Blog | FELD M

AI agents and web analytics: New rules for attribution | FELD M

Written by Dr. Ramona Casasola-Greiner, Dr. Matthias Böck | Nov 19, 2026, 9:54:37 AM

AI traffic series - Part 5 of 5 

In this final instalment of our five-part AI bot traffic series, we'll be looking at how bot traffic has changed the game for attribution.

Why's that, you ask? Because AI agents behave differently from traditional crawlers. We're specifically talking here about AI-powered assistants like ChatGPT with browsing mode, Bing Chat (Operator), Perplexity, or specialized GPTs with tools.

Traditional web crawlers, such as search engine bots like GoogleBot, systematically scan pages for indexing and follow predefined rules.

AI agents, on the other hand, respond to users' specific requests and goals. Rather than systematically crawling the web, they access and retrieve relevant pages as needed, often using a browser to navigate sites in real time.

Series overview

Need a recap? Here's an overview of what we've covered so far in this series:

What's the difference between AI agents and traditional bots and crawlers?

Agentic AI crawlers can behave like human visitors. They can scroll, click links or buttons, and even fill out forms.

One example is OpenAI’s ChatGPT “Operator,” introduced in January 2025. Operator uses a Chromium-based browser to perform tasks on the web. So unlike a typical bot, it doesn't just request static URLs, but can instead independently search the web and click on individual search results.

Depending on the agent's instruction, this traffic might show up differently in your analytics. If an agent is asked to open a specific page directly, your analytics might register it as direct traffic. But let's say it's asked to “find” a company’s website, it may start a Bing search and follow an organic result or even an ad. The visit is then recorded as Bing/organic or Bing/cpc.

While traditional crawlers can often be identified through distinctive User-Agents or IP ranges and usually don't trigger JavaScript-based analytics, AI agents are designed to appear as human-like as possible. They fully load pages, including scripts, meaning their activity can show up as regular sessions in web analytics.

AI agents also operate contextually. Rather than crawling pages indiscriminately, they retrieve specific content in response to complex inputs and carry out tasks on behalf of users.

This intent-driven behavior, combined with human-like technical interaction, makes AI agents a new category of web traffic that is rapidly gaining importance.

Current data shows a sharp increase in these types of requests. Within just 30 days, more than 976 million requests from OpenAI-based crawlers were recorded, 92% of them generated by ChatGPT. That already represents 4.5% of all “good” bot traffic.

The line between human and automated traffic is beginning to blur as AI agents retrieve content in real time and support users throughout the decision-making process.

 

So what does all that mean for web analytics and traffic attribution? 

The growth of AI agents creates new challenges for web analytics and attribution. One of the biggest challenges is referrer masking. Many AI-powered tools don't pass a Referrer Header when sending users to websites.

Tests showed that only a few – around six – AI platforms were recognized as such and listed under “Referrals” in Google Analytics. Most AI-driven sessions instead appeared as direct traffic or “unassigned.” This makes source attribution considerably more difficult.

Another problem is that, unlike traditional search, there's no information about user intent, in this case meaning the prompt that triggered the visit. In the past, SEO analysts could analyze the keywords users searched for. With AI agents, the question or task the AI assistant was handling remains hidden.

This “black box” makes it difficult to reverse engineer and connect content with user needs. And nor do standard tracking mechanisms like UTM parameters solve the problem, because AI systems don't automatically append UTM tags to links. As a result, AI recommendations cannot easily be reported like campaigns or channels.

AI agents can also skew your traffic metrics. Because their activity can resemble human browsing, it may slip past bot filters and get mixed in with data from genuine visitors.

For example, an AI agent acting on behalf of a user might click through several pages on a website. In your analytics tool, it might pass as an unusually engaged visitor.  This can lead to inflated engagement metrics: artificially low bounce rates, high page views, and long session durations can distort the overall picture. AI agents can also create incomplete funnels, particularly when they analyze content without going through a complete user journey.

For example, an AI assistant might present a product recommendation directly to a user, who then clicks once and goes straight to checkout. From a web-tracking perspective, this looks like a sudden purchase with no preceding journey on the site.

The same applies when AI recommendations later lead to an action, but there is no session connection. According to Adobe Research, 34% of consumers use generative AI for product research and then switch to traditional search or go directly to a website to complete their purchase.

These attribution gaps mean that AI’s initial influence can remain invisible in your analytics. Conversions may appear as direct or organic search, even though AI was the original trigger.

Overall, standard analytics reports often fail to capture the origin of AI traffic accurately. Without adjustments, it becomes difficult for analysts to understand which visits were influenced by AI recommendations and to measure and optimize this emerging channel.

Does your users suddenly land on your website with no clear path, or show up only as direct traffic?

We can help you better understand this new user behavior and show you how to make it visible and actionable in your analytics strategy.

68% of searches end without a click: AI assistants are changing user behavior

The growing use of AI agents and generative search assistants is noticeably changing how users behave on the web. One key phenomenon is the rise of “zero-click” journeys, where users get the answer they need without visiting an external website.

AI-powered search experiences, such as Google’s generative AI Overviews or Bing Chat, provide answers directly within the search experience, shortening or completely replacing traditional click paths. According to recent studies, around 80% of consumers rely on these zero-click results for at least 40% of their searches. As a result, organic traffic to websites is estimated to have declined by 15–25%.

Overall, around 68% of all searches now end without users visiting an external website, because they've already been provided with the answer they needed within the AI-powered search experience.

