What 2 million LLM sessions reveal about AI discovery

We analyzed nearly two million LLM sessions across nine industries from January through December 2025. We started with a simple assumption: ChatGPT dominates, usage patterns are uniform, and the volume is small and inconsequential.

The data proved us wrong.

ChatGPT commands 84.1% of trackable AI discovery traffic, but it functions primarily as the default tool for broad-market discovery. That reality changes the strategy.

Brands can no longer rely on a single, discovery-first approach. You need a multi-platform strategy that aligns with how users expect to be productive at different moments.

Success now depends on knowing which platforms actively enable user productivity and which simply support early discovery.

Different LLMs are winning in different industries, often by wide margins. The takeaway for 2026 is more nuanced than “focus on ChatGPT.”

Here’s what the data reveals.

The growth rate divergence: ChatGPT vs. everyone else

From January to December 2025, the major LLM platforms grew at very different rates:

• ChatGPT: 3x growth
• Copilot: 25x growth
• Claude: 13x growth
• Perplexity: 1x growth
• Gemini: 1x growth

ChatGPT grew, but Copilot and Claude grew eight to 10 times faster. Perplexity and Gemini effectively flatlined, or, more accurately, reinforced usage within tightly defined knowledge workflows.

These aggregate numbers reflect deeper strategic priorities.

• Satya Nadella publicly highlighted Copilot reaching 100 million monthly users.
• Dario Amodei announced that Anthropic’s revenue grew from $100 million to $8–10 billion in under two years.
• Aravind Srinivas posted that he’s “really encouraged by the interest in Perplexity Finance,” even positioning it as an alternative to Bloomberg Terminal.

These CEOs are focused on growth because growth signals real user value:

• Copilot wins by serving Microsoft ecosystem users.
• Claude wins with developers.
• Perplexity wins with finance professionals.

Different LLMs are winning different industries at dramatically different rates.

Pattern 1: Copilot dominates where work happens

Copilot’s 25x aggregate growth is striking, but the industry breakdown makes the pattern obvious. Copilot wins in B2B verticals where work already happens inside the Microsoft ecosystem.

SaaS

• ChatGPT: 2x growth
• Copilot: 21x growth
• Copilot adoption mirrors how modern SaaS teams operate. Companies embed LLMs directly into workflows to extract insights from proprietary and third-party data, driving efficiency, personalization, and product innovation inside Microsoft tools.

Education

• ChatGPT: 6x growth
• Copilot: 27x growth
• Copilot benefits from a culture of knowledge sharing and research synthesis. Institutions and publishers cite, expand, and contextualize existing material, making LLM-assisted discovery a natural extension of how educational content is created and consumed.

Finance

• ChatGPT: 4.2x growth
• Copilot: 23x growth
• Finance aligns strongly with Copilot because many tasks are automated and context-dependent. Analysts need models that can source, reconcile, and reason across authoritative reports, filings, and datasets inside trusted environments.

The key insight isn’t just Copilot’s growth. It’s where that growth occurs. Copilot accelerates fastest in industries where professionals already depend on Microsoft tools to analyze data, synthesize knowledge, and complete tasks.

A finance analyst doesn’t leave Excel to “search.” They ask Copilot to interpret, compare, and contextualize data in place. A content or product strategist doesn’t open a new tab to research competitors. They prompt Copilot inside their working environment.

What it means

If your audience lives within enterprise workflows — SaaS teams, financial professionals, educators, and B2B decision-makers — AI discovery is moving into LLMs as work happens. Visibility is no longer won during early research. It’s won during execution, when intent is highest and decisions are already forming.

Pattern 2: Perplexity only survives in finance

Perplexity’s overall growth sits at 1.15x, effectively flat. But when you isolate finance, a different picture emerges.

In finance, Perplexity holds a 24% market share.

This is the only industry where Perplexity maintains meaningful, sustained traffic. Everywhere else, its share has collapsed:

• SaaS: down from 14.9% to 7.3%
• E-commerce: down from 13.9% to 3.4%
• Education: down from 28.5% to 5.2%
• Publishers: down from 41.5% to 3.6%

Finance behaves differently because financial decisions demand verification.

When users compare investment platforms, evaluate loan terms, or research compliance requirements, a single synthesized answer isn’t enough. They need citations they can trace directly back to source documents.

Perplexity is built for this use case. Through partnerships with Benzinga, FactSet, Morningstar, and Quartr, it provides direct access to earnings transcripts, SEC filings, analyst ratings, and real-time market data.

Its Enterprise Finance product adds scheduled market updates, custom answer engines, and live data visualizations. These features serve professionals who require auditable, institutional-grade information, not just fast answers.

Every answer includes visible sources that users can click to verify each claim.

In most categories, convenience wins. In finance, trust and verifiability are non-negotiable.

What it means

Success in AI discovery means choosing the right platform for your users and being present in the sources and citations the models themselves trust.

Financial responses rely on networks of licensed data, institutional partners, and authoritative third-party references. If your brand isn’t visible, cited, and validated inside those ecosystems, you won’t surface, no matter how strong your content is.

