AI Referral Traffic is Unpredictable. Here’s Why

Marketers being asked by execs how their company is “performing in AI” is the new “what’s our ROAS”. A simple question, a not so simple answer. While people continue to adopt alternative/AI search engines, we need to get to the bottom of how AI citations influence overall website AI referral traffic and more importantly how we can optimize it.

The most important thing at Simple Search is having reliable data to inform decision making. Without properly understanding sources of traffic and related behavior, it’s hard to make those informed decisions.

Capturing AI Referral Traffic in GA4 (and other Analytics)

In 2024, we began using regular expressions in GA4 to capture the traffic that AI-powered conversational engines referred to client sites.  In May 2026, GA4* made it easier to track these sources as a channel “AI Assistant”.  Great, right?  Not so fast.

 Analytics platforms themselves don’t give any picture of why fluctuations happen. They show us cross-platform performance, but not how/why different modalities retrieve information. When these platforms are always evolving and updating – especially displaying (or not) and changing placement and presentation of citations – fluctuations are bound to happen. 

Here’s a timeline of the AI answer engines we consistently see delivering the most traffic in our clients’ analytics:

Late 2022: Perplexity launched as as a citation-first answer engine with links to cited websites. It was first built as a conversational search engine with inline and right hand side citation listing mandatory for virtually every single search (though not all).  This continues to cite at least 3 or 4 prominent links per query.

Important to note – the 3 engines below DON’T always cite results, but when they do…..

Early 2024: Google Gemini embeds citations inline or relies on underlying Google Search grounding, favoring brand-owned and structured schema domains.  It integrates deeply with Google’s search index to attach rich source links and media references (like YouTube).

Late 2024: ChatGPT added more web search citations in through ChatGPT Search, embedding inline links.  It generates the highest absolute count of reference links when search is triggered, heavily favoring encyclopedic and structured media domains. 

Spring 2025:  Anthropic adds a web search feature to Claude, where before it relied on its own training data.  Now you can see it building an answer with the urls it’s referencing before they are embedded in results.  Again, not for all queries.

What Primary Audience Does Each AI Platform Attract?

Perplexity: Researchers, market analysts, journalists, and professionals doing continuous fact-checking.

Gemini: Gemini is deeply embedded into the Google Workspace ecosystem making it easy for everyday users to naturally lean into Gemini as it’s integrated into their native platform (raises hand – love it when that little diamond pops up…..sometimes). Android users are also a big segment as the Chrome app is the go to app for Search.  

ChatGPT: Three-quarters of conversations focus on practical guidance, seeking information, and writing, with writing being the most common work task. This encompasses a massive audience ranging from high school students writing research reports to CEO’s sending company wide emails.

Claude: An emerging source in analytics data!  While it isn’t usually in our top 3 referring sources, that doesn’t mean it’s not doing heavy lifting. With an audience that is dev/engineering heavy, it still cites other urls.  Thus far, it appears that people are building and doing more in Claude vs. researching. 

How Many Users Does Each AI Platform Have?

To understand demand and potential audience size, it’s important to know what the current TAM of each one is. Here’s a timeline of recent milestones hit by the core platforms in 2026.

June 2026: ChatGPT hits 1 billion monthly active users. 

June 2026: 245 million people worldwide were using the Claude AI chatbot 

July 2026:  Perplexity maintains growth in users after offering premium free to some users in India to average about 14 million/month.

August 2026: Gemini app hits 1 billion monthly active users.

Ok, back to the main theme!

Why does AI Referral Traffic fluctuate so much?

Well, ChatGPT makes up 92% of AI referral traffic!  Even minor changes in how your website appears in Chat dramatically skew this traffic. For example, CTR changes based on chips vs. in-line citations. Chips = lower CTR.  When ChatGPT shifted to inline links in May 2026, especially for brands, we (and many others) saw traffic increases from ChatGPT for many clients, especially those with strong brand presence.

What can I do to optimize for AI Performance?

AEO is still a query game, although visibility is limited in terms of what people are actually querying (AND we are at the whims of how the engines link or don’t back to websites in general).  Standard SEO-keyword tools are still good directionally.  Shout out to Bing Webmaster Tools for giving us SOME insight into queries/themes/etc.

Simply put, the engines are more dynamic and follow up questions are the norm, so mapping where the engines are pulling information for fan out queries is critical. 

Traditional SEO:
enterprise SEO consulting – optimize a page for that keyword.

Next Level SEO/AEO: Fan-out analysis:
enterprise SEO consulting  – determine all the searches an AI might need to perform before it can confidently recommend a company.

1. Start with 20 ish priority prompts that represent questions you actually want the brand to appear for, like: What are the best companies for enterprise SEO consulting?
2. Then, for each prompt, generate fan-outs across several predictable “types” like the following:

3. Reverse engineer! Where are the engines pulling urls cited? How can you influence? 

Reverse engineering is always a fine idea – figuring out what can be accomplished on-site vs. 3rd party sites is the first step.  Step 2 is assembling the right people to cultivate and monitor. Step 3: roll up your sleeves, traditional SEO just got a lot more exciting and will rely even more on partnerships within your organization. 

Will AI Referral Traffic Grow or Decline?

Who knows what the next algorithm will bring and what sources it will favor. At least for now, marketers are inching towards better attribution and optimization for AI referral traffic. Make sure adopting best practices for LLM optimizations are routinely checked and updated. If you are looking for a partner who knows the space, go ahead, hit the contact button and let’s get the conversation started. 

Contact Us

*sidenote – data degradation in GA4 from CMP is a whole other topic

*caveat – this treats Google AI as a separate entity so we haven’t included it in this post. That’s for later!