Even skeptical users are adopting this convenience. Dedicated AI search assistants such as ChatGPT, with 2.5 billion prompts per day in 2026, and Perplexity, with 45 million monthly users, are rapidly gaining popularity.

For website owners, this represents a major shift in the customer journey. Many users are skipping the early stages of information gathering on company websites because AI assistants already provide a shortlist or summary.

We've seen this shift in our own inbound leads, too. Part of the awareness stage now happens within AI-generated answers, where users may discover a brand through an assistant's mention or recommendation without immediately visiting its website.

When they eventually click through to the company's website, they are often further down the funnel and highly qualified. AI recommendations act as a filter. Users who reach your website through ChatGPT and similar tools are generally already convinced that your content is relevant. And it's reflected in their behavior: AI-referred visitors tend to have significantly lower bounce rates and longer sessions than average. Adobe data, for example, showed that by February 2025, AI referral traffic had a 23% lower bounce rate than other traffic and users stayed on the site 41% longer.

The gap is narrowing on the conversion side as well. In mid-2024, AI traffic was still 43% less likely to convert than regular traffic. According to HubSpot, however, 11.4% of AEO visitors now become customers, compared with only 5.3% of traditional SEO visitors.

In other words, AI-prequalified users are rapidly catching up when it comes to purchases. In some research-intensive categories, such as electronics and travel, AI-referred visitors are already outperforming traditional visitors in engagement and revenue per visit.

 

AI's influence remains invisible in traditional attribution models

However, this also creates an attribution challenge.  If AI assistants influence the decision-making process in the background, but the user ultimately converts through a traditional channel, that AI influence remains invisible in existing attribution models.

Companies therefore need to rethink how they connect awareness and conversion. Ultimately, the entire journey is becoming more compressed and increasingly influenced by sources external to the company’s own website. Experts predict that traditional SEO traffic may decline as generative AI platforms become the new home of information discovery.

Marketing teams need to prepare for the fact that the rules of the digital customer journey are changing: From visibility in AI answers, also known as Generative Engine Optimization (GEO), to measuring the success of touchpoints that happen outside their own web properties.

 

How can AI-driven traffic actually be identified and attributed? 

 

TL;DR:

  • Referral URLs containing LLM domains
  • Grouping known AI referrals into a dedicated channel group using custom filters or Regex rules such as chatgpt\.com|perplexity\.ai|gemini\.google\.com|...,
  • Monitoring indirect indicators such as a sudden increase in Bing Organic Search traffic
  • Unusual behavioral patterns, e.g. extremely short interactions followed by an immediate conversion, or Chrome versions that are rarely seen among your regular users.

More detailed explanation: 

The first step is to identify visible traces of AI agents within analytics systems. Some visits can be recognized through their source URLs. For example, referrals from ChatGPT’s browsing mode may appear as chatgpt.com. Similarly, domains such as perplexity.ai, bing.com for Bing Chat, or gemini.google.com can provide clues.

In Google Analytics 4, it makes sense to group known AI referrers into a dedicated channel group. Using custom filters or Regex rules, for example:


 

chatgpt\.com|perplexity\.ai|gemini\.google\.com|...,

 

For platforms that provide referrer data, common AI sources can be classified as “AI traffic.”

A significant share of AI-assisted visits still appear as direct traffic. That makes it important to monitor indirect indicators as well. For example, a sudden increase in Bing Organic Traffic could indicate that AI agents such as ChatGPT Operator are increasingly finding your website through Bing Search.

Unusual behavioral patterns can also point to AI-agent visits, e.g. extremely short interactions followed by an immediate conversion, or Chrome versions that are rarely seen among your regular users.

 

New solutions are still needed

Specialized bot-management services are already developing fingerprinting methods to identify AI agents more reliably.

One security provider, for example, was able to develop a distinct signature for ChatGPT Operator traffic by combining technical characteristics such as Linux as the operating system and a characteristic Chrome 130.0 User-Agent with recurring behavioral patterns.

This example highlights the need for new solutions. Traditional defenses such as CAPTCHA or user agent blocking are increasingly ineffective against well-disguised AI agents.

For analytics teams, the challenge is now to measure AI traffic without mixing it with genuine user behavior.

In the short term, pragmatic approaches can help:

  • Create dedicated segments and reports for suspected AI referral traffic.
  • Compare analytics data with server logs.
  • Use opt-out or robots directives for known AI crawlers where available.

In the long term, new industry standards will need to be developed. Experts are already discussing new protocols like llms.txt or ai.txt, similar to robots.txt or sitemaps, that specifically address AI agents. Until such standards are widespread, companies should closely monitor how AI traffic shares evolve and establish internal guidelines for segmenting and analyzing these visits.

It's also worth actively correcting KPI distortions caused by AI agents, for example by filtering known AI traffic out of funnel analyses or creating separate benchmarks for AI-driven users. You can also consider integrating new tools that provide AI traffic analytics and make the performance of this visitor group visible.

Ultimately, web analytics needs to evolve to distinguish between humans, traditional bots, and AI agents. Only then can digital strategists and analytics leads maintain a clear picture as AI agents become an increasingly important source of website traffic in the months and years to come.

If you start collecting and understanding reliable data about this phenomenon early, you'll gain a clear advantage. You'll be able to adapt your marketing and customer experience strategies before AI agents evolve from an exception into the mainstream.

 Previous articles: