exceeds or matches Informational.

AthenaHQ data by vertical for AI Overview:

• Tech: Comparative/Selection — ~25-30%, Informational — ~23%. Comparative is higher.
• Logistics: Comparative/Selection — 32.25%, Informational — 31.65%. Comparative is higher.
• Government: Comparative/Selection — 25.25%, Informational — 23.46%. Comparative is higher.

This is an anomaly compared to all other models. For ChatGPT, Gemini, and Perplexity, Informational always leads by a significant margin. AI Overview is the exception.

Another distinctive feature of AI Overview: a notably high Learning/Education intent share (17-21% depending on the vertical). These are educational queries — "how does this work," "explain the steps," "what should I learn." AI Overview, unlike Copilot, covers this category well.

The reason for this anomaly is clear: Google historically is a search engine for commercial decisions. Users come to Google when they want to choose — and AI Overview serves exactly these query types.

Perplexity: The Most Balanced Profile

Perplexity is the hardest provider to "target" precisely. Its intent distribution is the most even among all 6 models.

AthenaHQ data for Perplexity:

• Informational: 30-42% (varies by vertical)
• Comparative/Selection: 12-27%
• Acquisition/Obtaining: 7-19%
• Remaining 6 intents: 3-10% each

Perplexity does not ignore any intent type. It is a research-oriented tool that people use for deep exploration of topics — which is why it "accepts" diverse content.

For brands, this means: Perplexity can be covered with multiple content types. An informational article, a comparative review, a buying guide — all have a chance of appearing in responses. Strategically, Perplexity favors brands with a broad content portfolio.

For more on optimizing for Perplexity, see How to increase cited sources in Perplexity.

ChatGPT: A "Middle of the Road" Profile Without Extremes

ChatGPT demonstrates a moderate profile close to the market average. It is not as informational-biased as Copilot and not as comparative-oriented as AI Overview.

AthenaHQ data for ChatGPT:

• Informational: 31-43%
• Comparative/Selection: 14-28%
• Acquisition/Obtaining: 9-21%

In essence, ChatGPT strikes a reasonable balance where Informational leads but other intents are meaningfully represented. This makes ChatGPT a versatile platform — it responds to educational content, commercial content, and comparative content alike.

However, a "middle" profile does not mean "easy." For ChatGPT, source credibility and brand recognition are critical: the model knows large international brands and authoritative independent domains well, but may not know local brands even with high-quality content. More on this in How ChatGPT selects sources.

Gemini: Informational Plus a Learning Lean

Gemini is the second provider with a pronounced Informational bias, though less extreme than Copilot.

AthenaHQ data for Gemini:

• Healthcare: Informational — 50.31%
• Real Estate: Informational — 51.80%
• Learning/Education: 13-20% — noticeably higher than most other providers

Unlike Copilot, Gemini is more active with Learning/Education intent. This is explained by its integration into Google's educational products and a general orientation toward explaining and teaching. Where Copilot explains "what this is," Gemini more often explains "how this works" and "how to learn this."

For brands in education, healthcare, and financial services — Gemini is a strong platform for deep educational materials. Step-by-step guides, course-style content, mechanism explanations — this content performs well.

Claude: The Careful Investigation Model

Claude is the most "cautious" model in the context of commercial recommendations. AthenaHQ data on Claude is less granular by vertical than for other providers, but general patterns are documented.

Claude consistently shows:

• A lower share of Acquisition/Obtaining intent — Claude avoids direct "buy here" recommendations and transactional content.
• A higher share of Investigation/Research — deep, analytical content. Claude responds better to research materials that include citations, data, and sources.
• Caution with Comparative/Selection — Claude rarely delivers categorical "A is better than B" recommendations, preferring balanced analysis.

For brands, this means Claude is not a platform for direct commercial push. But it is an excellent platform for brands investing in thought leadership: research reports, white papers, analytical pieces, data-driven content.

Intent Preferences Matrix by Model

AthenaHQ State of AI Search 2026 data (Q1 2026, 8M AI responses, averaged across verticals).

