Kevin Indig did a great roundup of AI Chatbot referrals over at Search Engine Journal. The data in his article comes from Similarweb, which is sort of like Nielsen for websites. Calling out some of the highlights here with thoughts.
AI Traffic is Insignificant as a share of organic visits.
Kevin reaffirmed this claim and referranced an article from December, where he noted that AI chatbots represent less than 1% of organic traffic (and remember, organic traffic is less than half of total traffic for the typical commercial website). This lines up with what I wrote recently about not getting swept up in the AI referral growth hype.
AI Chatbots Prefer the Same Sites Traditional Search Favors
If you were hoping a shift to chatbots for search was going to disrupt giant companies that are entrenched in traditional search, don't hold your breath.
Image credit: Kevin Indig - Published at https://www.searchenginejournal.com/transactional-ai-traffic/539548/
This is no surprise. These brands appear in the search results that some of these models use for building out their own results. They are frequently mentioned in training data. And to the extent user behavior may influence output, a lot of users will click to amazon/walmart/etc.
AI Chatbot Traffic MAY be More Engaged.
I am quite skeptical. Kevin posts some stats that are pretty clear.
The trend is crystal clear: AI chatbot referrals stay more than a minute longer than Google traffic.
The front runner is Copilot, with an average of 11.6 minutes per session. On average, AI chatbot traffic stays for 10.4 minutes, compared to Google’s 8.1 minutes.
He also makes note about how this reflects number of pages visits.
AI chatbot referrals visit 12.4 pages on average, compared to Google’s 11.8.
Again, more pageviews within the same session correlate with a higher chance of conversion when looking at transactional traffic, so this means AI chatbot traffic has a higher quality.
That last sentence is a pretty bold assertion. Especially when you consider my next takeaway.
AI Chatbots Send Traffic to Home Pages More Frequently
Here Kevin makes the following observation:
Folder depth for referral traffic varies significantly between the difference referral sources, even among AI chatbots.
For example, ChatGPT sends over 50% more traffic to homepages than Copilot or Perplexity and over three times more than Google Search.
...
On average, AI chatbots send more traffic to the homepage than Google, which sends more traffic to the second and third folder levels.
This should create some pretty obvioius disonnance when considered against the previous point. If you are performing a transactional/product search and are sent to the home page of a site, you will need to spend more time and visit more pages in order to complete your shopping mission. Google more often sending people deeper into a website means you can more efficiently achieve your goal.
Without the inclusion of conversion data (did the user ultimately purchase), it is reckless to speculate that AI chatbot referrals are of "higher quality."
This is somewhat addressed later.
The data for Google Search raises the question of whether AI chatbot traffic would engage even better if less went to the homepage.
Since AI chatbots understand context so much better than (Google) Search, at least outside of AI Overviews, sending users to more specific pages should be a better experience.
The reason this isn’t happening yet could be that AI chatbots haven’t crawled enough of the web yet and that search functionality is still very nascent.
The dissonance with the engagement point isn't really addressed and making that point in the first place still strikes me as a massive leap. But some interesting points here nonetheless.
Briefly on the statement about how "AI chatbots understand context so much better than (Google) Search)," I'd love to know what the rationalization is for this. From reading, it appears this is not actually based on "understanding" (let's just pretend these machines actually understand anything, for the sake of making this easier to follow) but on the additional context input by the user. People type longer explanations and questions into chatbot prompts compared to the short phrases and single words largely input into a traditional search prompt. E.g. a typical use case would be going to Google and typing "mens neutral winter running shoes". Where a prompt to ChatGPT might be "what are good running shoes for neutral pronation that offer traction for running in Wisconsin during winter?". You can't declare anything about the ability of a tool to "understand" context unless you have equal context for the inputs from users.
For reference, here's how ChatGPT 4o handles those two queries. The more context in the second query offers results that are more appropriate for the conditions (or you could argue it at least emphasizes important details more to help a user make a decision on what to click). So even with the allegedly superior "understanding" of AI, more context on input makes a big difference.
(okay that was not as brief as I intended, but it hit on a common frustration of mine with new technology)
If you buy the assumption that AI bots haven't crawled as much of the web as Google (this is likely true), and you presume that is a negative, then at least in the case of Copilot and ChatGPT it's curious that they haven't found ways to better leverage Bing results as part of responses.
But I would propose that, at least for now, the reason AI Chatbots are not sending people deeper into websites is less about capability and more about design. Outside of individual product links (which are increasingly common), LLM responses still largely treat website links as a shorthand way of citing sources. The intent is not to send people away from the app and more about showing where they got information and giving people some options to explore if they really want to. Conversely, the original purpose of Google (and all traditional search engines) was to send people to websites. Sure, motivations have changed, but that is still a core concept in the search engine experience.
As always, try to avoid getting too swept up by the hype.
