The Web Is Splitting in Two

On this page

Key Takeaways The Scale of the Machine Audience AI Visits Your Website and Gets It Wrong Your Website Now Has Two Audiences Why AI Search Changes Everything What Structured GEO Programs Actually Achieve Two Versions of the Internet What This Means for Your Business What You Can Do About It When You Cannot Close the Gap In-House FAQ Related Reading Sources

For 30 years, the internet had one audience: people. People clicked links, read pages, compared products, and made buying decisions. Every website was designed for that workflow. That era is ending. Not because humans are leaving the internet, but because a second audience showed up, and it is growing faster than anyone expected. Ahrefs estimates that roughly 25% of all web requests now come from AI bots. AI-driven search traffic grew over 1,300% between 2023 and 2025 (Similarweb). This article explains why the web is splitting into two audiences, what that means for your business, and what you can do about it.

Key Takeaways

AI bots now account for roughly 25% of all web requests according to Ahrefs, and user-action AI crawling grew 15x in 2025 (Cloudflare).
Most of that crawling does not send anyone back. Anthropic makes 38,065 crawls for every single human visitor it refers back. OpenAI's ratio is 1,091:1 (Cloudflare).
AI-referred traffic converts at 15.9% compared to 1.76% for Google organic, based on GA4 data across seven months (Seer Interactive).
Brands are 6.5x more likely to be cited through third-party sources like Reddit and review sites than through their own domains (All About AI).
Content with schema markup has a 2.5x higher chance of appearing in AI-generated answers (SchemaApp).
Companies running structured GEO programs see 3-10x citation rate improvements within 60-90 days. Ramp grew AI visibility 7x. Popl moved from #5 to #1 in AI Share of Voice with 1,561% ROI.

The Scale of the Machine Audience

To understand how big this shift already is, look at the crawler data.

Cloudflare's 2025 analysis shows GPTBot traffic surged 305% in a single year, climbing from 4.7% to 12.8% of all AI bot requests. PerplexityBot grew 157,490%. ClaudeBot now accounts for 11.4% of AI crawler traffic. There are 21 major AI bots actively crawling the web, and the number keeps growing.

User-action AI crawling, the kind triggered when someone asks ChatGPT a question, grew 15x in 2025 alone.

But here is the uncomfortable part: most of that crawling does not send anyone back. Cloudflare's "crawl-to-click gap" research found that Anthropic makes 38,065 crawls for every single human visitor it refers back to a website. OpenAI's ratio is 1,091:1. Training accounts for 80% of AI crawling. Search accounts for only 18%.

That means the value of being read by AI is not in clicks. It is in being cited. Being recommended. Being the answer. If AI reads your site and cannot extract clean information, it does not just skip you. It recommends someone else.

AI Visits Your Website and Gets It Wrong

Here is the part most business owners miss.

Your website looks great to humans. Clean design, pricing cards, testimonials, feature comparisons, product pages with all the right details. Everything a potential customer needs.

But when an AI agent visits that same page, it often sees something very different: navigation menus repeated on every page, cookie consent banners, tracking scripts, CSS and JavaScript that has not finished loading, and content hidden behind client-side rendering.

So when someone asks ChatGPT "How much does [your company] charge?" or "Is this product HIPAA compliant?" the AI might answer incorrectly. Not because you did not publish the information. Because the AI could not extract it from the noise.

Brands are 6.5x more likely to be cited through third-party sources like Reddit and review sites than through their own domains. When the AI cannot read your site, it fills in the blanks from wherever it can.

That matters more than you think. Bain & Company found that 80% of consumers now rely on AI-generated answers for 40% or more of their searches. If AI gets your pricing wrong or your features wrong, that wrong answer reaches thousands of potential customers before you even know it happened.

Your Website Now Has Two Audiences

This is the shift. Every business website now serves two distinct readers with completely different needs.

Humans want beautiful pages, interactive experiences, videos, animations, smooth checkout flows, and modern design. They judge your brand partly on how your site looks and feels.

AI agents want structured facts. Clean text. Pricing in a format they can parse. Product features they can compare. They need "what matters" without the visual noise.

Here is the hard truth: AI does not care about your website design. It cares about whether it can extract accurate information. A gorgeous $50,000 website redesign is invisible to ChatGPT if the product data is buried in JavaScript that never renders for crawlers.

Content with schema markup has a 2.5x higher chance of appearing in AI-generated answers. These are not SEO tricks. They are basic requirements for communicating with the machine audience.