Legend: ★★★★★ — dominant intent (over 45%), ★★★★ — high (30–45%), ★★★ — moderate (15–30%), ★★ — present (5–15%), ★ — low (under 5%).

Practical Mapping: What to Write for Each Model

Intent preference data translates directly into content strategy. Here is how it works in practice.

For Copilot: Build an Information Hub

Copilot needs content that explains and informs, not content that sells.

What works:

• Detailed explanatory articles ("What is X," "How Y works")
• Industry terminology glossaries
• Criteria selection guides (not "buy us" but "here is how to evaluate in general")
• "How to improve X" and "how to optimize Y" materials (Optimization/Improvement intent)
• Wikipedia-style: facts, definitions, structured information

What does not work: landing pages with CTAs, "why we are the best" lists, direct purchase prompts.

For AI Overview: Comparative Content

AI Overview wants to see your brand in the context of comparisons and choices.

What works:

• "X vs Y" pages (direct competitor comparisons)
• "Best solutions for Z" roundups — with your brand included
• "How to choose between A and B" articles with evaluation criteria
• Educational materials with practical steps (Learning/Education intent)
• Category hub pages with comparison tables

What does not work: purely informational articles with no commercial angle.

For Perplexity: A Diversified Content Portfolio

Perplexity is a platform where almost everything works. Prioritize based on audience volume and keyword opportunities.

What works:

• Any high-quality long-form content with sources and citations
• Research and data pieces (Perplexity frequently cites specific facts)
• News and updates (Update/News intent)
• Commercial content with specific product characteristics and specs

Strategy: create content covering at least 5 of the 9 intent types — Perplexity will cover the distribution.

For ChatGPT: Source Authority and Trust

ChatGPT is well-balanced across intents but sensitive to source credibility.

What works:

• Content on authoritative domains with strong reputations
• Mentions in independent media and review sites
• A mix of informational and commercial content
• International presence (ChatGPT knows brands with an English-language footprint better)

Strategy: invest in earned media — mentions in authoritative sources outperform owned content.

For Gemini: Education and Expertise

Gemini responds to deep educational and expert-level content.

What works:

• Step-by-step guides and tutorial-style materials
• Content explaining mechanisms and underlying principles
• E-E-A-T signals: expert authorship, data, cited research
• Integration with the Google ecosystem (Google Business Profile, structured data)

For Claude: Research and Analysis

Claude values depth and sources; it avoids commercial pressure.

What works:

• Research reports and white papers
• Analysis with data and primary source citations
• Content presenting multiple perspectives without categorical conclusions
• Thought leadership with first-hand expert experience

What does not work: aggressive commercial content and explicit "choose us" messaging.

Checklist: Auditing Your Content Strategy by AI Provider

Use this checklist to evaluate your current content portfolio:

• Run an intent audit: identify what content types you already have — informational, comparative, educational, transactional
• Define priority models: where does your audience spend time — B2B Copilot users, Google searchers with AI Overview, Perplexity researchers?
• Create comparison pages if targeting AI Overview — an "X vs Y" page for each key competitor
• Build an information hub if targeting Copilot — at minimum 10-15 educational articles covering key niche topics
• Add Investigation/Research content to gain visibility in Claude — studies with data, analytics, industry reports
• Diversify for Perplexity: cover at least 5 of the 9 intent types with quality content
• Update structured data: Article, FAQPage, HowTo schema — they help AI models classify your content by intent type
• Set up per-provider monitoring: track in which models your brand appears for informational vs commercial queries — GEO Scout shows this broken down by prompt type
• Check YandexGPT and Alice AI: for the Russian market — build a separate strategy for the Yandex ecosystem
• Repeat the audit in 30 days: model intent preferences shift with updates — monitoring must be ongoing

A content strategy that ignores the intent preferences of specific AI models is a strategy executed blindly. AthenaHQ State of AI Search 2026 data provides a clear enough picture: Copilot wants information, AI Overview wants comparisons, Claude wants research. Your job is to make sure you have the right content for each provider where your audience is.

To understand how your brand currently appears across different AI models for different query types — start monitoring on GEO Scout.