AI Search Market Share 2025 Report [User & Referral Traffic]

When ChatGPT became the fastest-growing consumer application in history, it was easy to glorify it as the “ultimate” AI search engine. With over 700 million monthly active users (MAU), that number alone seemed to tell the whole story; the more users, the better the chance you have at gaining visibility and traffic. Seems logical, right? But if you’ve been focused solely on that one metric, you're missing the bigger picture of what's happening in AI search.

The real story isn't just about active users; it's also about referrals. What are referrals? They’re the number of visitors who click a citation or link they discovered in an AI search platform and are sent to your site. This metric is a defining factor in the new organic search landscape as it tells you where people are coming from when they land on your site. 

Here’s what may shock you: while chat-first engines like ChatGPT and Gemini are winning at referrals, platforms with built-in distribution (think Google AI Overviews and Meta AI) are winning users. Between February and July of 2025 alone, roughly 1 billion people actively used Meta AI, placing it second only to Google AI Overviews in terms of MAU, and capturing around 22.6% of the total AI search market share.

All of this data highlights our new reality. The battle for users versus the battle for referrals is defining the AI search world, and understanding the difference is critical for navigating organic search today. We're going to help you navigate it by giving you a clear read on this market using two complementary datasets:

• Referral Share: 2,802,519 AI-search referral sessions captured via Google Analytics across 41 brand sites (May 1-Aug 31, 2025).‍
• Distribution Baseline: Public MAU and weekly active user (WAU) figures from earnings, investor materials, and reputable reporting (all cited).

The Two Lenses Marketers Need

Glorifying ChatGPT based on a one-dimensional view was a mistake. Why? Because it solely focused on users, not referrals; and surprise surprise, ChatGPT isn’t even the top player for users.

To truly understand AI search, you have to look at it through two distinct lenses: who is sending traffic to your site, and who has the users. 

Who Has the Users? (The Distribution Lens)

This refers to platforms that have a built-in, massive audience, measured here in monthly active users (MAU). These are often not chat-first engines but existing platforms that have integrated AI to enhance their core user experience. 

The leaders here are Google, with its AI Overviews baked directly into the world's most popular search engine, and Meta, which has integrated AI into its billion-user-strong ecosystem of Facebook, Instagram, and WhatsApp. The goal of these platforms is often to answer a user's query directly and keep them within their ecosystem.

Who Sends You Traffic? (The Referral Lens)

But user counts don’t tell the full story. When you shift to the referral lens, which tracks outbound clicks, the leaderboard flips.

This refers to platforms that are designed to cite external sources and drive users to your website. These are the chat-first engines whose business models are built around a more conversational, source-citing user journey. 

Platforms like ChatGPT and Perplexity are dominant here, and their nature means they are far more likely to convert a user session into an outbound click.

But why does all of this matter? We’ll break it down for you:

1. Brand Visibility vs. Conversion: AI Overviews on Google offer a new form of brand visibility, where your company can be the featured "answer" without a click ever occurring. This is a brand victory that you need to be able to measure and report on, as AI Overviews show up now for 13% of searches. Conversely, platforms like ChatGPT are a direct ROI play, as they are proven to send high-intent, high-value referral traffic that leads to conversions and revenue. 
2. Budget Allocation: The two lenses directly impact budget allocation. You have to decide whether to invest in "distribution defense" (optimizing for visibility on Google and Meta's "zero-click" platforms) or "referral harvest" (optimizing for clicks on chat-first engines). Failing to understand the difference between these two types of AI platforms means working with a one-dimensional strategy in a multi-dimensional search world, leading to misallocated resources and missed opportunities.

Let's take a deeper look at the data. 

Deconstructing the Data: Referral Share & User Distribution

The data reveals two distinct realities in AI search. Here, we’ll outline the numbers, first with Referral Share, which is the actual traffic sent to websites, and then by examining Distribution Baseline, the platforms with the largest user bases. By looking at these two metrics together, we can understand the full picture of which platforms are dominating and why.