Why AI Search Changes Everything

For two decades, "search" meant Google. Ten blue links. SEO rankings. Blog posts optimized for keywords.

That model is breaking down. 60% of Google searches now end without a click. Gartner projects that traditional search volume will drop 25% by 2026 as users shift to AI assistants.

People are also skipping Google entirely. They ask AI directly: "What's the best CRM for small teams?" or "Compare these two project management tools." The AI does not return ten links. It gives one answer with two or three recommendations. ChatGPT alone now handles 5.4 billion monthly visits, exceeding Bing's 1.9 billion.

That changes the competitive dynamic completely. You are no longer fighting for a spot on page one. You are fighting to be one of the two or three brands the AI mentions at all.

And the conversion numbers back this up. Seer Interactive found that ChatGPT referral traffic converts at 15.9% compared to 1.76% for Google organic, based on GA4 data from October 2024 through April 2025. Perplexity converts at 10.5%, Claude at 5%. Visitors from AI have already done their research inside the AI conversation. By the time they click through to your site, they are ready to buy.

A caveat: AI referral volume is still small, roughly 0.07% of organic traffic for most sites. But it is growing fast, and the visitors it sends are far more qualified.

What Structured GEO Programs Actually Achieve

The companies that have adapted to this split are seeing measurable results. Here are published benchmarks from named companies running structured generative engine optimization programs:

Company	Category	Key Result	Timeframe
Ramp	Fintech SaaS	AI visibility 3.2% to 22.2% (7x), 300+ citations	1 month
Airbyte	Data Integration SaaS	ChatGPT visibility 9% to 26% (3x), $100K deal from ChatGPT	1 week initial lift
Popl	Digital Business Card SaaS	AI Share of Voice #5 to #1, 1,561% ROI, 18-day payback	Ongoing
Tinybird	Real-time Analytics	Share of Voice 11% to 32% (3x), LLM traffic +370%	3 months
BairesDev	Software Outsourcing	Third-party presence 16% to 78%	60 days

The pattern: companies that combine structured content, technical optimization, and continuous execution see 3-10x improvements in AI citation rates within 60-90 days. AI-referred traffic converts 4.4x better than standard organic search, with average engagement times of 8-10 minutes versus 2-3 minutes from traditional Google.

Two Versions of the Internet

We are entering a world where every business effectively needs two versions of its web presence.

The human web is what your customers see today: your normal website with its design, branding, and interactive elements.

The machine web is what AI consumes: simplified, structured versions of your key pages. Clean text. Explicit pricing. Product specs in formats AI can parse without guessing. Think machine-readable layers, schema markup, and protocols like llms.txt that tell AI crawlers what to read and how to cite it.

Same business. Same information. Two formats.

Companies that only maintain the human version will gradually disappear from AI answers. Not from the internet itself, but from the recommendations people actually act on.

What This Means for Your Business

If AI cannot read your website properly, the consequences are concrete:

You do not appear in AI recommendations for your category
Competitors become the default answer
Potential customers choose someone else before ever visiting your site
Wrong information about your pricing or features spreads at scale

This is the new version of being invisible on Google. Except it is worse. Google shows ten results, so even position seven gets some clicks. AI gives one answer. You are either in it or you are not.

What You Can Do About It

Three practical steps any business can start with today.

1. Test whether AI can actually read your site

Open ChatGPT, Perplexity, and Gemini. Ask them about your company. Ask about your pricing, your product, how you compare to competitors. See if the answers are accurate. Most businesses are shocked at what they find.

2. Create a machine-readable layer

This means structured data (schema markup), server-side rendered content, clean text versions of key pages, and formats like llms.txt that help AI crawlers understand your site. The goal: make it easy for AI to get your information right. For a deeper dive on what this involves, see our guide on building a machine-readable layer.

If you want to handle this yourself, start with schema markup on your highest-traffic pages and work outward. Ensure AI crawler bots (GPTBot, PerplexityBot, ClaudeBot) are not blocked in your robots.txt. Add an llms.txt file that tells AI models what content to read. Deploy JSON-LD structured data on your product, pricing, and comparison pages.

3. Start tracking AI visibility

You already track SEO traffic, Google rankings, and ad performance. Now you also need to monitor how often AI mentions your brand, whether those mentions are accurate, and how much traffic AI referrals actually drive. This is where the growth is shifting.