Optimization now means earning relevance across the full web of sources each model draws from, not just ranking in a single interface.

Pattern 3: Claude dominates standalone analysis

Claude represents just 0.6% of total AI discovery traffic, which makes it easy to dismiss. But where that 0.6% concentrates is revealing. Claude wins with professionals who research, write, and analyze, not consumers who shop.

• Publishers: 49x growth
• Education: 25x growth
• Finance: 38x growth
• SaaS: 10.3x growth

Why does Claude win in these verticals when Copilot already dominates knowledge work?

The difference is the type of work. Copilot lives inside operational tools like Excel, Word, and PowerPoint, helping professionals execute tasks within existing workflows. Claude is where professionals go for standalone strategic thinking.

• A publisher uploads an 80,000-word manuscript and asks, “Is this argument coherent across chapters three through seven?”
• A finance analyst uploads three years of earnings transcripts and asks, “How has management’s language around capital allocation changed?”
• A developer pastes an entire legacy codebase and asks, “Map the data flow and identify architectural bottlenecks.”

Claude’s 200,000-token context window enables this. The value isn’t efficiency inside a workflow. It’s having a reasoning partner for work that requires synthesis, critique, and strategic judgment.

What it means

If you target technical audiences or strategic decision-makers, Claude optimization demands analysis-grade content. Publish deep case studies with clear methodology and detailed implementation paths, not 500-word summaries.

Structure content for reasoning. Use explicit frameworks and comparative analysis. The audience is smaller, but the influence is higher. A developer who uses Claude to deeply analyze your API documentation becomes an internal champion.

Pattern 4: The Gemini measurement crisis

Gemini’s tracked traffic tells a confusing story:

• Education: −67% tracked traffic
• SaaS: +1.4x growth
• Finance: +1.3x growth
• E-commerce: +2.7x growth

This likely isn’t a user decline. It’s an attribution collapse.

Over the past 13 months, Gemini has increasingly kept users inside its interface. It delivers AI-generated answers without prominent, clickable source links. Users research, absorb the answer, and either convert directly or search brand names later. That journey never shows up as AI discovery.

Google still controls the largest search distribution network in the world, and Gemini is deeply embedded in it. It’s unlikely Gemini users are abandoning AI discovery while ChatGPT grows 3x and Copilot grows 25x.

What’s more plausible is that Gemini-driven discovery still exists, but it’s becoming invisible.

Unlike Perplexity, which surfaces sources, or Copilot, which operates inside traceable workflows, Gemini synthesizes answers and retains users in Google’s ecosystem.

A user asks Gemini about project management software, gets a complete answer, then searches “[your brand]” days later. Analytics record branded search, not AI influence.

This creates a real strategic risk.

The commonly cited “0.13% AI penetration” metric is almost certainly understated. If even 30% to 40% of Gemini-assisted discovery goes untracked, true AI-driven research volume could be two to three times higher than what we can measure.

What it means

• Monitor branded search lift alongside AI optimization efforts.
• Build measurement models that account for multi-session, cross-platform journeys.
• Invest in brand strength and recall, not just clicks.
• Track time-lagged conversions as research and conversion drift further apart.

Last-click attribution is breaking. AI-assisted conversions — where users research in one system, synthesize in another, and convert through branded or direct search — are becoming the default. Flat or declining Gemini traffic likely signals measurement failure, not user absence.

How to choose your LLM strategy based on your audience

AI discovery isn’t consolidating around a single platform. It’s fragmenting by industry, use case, and user intent.

• If your audience works in enterprise environments: Copilot is where discovery happens. SaaS buyers, financial analysts, educators, and B2B decision-makers research inside Microsoft tools like Excel, Outlook, and Teams. Discovery occurs at the moment decisions form, not during separate “research” sessions.
• If your audience makes high-stakes decisions: Perplexity matters. Finance is the only industry where a secondary platform holds a 24% share alongside ChatGPT. These users need citations, not synthesis. Optimization means earning visibility inside institutional data networks such as FactSet, Morningstar, and financial news, not just ranking in the interface.
• If your audience includes technical evaluators: Claude’s 0.6% share understates its influence. Developers, strategists, and researchers use it for deep analysis by uploading full documents and datasets. They are fewer, but they shape buying committees. Content must go deep: detailed case studies, clear methodology, and analysis-grade research.
• If you’re in an emerging category: Legal, events, and insurance show 15x to 90x growth because AI discovery just arrived. Start with ChatGPT’s broad reach, then watch for platform migration as your audience matures.
• If measurement is breaking: Gemini’s declining tracked traffic likely reflects attribution collapse, not user loss. Monitor branded search lift. Track time-lagged conversions. Build models that account for multi-session, cross-platform journeys.
• Across all categories: Expect attribution gaps. Traditional last-click attribution is breaking as AI-assisted conversions become the norm.

The future of AI discovery isn’t about ranking on ChatGPT alone. It’s about understanding where your audience discovers and which platforms actually serve their needs.

The full study. 2025 State of AI Discovery Report: What 1.96 Million LLM Sessions Tell Us About the Future of Search