1. Referral Share (Measured Clicks From AI Platforms)

Referral share tells us which AI platforms actually send users to brand websites. When an AI model cites your content, how often does that translate into a click? And which ones produce the most referrals? 

From 2.8M sessions across 41 sites (May 1-Aug 31, 2025):

1. ChatGPT: 89.10%
2. Microsoft Copilot: 3.20%
3. Perplexity: 3.10%
4. Google Gemini: 2.40%
5. Claude: 1.35%
6. Grok: 0.55%
7. DeepSeek: 0.20%
8. Hugging Face: 0.06%
9. Qwen: 0.04%

Important: This dataset excludes Google AI Overviews and AI Mode, because those appear inside Google Search and are not attributed as Referral sources in Google Analytics. Amazon Rufus is also excluded as it is part of a closed ecosystem.

2. Distribution Baseline (Who Actually Has The Users)

Distribution tells us which AI platforms are winning the battle for user volume. Which platforms have the largest number of active users? And how does that reach then impact their overall market share?

Latest public figures show the real reach picture:

• Google AI Overviews: 2.0B MAU (global)
• Meta AI: ~1.0B MAU (across apps)
• ChatGPT: ~700M WAU
• Gemini app: >450M MAU
• Google AI Mode: >100M MAU (US & India, still rolling out)
• Microsoft Copilot: >100M MAU
• Perplexity: ~22M active users; ~780M monthly queries (May)
• Grok (xAI): ~30–35M MAU (estimates vary, not company-reported)
• Claude: ~16–18.9M monthly users/unique visitors (tracker-based)

What The Data Actually Says (& How to Act on It)

The data tells a story of two parallel realities. On one side, we have the user distribution; the sheer volume of people using platforms like Google AI Overviews and Meta AI. On the other hand, we have referral share, a metric that reveals which of these platforms is actually sending traffic to sites.

The first takeaway is a wake-up call for anyone focused on the traditional organic search model. The platforms with the biggest user base are not currently the ones driving the most referrals. This disconnect highlights a critical missed opportunity and presents a new challenge for marketers: you can be present in a billion-user ecosystem without receiving a single click. 

1. Google Already Owns AI Distribution; You Just Don’t “See” It in Referral Logs Yet

Between its colossal AI Overviews (2 billion MAU) and the rapidly expanding AI Mode (100M+ MAU), Google’s AI surfaces dwarf every other chat engine in terms of raw user reach. 

The catch? Google’s AI doesn’t show up in referral logs. Every AI Overview or AI Mode click is buried under “organic search” in GA4.

Think of classic SEO as the building blocks for visibility inside AI Overviews and AI Mode by prioritizing schema, trustworthy source coverage, and retailer and community signals, not only chat citations. You can do this through: 

• Authoritative Content: Building undebatable topical authority and a brand that demonstrates H-E-E-A-T (Helpfulness, Expertise, Experience, Authority, and Trust) through unique data, expert bios, and third-party mentions.
• Schema: Using schema to help Google's AI models understand your content's key facts, relationships, and entities.
• Brand Mentions: Optimizing for brand visibility by being a trustworthy source cited by other reputable publishers. Your goal here is to be the answer, not just a link in a list.

2. Meta AI Is the Sleeping Giant for Top-of-Funnel

With almost 1 billion MAU across WhatsApp, Instagram, and Facebook, Meta AI has unparalleled ambient reach. It’s a tool that users engage with casually throughout their day. Today, linking out is limited, but the primary function of Meta AI is to provide instant answers and recommendations, which means it's a powerful layer for awareness, consideration, and branded answers.

Treat Meta AI like you would an early-stage distribution channel like Instagram Stories in 2018. The goal isn't last-click attribution, but brand seeding. Your strategy should focus on:

• Snackable Content: Creating authoritative, bite-sized content and branded entities that are easy for the model to understand and quote.
• Audience Building: Using your presence to build trust with users who may not be in a search-first mindset, but are looking for a quick answer from a trusted source. This is about building a brand that a user might remember and search for later.