When You Cannot Close the Gap In-House

Most companies get through the audit and testing phase before hitting a wall. Content teams have no bandwidth to create a parallel content program with different formatting requirements. Engineering has a six-month sprint backlog. Nobody on the team has deep expertise in how LLMs select and cite sources. The monitoring dashboard becomes an expensive report nobody acts on.

Disclosure: Mersel AI is the publisher of this article and offers the managed service described below. We have made every effort to present the DIY path fairly and completely above.

For companies that lack the internal bandwidth to execute, Mersel AI runs the two-layer system as a fully managed program:

Layer 1: Citation-first content engine. We build prompt maps from sales call recordings, competitor citation patterns, and your category's AI answer landscape. From that map, we publish structured content directly to your CMS on a continuous cadence, connected to Google Search Console and GA4 to track which posts earn citations and refine based on real performance data.

Layer 2: AI-native infrastructure. We deploy a machine-readable layer behind your existing website. Clean entity definitions, structured schema markup, llms.txt configuration, and AI-crawler-optimized rendering. Human visitors see nothing different. No engineering resources required.

Client results from this approach:

A Series A fintech startup building a unified finance OS saw AI visibility increase from 2.4% to 12.9% over 92 days, with non-branded citations up 152% and 20% of demo requests influenced by AI search.

A publicly traded quantum computing company saw AI citation rates increase from 1.1% to 5.9% over 123 days, with 214 citations across quantum computing prompts and a 16% quarter-over-quarter increase in AI-influenced enterprise leads.

FAQ

What does it mean that the web is splitting in two?

It means every website now serves two distinct audiences: human visitors who browse visually, and AI agents (ChatGPT, Perplexity, Claude, Gemini) that extract structured information to generate answers. Most websites are only built for the first audience, making them partially invisible to the second. Ahrefs estimates 25% of all web requests now come from AI bots, and that share is growing.

How much web traffic comes from AI bots?

Approximately 25% of all web requests come from AI bots according to Ahrefs. User-action AI bot crawling grew 15x in 2025 (Cloudflare), and nearly 69% of websites now receive some AI-driven traffic. GPTBot alone surged 305% in one year, climbing from 4.7% to 12.8% of all AI bot requests.

Do AI-referred visitors actually convert?

Yes, at significantly higher rates than traditional search. Seer Interactive found ChatGPT referral traffic converts at 15.9% compared to 1.76% for Google organic. Perplexity converts at 10.5%, Claude at 5%. However, AI referral volume is still small, roughly 0.07% of organic traffic for most sites. The visitors are few but highly qualified.

What is a machine-readable layer?

A machine-readable layer is structured content added to your existing website specifically for AI agents to parse. It includes schema markup (JSON-LD), server-side rendered content, clean text versions of key pages, and formats like llms.txt that guide AI crawlers. Human visitors see no difference. For a full technical walkthrough, see what is a machine-readable layer for AI search.

How do I check if AI agents can read my website correctly?

Open ChatGPT, Perplexity, and Gemini and ask them about your company, pricing, and features. Compare their answers to your actual information. Most businesses discover significant inaccuracies in how AI represents their brand. This takes five minutes and costs nothing. For a more systematic approach, see how to measure AI visibility.

Ready to see how AI currently reads your site? Book a free 20-minute audit and we will show you exactly what ChatGPT, Perplexity, and Claude see when they visit your pages.

Want to understand the full framework first? Read our complete guide to generative engine optimization for a breakdown of how AI search works and what drives citations.

Sources

Ahrefs. "AI Search Statistics." ahrefs.com
Adobe Digital Insights. "AI traffic to retail sites, 2025." adobe.com
All About AI. "AI Hallucination Statistics 2026." allaboutai.com
Bain & Company. "Goodbye Clicks, Hello AI: Zero-Click Search Redefines Marketing." bain.com
Cloudflare. "AI Crawler Traffic by Purpose and Industry." cloudflare.com
Cloudflare. "From Googlebot to GPTBot: Who's Crawling Your Site in 2025." cloudflare.com
Cloudflare. "The Crawl-to-Click Gap." cloudflare.com
SchemaApp. "What 2025 Revealed About AI Search and Schema Markup." schemaapp.com
Seer Interactive. "6 Learnings About How Traffic from ChatGPT Converts." seerinteractive.com
Similarweb. "AI Search Traffic Growth." similarweb.com
Similarweb. "Generative AI Statistics 2026." similarweb.com