ChatGPT Has 3.2% of Search. That Number Is Both Reassuring and Dangerous (Part 3)

In Part 1 I examined SparkToro's methodology and why the 73.7% headline needs context. In Part 2, I explained why lumping all platform searches together creates a category error. Now let me address the finding that most people in the SEO industry latched onto with visible relief: AI tools represent just 3.2% of desktop search.

When SparkToro's data dropped, I watched a wave of "See, AI is overhyped!" reactions roll through the marketing community. And I understand the appeal of that interpretation. If you have built your career on Google SEO, a data point that says AI search is tiny is comforting.

But comfort and accuracy are not the same thing.

SparkToro found that all AI tools combined (ChatGPT, Claude, Deepseek, CoPilot, Gemini, and others) accounted for roughly 3.2% of desktop search activity across their 41 analysed domains. ChatGPT alone was smaller than Amazon, Bing, and YouTube in search volume.

I wrote about similar numbers in my article on the Google Zero myth back in October 2025. ChatGPT had roughly 51 million visits compared to Google's 3.4 billion. The scale difference was enormous. AI referrals had reached 1.1 billion total, but Google still generated 191 billion referral visits. The maths was clear: AI search was a rounding error compared to Google's volume.

And that is still true today. SparkToro's 3.2% confirms what the traffic data already showed. On pure volume, AI search is small.

So why am I not reassured?

Volume Is The Wrong Metric

Here is an analogy I keep coming back to when I think about this.

When I was living in Japan and commuting on JR trains, I would sometimes watch the shinkansen pass through a local station at full speed. The local train I was standing on had more passengers at that particular station. More people getting on and off. More "traffic." But the shinkansen was clearly the more consequential piece of infrastructure. It connected entire cities. It changed economic geography. The local train served a neighbourhood.

Measuring AI search by current search volume is like counting passengers at a local station and concluding the shinkansen does not matter.

The metrics that actually matter for understanding AI search's impact are different:

Conversion quality. In our research on AI curating buying decisions, we cited Adobe Digital Insights data showing AI-referred traffic to retail sites grew 4,700% year-over-year, with 10% higher engagement and 32% longer session times. The visitors are fewer, but they convert better, because AI pre-qualifies the match before sending the user anywhere.

Purchase influence. From the same research, LANY's survey found 87.3% of B2B respondents said AI recommendations influenced their contract decisions. And 46.4% contracted a vendor they discovered through AI, a vendor they had not previously considered. That is not a 3.2% kind of influence, even if the raw search volume is small.

The zero-click mechanism. This is the one that should keep you up at night. As I covered in Part 1 of our AJSA seminar report, 60% of Google searches already result in zero clicks. AI Overviews, which appear on roughly 16% of Google SERPs, cut click-through rates in half. The user gets an answer and never visits your site. The "search" still happens on Google (so it counts in Google's market share), but the value extraction has shifted to AI.

This is the point I made in our article on GEO not being optional: Mr. Tsuji's data from Faber Company showed that organic search traffic in Japan had "barely declined." The water level in the river was the same. But the riverbed was shifting course. SparkToro's 3.2% is measuring the water level.

Google's Own AI Is The Biggest AI Search Engine

SparkToro buries what I think is the most important insight in a single sentence: "Most AI Search and AI Answers happen on Google."

If 16% of Google SERPs show AI Overviews, and Google processes roughly 5 trillion searches a year, that means Google's own AI search features handle hundreds of billions of queries annually. That dwarfs every standalone AI tool combined by at least an order of magnitude.

So when SparkToro says AI tools have 3.2% of search, they are counting the standalone AI chatbot interfaces. They are not counting the AI that is already embedded inside Google itself. The AI revolution in search is not primarily happening on ChatGPT. It is happening inside Google, and it is cannibalising Google's own traditional search results.

This is exactly what makes the shift so hard to see in market share data. Google's share of "search" stays high. But the nature of that search, and what it means for websites that depend on organic traffic, is fundamentally changing.

The Half-Of-Visitors Problem

One of SparkToro's more interesting findings was that only about half of ChatGPT's desktop visitors actually enter prompts. The rest are viewing shared conversations, checking API credits, or browsing without searching.

This matters because it means the already small 3.2% AI search share is based on visit data that overstates actual search behaviour. SparkToro acknowledged this and even updated their own product to measure search-specific behaviour rather than visits. Good on them for that.

But it also reveals something about how we measure AI tools that is easy to miss: visits and prompts are not the same thing. A single ChatGPT prompt session can replace what would have been 5-10 Google searches. The user asks a complex question, gets a synthesised answer, asks follow-ups, and resolves their information need in one session. On Google, that same information need would have generated multiple searches across multiple pages.

SparkToro's methodology counts each as "one search." But the information resolution per search is completely different. Three percent of search volume might be resolving a much larger percentage of information needs.

What This Means For Your Strategy

I keep coming back to the practical question, because that is what matters for the people reading kafkai.ai. Here is how I think about the 3.2% number:

1. Do not use it as an excuse to ignore AI search.

The businesses that will benefit most from AI search are the ones that prepare before it becomes the majority channel. First-mover advantage is real. We outlined specific steps in Part 2 of our AJSA seminar report: structured data, FAQ schema, original first-party data, and content that AI systems can extract and cite.

These investments cost relatively little and they improve your traditional SEO at the same time. As we noted in our article on AI and structured data for buying decisions, better information architecture for AI simultaneously improves Google rankings. It is the same discipline applied to two interfaces.

2. Pay attention to AI Overviews, not just ChatGPT.

Most people in the "AI search" conversation focus on ChatGPT. But Google's AI Overviews are the AI search feature that actually touches your organic traffic right now. If you appear in an AI Overview, your click-through rate drops by roughly half compared to a regular search result. If you do not appear, you might as well not exist for that query.

Tracking whether your content gets cited in AI Overviews is a more immediately useful metric than worrying about ChatGPT search share. I covered the practical how-to for this in our article on incorporating AI into your SEO workflow.

3. Focus on being citation-worthy.

This is the thread that runs through everything I have written on this topic. Whether the AI answering the user's question is Google's AI Overview, ChatGPT, Claude, or whatever comes next, the content that gets cited has specific qualities: it contains original data, it is clearly structured, it demonstrates genuine expertise, and it is honest about trade-offs and limitations.

Generic content that repeats what is already common knowledge will not be cited. As we covered in our AJSA seminar report, "if it's common knowledge, AI won't use it because it already has it." Primary, original information is what AI systems need. But you can only produce genuinely distinctive content if you know what already exists in your space. Understanding what competitors have published (and where they have left gaps) is the prerequisite for creating something citation-worthy. That competitive visibility is what Kafkai provides.

4. Diversify, but do not spread thin.

SparkToro's data shows search happens on 41+ platforms. That does not mean you need to be on all of them. Pick the 2-3 platforms where your specific audience searches, go deep, and do those well. For most B2B businesses, that is Google, LinkedIn, and one AI tool. For e-commerce, it is Google, Amazon, and social. For content creators, it is Google, YouTube, and social.

I have argued consistently for channel diversification as protection against algorithmic dependency, and nothing in SparkToro's research changes that view. But diversification means being strategic about where you invest, not trying to be everywhere.

The Real Story SparkToro's Data Tells

If I step back and look at what SparkToro's research actually reveals, it is this: search behaviour in 2025 is more fragmented than the traditional "Google has 90%+ market share" narrative suggests, but far less disrupted by AI than the "Google is dead, ChatGPT is eating everything" narrative claims.

Both narratives are wrong. The truth is in between, and it is more nuanced than either side wants to admit.

Google is still overwhelmingly dominant, but its dominance is being eroded at the edges by platform-specific search (Amazon, YouTube, social) and by its own AI features that reduce the value of organic rankings. AI chatbots are small in volume but disproportionate in influence. And the structural shift from "search and click" to "ask and get an answer" is accelerating regardless of what the market share numbers say.

The businesses that will navigate this well are the ones that focus on what all these platforms and AI systems have in common: they reward content that genuinely helps the person asking the question. That principle has not changed. The interfaces through which the question gets asked have multiplied, but the answer to "what should I create?" remains the same.

Be useful. Be specific. Be honest. And do not let anyone's proprietary data, no matter how nicely graphed, substitute for understanding your own audience.