TL;DR: While Google and Meta AI have the most users they don’t show up in referral logs. This means marketers need a new strategy. To win Google's AI Overviews and AI Mode, you should optimize for SEO fundamentals like authoritative content, schema, and brand mentions. For Meta AI, which has a huge reach but limited linking, the focus should be on building brand awareness and creating "snackable" content that the model can easily use for instant answers. And remember that both platforms are designed to keep users within their ecosystem, not link them out. 

3. ChatGPT Dominates Referrals Because It’s Built to Link

Our dataset shows that 89% of measured AI referrals come from ChatGPT. This is a dramatic over-indexing relative to its ~15.8% share of the overall blended distribution baseline. What this really means is that ChatGPT sends far more clicks than its market share would suggest, because it’s designed to, unlike much of Google and Meta AI’s integrated features. 

To help you understand the true value of this data, we’re gonna give you a new metric that breaks down these numbers even further. We call the comparison between these two datasets the Referral Efficiency Index (REI): referrals relative to distribution (referrals / active users). Based on this formula, ChatGPT’s REI is approximately 5.6x.

This phenomenal efficiency is a direct result of its design: it's a "chat-first" engine that is built to provide an answer and then, crucially, cite its sources with clear, clickable links.

You should optimize for ChatGPT when your priority is getting traffic, now. This requires a strategy focused on:

• Citation-Ready Content: Creating concise, authoritative content that is easy for the model to pull from and attribute. Think data-heavy articles, definitive guides, and well-structured Q&A pages.
• Publisher Coverage: Earning mentions and backlinks from credible third-party publishers that the model loves to quote, acting as a force multiplier for your brand's authority.

4. Perplexity Punches Far Above Its Weight

With a relatively small user base (~0.5% baseline share), Perplexity still drove a remarkable 3.1% of all referrals in our data. This gives it an REI of approximately 6.2x, the highest in our analysis (move over, ChatGPT). 

In other words, a single Perplexity user is far more valuable for generating traffic than a user on other platforms.

Why? Its product is designed to cite and for users to click out. If your content is in a research-heavy, technical, or comparison-driven category, you can harvest real, high-quality traffic from Perplexity with:

• Expert Guides: Long-form, in-depth content that addresses complex topics.
• Structured Answers: Using clear headings, bullet points, and an easy-to-scan format that makes it simple for the model to synthesize and cite your information.

5. Copilot Has Reach, But Middling Referral Yield

Despite its integration into the Windows and Microsoft 365 ecosystems and a user base of over 100M MAU, Copilot delivered only 3.2% of our measured referrals (an REI of ~1.4x).

This low efficiency is intentional; Copilot's strength is task completion and in-app assistance inside Microsoft's surfaces, not pushing users outward to the open web.

If you sell to enterprise or developer audiences, your focus for Copilot should not be on traffic. Instead, it should be on ensuring your brand and product are:

• Accurate & Discoverable: Prioritizing the accuracy of your brand's information so that when users ask about your products or services, the model provides a correct, authoritative answer.
• Integrated: Exploring opportunities to integrate your product or service with Microsoft 365 and other Microsoft properties, making it an essential part of a user's workflow rather than an external destination.

While it might seem that Copilot is underperforming, you need to remember that this platform has a different purpose than others. It’s designed for users to complete tasks within the Microsoft ecosystem, not send them back out onto the web. So, for the purpose it was designed for, Copilot actually isn’t failing; it’s performing exactly as intended. 

Quick Comparison: Who Reaches Users vs. Who Sends You Traffic

This chart provides a side-by-side comparison of the AI search landscape, moving beyond a single metric to reveal the full story. 

On one side, we have Distribution Footprint, which shows the raw number of users each platform commands. 

On the other hand, we have Referral Share, which measures the actual traffic each platform sends to websites. By analyzing these two metrics together, it’ll help us decipher what each platform is uniquely good for.

* REI (Referral Efficiency Index) = Referral share ÷ Blended baseline share (only where apples-to-apples was possible).
† The Gemini line mixes app MAU with web/app referrals; treat the ratio directionally.

Measurement: Why Your GA4 is Undercounting AI Citations (& What to Do)

In the past, you could look at a Google Analytics referral log and get a clear picture of where your traffic was coming from. That's no longer the case; AI has created a measurement chasm, making it more difficult than ever to attribute success accurately. 

Here’s why your standard GA setup is undercounting your AI visibility and how you can fix it.

1. The Invisible AI Traffic: Google's AI Overviews

Google AI Overviews and AI Mode, despite having over 2.1 billion MAU combined, don't show up in your GA4 referral logs. They're a part of Google Search, so you’ll see outcomes as organic impressions and clicks, not as a separate referrer.

• The Problem: You could be a top-cited source in an AI Overview and have no way to see that directly in a "referrals" report. This hides a huge portion of your visibility and brand authority.
• The Action: Don't rely on referral data alone for Google's AI. Instead, you need to conduct SERP-level audits and use AEO observability to see your content's presence. Look for pages with high impressions in Search Console but low clicks, as this is a prime indicator of an AI-driven "zero-click" victory.

2. Instrumenting Direct LLM Referrals in GA4

Traffic from chat-first platforms like ChatGPT, Perplexity, and Claude is often miscategorized under generic "referrals" or even "direct traffic.” To get a clear picture, you must explicitly configure your GA4 to track these sources.

• The Problem: Standard GA4 setup doesn't have a parameter for AI traffic. This means you can't easily compare the performance of ChatGPT with that of your organic search or social media channels.
• The Actions:
  
  • Custom Channel Group: The most effective method is to create a custom channel group in GA4's Admin section. Define a new channel called "AI Search" and use a regular expression (regex) to match known AI referrers (e.g., chatgpt.com, perplexity.ai, copilot.microsoft.com, claude.ai, gemini.google.com). This automatically categorizes historical and future traffic from these sources.
  • Custom Dimensions: For a deeper analysis, you can create a custom dimension (e.g., traffic_source = "AI Search") to enrich session data. This allows you to report on specific AI platforms and their performance, directly within your reports.

3. Tracking AI Crawlers

Before AI can cite your content, its specialized web crawler must crawl and index it. These crawlers (like GPTBot or PerplexityBot) are a key indicator of public interest.

• The Problem: Standard GA4 may not differentiate between a standard search bot and an AI agent, so you're missing the earliest signs that a model is engaging with your content.
• The Action: Monitor your server logs for hits from AI user agents. You can also create an LLMs.txt file on your site to set specific rules and hints for how these crawlers should summarize and interact with your content. This is a proactive step that gives you a measure of control over how your brand is represented in AI responses.

4. The New Attribution Reality

You might think a user’s journey from visibility to conversion with AI might look something like this: visibility → mentions → citations → referrals, but in fact, it’s a non-linear chain. Here’s why:

• The Problem: Your analytics may show a "direct" or "organic" conversion when a user first discovered your brand through an AI chat. This makes it impossible to accurately attribute the initial touchpoint.
• The Action: Use a combination of tools and metrics (model-level observability and web analytics) to see the full picture.
  
  • Web Analytics: Use your custom GA4 setup to track the direct referrals you are able to capture.
  • AI Observability Tools: Utilize third-party tools like Goodie that actively monitor when your brand is cited or mentioned in AI model responses, regardless of whether a click occurs.
  • Qualitative Data: Add questions to your user surveys or lead forms asking how they first heard about your brand. This qualitative data can help fill in the gaps left by your web analytics.

Platform Playbook 

With the data deconstructed, it's time to translate insights into action. This playbook outlines the specific strategies you need to employ for each major AI platform to capitalize on its unique strengths, whether for user distribution, referrals, or both.

