Websites were designed for people. Every layout decision, every button colour and every checkout step assumes a human on the other side of the screen, scrolling, hesitating and clicking. That assumption is breaking down. AI agents that browse, compare and complete tasks on behalf of users are now arriving on websites in growing numbers, and they behave nothing like the visitors that analytics tools and conversion strategies were built around.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
From Crawlers to Agents
Automated traffic is not new. Search engine crawlers, price scrapers and monitoring bots have visited websites for decades. What is new is the type of bot. Traditional crawlers index pages. AI agents act: they read product details, fill in forms, compare options and in some cases complete a purchase or a booking for the person who sent them.
Consumer agent products moved quickly from demos to general availability. OpenAI launched ChatGPT agent in July 2025, folding its earlier Operator tool into a system that can browse visually and through a text browser. Perplexity launched its Comet browser the same month and made it free to all users in October 2025. Google has rolled agentic browsing into Gemini and Chrome. The market has also been volatile, with some early products already retired or merged into others.
What Agents Do Differently
An AI agent does not browse the way a person does. It does not get distracted by a hero image, linger on a promotional banner or respond to urgency messaging. It reads the page structure, looks for the information it was asked to find and moves on as soon as it has it.
That changes the meaning of familiar metrics. A short session with no scrolling might signal a frustrated human, or it might be an agent that found the price in two seconds and left satisfied. A form abandoned halfway through might be a confused customer, or an agent that hit a field it could not interpret. Bounce rates, time on page and scroll depth all become harder to read when a meaningful share of visitors are not people.
Analytics and A/B Tests Get Noisy
Conversion work depends on clean data, and agent traffic can quietly distort it. If agents are counted as visitors, conversion rates may fall even when human behaviour has not changed. If an A/B test splits traffic randomly, agents may land in both variants and respond to neither design element in the way the test assumes.
That is why experimentation now needs a step that was rarely discussed a few years ago: identifying and segmenting automated sessions before drawing conclusions. Specialist partners such as CROforce work on exactly this kind of conversion problem, where the question is not only which variant wins but which visitors the result actually reflects. Without that segmentation, teams risk optimising pages for a mix of humans and machines that represents neither group well.
Bots Already Outnumber People
The scale of automated traffic is now hard to ignore. Imperva’s 2026 Bad Bot Report found that automated traffic made up more than 53% of all web traffic in 2025, meaning bots now outnumber people online. Not all of that traffic is AI agents, and much of it is malicious, but the trend points in one direction.
Measurements vary by network, and Cloudflare’s chief executive predicted in March 2026 that bot traffic would overtake human traffic on its network by 2027, noting that an agent shopping on a user’s behalf might visit far more sites than a human ever would. For website owners, that means a growing share of sessions will come from software making decisions for someone else.
Designing Pages Agents Can Read
The response from the web industry has been to make sites more legible to software without making them worse for people. Clear page structure, descriptive labels, consistent product data and accessible forms all help agents complete tasks accurately. Many of these practices overlap with accessibility, because agents often rely on the same underlying page structure that screen readers use.
Google has started testing what this looks like in practice. An experimental Lighthouse category for agentic browsing, reported by CMSWire, audits readiness using checks that include the accessibility tree, layout stability and the presence of an llms.txt file, which gives AI systems a structured guide to a site’s content. The category reports a pass ratio rather than a single score, reflecting how early this area still is.
Knowing Which Agents to Trust
Not every automated visitor deserves the same treatment. Website owners need to tell the difference between a legitimate agent acting for a real customer and a scraper or attacker pretending to be one. Industry efforts are forming around cryptographic verification, in which an agent signs its requests so a site can confirm who sent it. Cloudflare introduced signed agents in 2025, and the approach is being formalised through an IETF working group on web bot authentication.
For conversion teams, verification creates an opportunity. Verified agents can be counted separately, offered a streamlined path to complete tasks and excluded from tests designed for human behaviour. Unverified bots can be filtered or blocked.
Rethinking the Funnel
The classic conversion funnel assumes a person moves from awareness to interest to decision on the same site. Agents compress that journey. A user may ask an assistant to find the best option, and the agent may compare a dozen sites in seconds before returning a recommendation or completing the task itself.
In that world, the persuasive copy on a landing page matters less than the accuracy and completeness of the information behind it. Clear pricing, reliable availability data and straightforward checkout flows become competitive advantages, because they make a site easier for an agent to choose. Tech Insider’s comparison of Comet, Gemini Agent and ChatGPT Atlas, which found that only one survives after rivals were shut down or folded into other products, shows how quickly the tools themselves are changing, which makes flexible, standards-based site design the safest bet.
Two Audiences on Every Page
The arrival of AI agents does not make human visitors less important. It means every website now serves two audiences at once: people who want a fast, clear and trustworthy experience, and software that wants structured, accurate and accessible information. The sites that perform best over the next few years will be the ones that measure those audiences separately, design for both and stop assuming that every session belongs to a person.
For teams responsible for growth, the immediate steps are practical: segment automated traffic in analytics, check how experiments handle bots, review page structure for machine readability and watch the emerging standards for agent verification. The visitors are changing, and the way websites are measured has to change with them.
Related articles
- BragJack Attack Hijacks AI Browser Agents in 5 Browsers [2026]
- Agentic AI in Enterprise 2026: Inside the $9 Billion Market Reshaping How Businesses Operate
- Wix vs Squarespace 2026: 2x Lighthouse Gap and 900 vs 150 Templates [Tested]
