Latest Case Study: Is AI Agent Traffic Any Good?

AI agent traffic is small in volume but high in quality. Our GA4 data shows ChatGPT-referred visitors hitting a 72% peak engagement rate and the highest time on site of any traffic source, with steady 10 to 12% month-on-month engagement growth since April.

We review this site’s performance every month as part of our ongoing research into AI and AI agents. Because our work sits at the meeting point of search, AI-answer inclusion, and user behavior, we track more than volume. We track quality signals like engagement rate, time on site, and events per session, broken down by traffic source.

Chatbots have now taken a measurable share of traffic that used to come from traditional search engines. That shift made us ask a simple question: are the visitors arriving through AI agents genuinely valuable, or just curious one-click readers who bounce straight away?

How Good Is AI Agent Traffic Quality?

AI agent traffic quality is improving fast enough to surprise even us. When we pulled last month’s GA4 report, ChatGPT had become one of the best-performing traffic sources on the site, not by volume, but by every engagement metric we track. The headline figure was a 72% engagement rate for ChatGPT-referred sessions in the most recent month. A year ago, the same figure was less than half that. We first assumed it was a fluke caused by a low session count, so we pulled the previous three months to check for a pattern.

It was not a fluke. The three-month average settled at 56.87%, still the second-highest engagement rate of any traffic source, and the trend line showed steady 10 to 12% relative growth per month starting in April. The screenshot below covers that three-month window.

In the table, ChatGPT appears at position 7, which is effectively 6th, since position 4 is automated bot traffic. Its 56.87% engagement rate places it second only to Bing. More striking, ChatGPT beats every other source on both average time on site and events per session, which means referred visitors explore multiple pages rather than reading one article and leaving.

For context, overall traffic volume to this site has stayed roughly stable year over year, with a slight upward trend. The engagement jump is therefore not the result of fewer, more targeted sessions replacing a larger low-quality pool. The visitor count stayed similar. The behavior changed. Perplexity also shows up in the data with decent engagement quality, though not at ChatGPT’s level. If you want to understand how AI answer engines decide which sources to cite, our piece on Query Fan-Out and RAG explains the retrieval mechanics behind those decisions.

Are AI Agents Used Differently Than We Think?

High engagement from chatbot traffic is counterintuitive, and that is exactly why it matters. When a user queries Google, Bing, or DuckDuckGo and clicks a result, they are showing clear navigational or informational intent, and that click is a deliberate next step. Chatbot sessions should behave differently in theory: the user asked a question and got an answer inline, so why would they then follow a source link and spend several minutes on the destination page? The data raises three competing ideas, and we cannot yet confirm or rule out any of them.

• Better source surfacing: AI agents may have gotten better at matching cited sources to the user’s real intent, so the people who click through are already a good fit for that content.
• Declining answer quality: chatbot responses may be getting less complete or less reliable in some areas, pushing users to verify claims by visiting the original source.
• AI agents as pre-search filters: users may be using chatbots to skip ad-heavy or SEO-saturated results, arriving at a shortlist of sources they then explore with high intent.

All three ideas fit the observed pattern of low volume, high engagement, and steady improvement over time. Research into information-seeking behavior offers useful framing here. Traditional search satisfies a known-item need, while conversational AI may satisfy a more exploratory need, where the cited source becomes the start of a deeper investigation rather than the end of it.

What Does This Mean for AI Search Optimization?

Targeting AI agents as a traffic channel is worth prioritizing now, because the barrier to visibility is lower than in traditional organic search and the quality of the traffic looks higher. Getting cited by a chatbot does not require a domain authority score in the 80s. It requires content structured clearly enough for a retrieval-augmented generation system to pull out a useful, quotable answer.

This connects directly to Answer Engine Optimization and Generative Engine Optimization, the practice of structuring content so AI systems surface it for relevant prompts. If you are worried that organic search drops are eroding your traffic baseline, AI-agent visibility is one of the more realistic offsets available right now. Our post on post-SEO marketing strategy covers how these channels fit into a broader acquisition model.

A high engagement rate from a low-volume source does not prove that scaling that source will keep the quality. The type of person who follows a cited link from a chatbot may simply be a more deliberate researcher.

It is also worth being clear about what this data does not prove. Selection effects are likely at play, so the next research question is whether this quality holds up at scale, across more sites and longer time periods. For now, the signal is promising and worth tracking, not a guarantee.