Google (AI Overviews, AI Mode & Gemini)

Your strategy here is inclusion first, clicks second. Google’s AI features are designed to keep users on the search results page. Your goal is to be the trusted source that an AI model relies on for its answer, even if the user never clicks your link.

• Treat It Like Next-Gen SEO: The same core signals that help you rank in traditional search, trust, authority, and experience, are critical for AI. This is a battle for H-E-E-A-T.
• Optimize for Machine Readability: Use schema (like FAQPage and HowTo) to provide Google's AI with a machine-readable map of your content.
• Build Trust Signals: Focus on a strong backlink profile from authoritative sites, positive community reviews, and strong social proof. These are the signals that models use to determine if a source is trustworthy.
• Answer Directly: Write short, definitive answer blocks in your content that models can lift verbatim. For example, under an H2 (that should be phrased as a question), write out a one or two-sentence-long answer before delving into a deeper explanation.

Meta AI

Expect minimal link-out from Meta AI. With its unparalleled reach of nearly 1 billion users, this platform is a brand-building and mindshare channel, not a traffic driver.

• Focus on Brand Entities: Ensure your brand's core information (names, products, services, and unique value propositions) is easy to render in a single answer.
• Create Authoritative Content: Publish content that positions your brand as a definitive source for specific topics. Think comprehensive guides, research reports, and fact-based content that can be referenced.
• Monitor Mentions: Use Goodie to track how your brand is being mentioned in Meta AI responses to ensure accuracy and brand-safe sentiment.

Perplexity

This is your #1 near-term traffic driver. Our data shows that Perplexity is the most efficient platform for converting user sessions into outbound referrals, with an REI of ~6.2x.

• Publish Comparison Content: Perplexity excels at citing content that compares and contrasts products, services, or methodologies. Publish detailed comparison tables, spec sheets, and pros and cons lists to make it easy for the model to cite your content.
• Go Deep on Methodology: Users on Perplexity are often technical or research-focused. Publish content that explains the "how" and "why" behind your data or findings.
• Stay Fresh: Perplexity prioritizes new information. Keep your content up-to-date with new data and trends to remain a relevant citation source.
• Watch the Publisher Experiments: If you're a publisher, monitor Perplexity's news licensing and revenue-sharing experiments. This could be a significant new revenue stream for content producers.

ChatGPT

Treat ChatGPT like a turbocharged research engine. Its users are in research mode and are highly motivated to click out for a deeper dive. Its REI of ~5.6x proves that it punches far above its weight.

• Build Source Density: Models like ChatGPT prioritize information they see repeated across multiple trustworthy sources. Ensure your site, high-authority reviewers, credible forums, and third-party retailer product pages all tell a consistent story.
• Monitor Answers & Close Perception Gaps: Regularly check how ChatGPT answers questions about your brand or industry. If a model misrepresents your brand or provides inaccurate information, create new, targeted content that corrects the record and provides a factual source for the model to cite.
• Prioritize Citation-Ready Content: Create content with clear sections, headings, bullet points, and summaries that make it easy for the model to extract and cite as a source.

Copilot

Copilot is less about driving external traffic and more about productivity and enterprise workflows. Its primary value is in task completion inside the Microsoft ecosystem.

• Prioritize Accuracy: Ensure your brand and product information is correct and easily accessible for the model. Success is a brand-correct answer inside a customer's workflow.
• Explore Integrations: If your business sells software, services, or data, investigate opportunities to integrate your offering with the Microsoft 365 ecosystem.
• Focus on Internal Use Cases: Publish content that shows how your product or service can be used within the Microsoft suite (e.g., "How to use our product data with Copilot in Excel").

Claude

Claude is known for its long-form reasoning and empathetic tone, making it a valuable tool for customer service and complex problem-solving with a respectable REI of ~3.1x.

• Focus on Detailed Explanations: Create content that breaks down complex topics into empathetic and polished explanations.
• Build Customer Service-Style Content: Publish comprehensive FAQs, troubleshooting guides, and "how-to" content that answers common customer inquiries.
• Use Conversational Language: Write in a conversational, helpful tone that mirrors the way Claude interacts with users.

The Takeaway for 2025 Planning

The most dangerous assumption a marketer can make in 2025 is that AI search is a single, unified channel. It's not. It encompasses two distinct problems that require a dual-pronged strategy: reach (Google, Meta) and referrals (ChatGPT, Perplexity, Claude).

The Budget Split

As you plan your budget for 2025, a methodical split is highly recommended to reflect these two distinct problems:

• 60–70% for Distribution Defense: This is your budget for AI inclusion on platforms that prioritize reach over referrals. Think of this as your long-term investment in brand visibility. It includes efforts to optimize for Google's AI Overviews, create brand-safe content for Meta AI, build a strong community presence, and secure mentions in publisher coverage. This is a proactive defense against the "zero-click" phenomenon.
• 30–40% for Referral Harvest: This is your budget for AI traffic generation. It is a more direct, near-term ROI play. These funds should be dedicated to creating highly citable content, building source density across the web, and monitoring citation opportunities on platforms like ChatGPT, Perplexity, and Claude. This is about actively harvesting clicks from the platforms that are designed to send them.

How to Figure Out Where You Stand

You can't manage what you don't measure. You may not like it, but you need to abandon outdated KPIs and adopt a modern framework that reflects the nature of AI search. We’ve created a rubric to help you figure out your brand’s current standing that should help you identify potential opportunities and build a data-driven strategy. 

The Metrics that Matter

• Inclusion Rate (Distribution Defense): This measures how often your content is a cited source in AI Overviews and AI Mode for your top 100 queries. This is your primary metric for "reach." Since this traffic is not disaggregated in GA4, you'll need to use Goodie to track your performance.
• Citation Density (Referral Harvest): This measures how often you are a cited source across chat-first engines like ChatGPT, Perplexity, and Claude. This is your primary metric for "referrals." These platforms are far more likely to send traffic, making this a direct measure of your content's ability to drive clicks. You can also measure this in Goodie. 
• Referral Yield (ROI Generator): This goes beyond just sessions. It measures the number of sessions and, more importantly, the assisted conversions you get from AI referrers. You must instrument your analytics with custom channels and dimensions to accurately track this data. This metric directly connects your content strategy to your bottom line.

Perception Gap Delta (Brand Health): This is a critical metric for a brand-safe era. It measures the distance between how AI models talk about your brand and your canonical truth. Are models sharing accurate information? Is the sentiment positive, neutral, or negative? You must monitor this in Goodie to protect your brand in real-time.

Final Word

Distribution ≠ referrals. Google and Meta already command the audience. ChatGPT (and Perplexity) command the clicks and users’ mindshare. 

Winning AI search in 2025 means having a strategy that acknowledges both of these realities. You must simultaneously:

1. Defend the Defaults: On platforms like Google and Meta, your goal is to be the best source to lift when an answer forms. This is the long game of building trust and authority.
2. Harvest the Clicks: On platforms like ChatGPT and Perplexity, your goal is to publish valuable content so that it becomes a primary citation, driving direct traffic and conversions to your site.

This isn't about choosing one path over another; It's about recognizing that the AI landscape has split into two separate, but equally critical, challenges. Your success will be measured by how effectively you can master both being the answer and driving the click.

Appendix: Sources Behind the Distribution Baseline

• Alphabet Q2’25 (AI Overviews 2B MAU; AI Mode 100M+ MAU; Gemini app >450M MAU). 
• Meta Q1’25 (Meta AI “almost 1B” MAU).
• ChatGPT (~700M WAU, multiple Reuters updates). 
• Microsoft Copilot (>100M MAU, Reuters). 
• Perplexity (780M monthly queries, TechCrunch; ~22M active users, Business of Apps). ‍
• Claude (~16–18.9M users / visitors, Backlinko, and trackers).