SparkToro Says Google Has 73% of Search And I Have Questions (Part 1)

Rand Fishkin and SparkToro published research in March 2026 that reframed how we think about search on the internet. Their headline finding: Google is responsible for 73.7% of all desktop web searches across 41 major domains in the US. Not 90%. Not 85%. 73.7%.

That number immediately made the rounds. SEO Twitter (or whatever we call it now that Threads apparently has more daily mobile users) lit up. "See, Google isn't as dominant as you thought!" and "Time to optimise for Amazon and YouTube!" were the predictable reactions.

I will take whatever data someone tells you is the definitive picture of search behaviour with a bucketful of salt. And this research, despite being genuinely interesting, deserves that same treatment.

The research results that Rand Fishkin published is a long one but very interesting so I've decided to split my thoughts on it into three separate blog posts (which makes it easier for me to churn and I think so will my readers). This will be part 1 of the series.

Here goes.

What SparkToro Actually Did

The methodology is straightforward on the surface. SparkToro partnered with Datos (a Semrush company) and analysed clickstream data from millions of US and EU/UK desktop devices across 41 websites. They counted searches and AI prompts on each domain, then calculated market share. They then broke down the 41 sites into five categories: traditional search (80%), commerce (10%), social (5.5%), AI tools (3.2%), and other verticals.

The visualisations are genuinely impressive. They tracked quarterly trends, searches per searcher, and even the percentage of visitors who actually perform a search versus those who just browse. That last metric produced one of the more interesting findings: only about half of ChatGPT's desktop visitors actually enter prompts. Many are viewing shared conversations or checking their account.

But impressive visualisations do not automatically mean the underlying measurements are sound. The domains were "editorially chosen by Rand" from the top 250 most-visited sites, narrowed down to those with "relevant search behavior." That's the first thing that caught my eye.

Three Things That Give Me Pause

1. The domains were hand-picked.

Forty-one websites chosen by one person based on his judgement of what constitutes "relevant search behavior." The article acknowledges this openly, which I respect. But editorial selection is a significant methodological choice. Different selections would produce different market shares.

What if you included more niche e-commerce platforms? More regional search engines? More vertical-specific sites where professionals search daily? The 73.7% figure is a function of which 41 sites you put Google up against. Change the denominator and the percentage changes.

2. Desktop only, in 2026.

This is the caveat that should be in the headline but never is. A huge percentage of search happens on mobile browsers and apps, and mobile was not included. SparkToro acknowledges this, but the nuance gets lost the moment anyone shares the "73.7%" figure without context.

We know from our own research and from what we covered in our article on the evolution of search engines that mobile-first indexing and mobile search behaviour have been dominant for years. Google itself switched to mobile-first indexing as the default back in 2020. A desktop-only study in 2026 is measuring the smaller half of the picture.

Consider also that mobile search behaviour skews differently across platforms. Social search (Instagram, TikTok) is overwhelmingly mobile. E-commerce search through apps (not browsers) would not show up in this panel at all. AI tools like ChatGPT have dedicated mobile apps with significant usage. The desktop-only limitation does not just reduce the sample size; it systematically biases the results toward platforms and behaviours that are more desktop-centric.

3. Not all "searches" are the same thing.

This is the big one. SparkToro counted an Amazon product search, a YouTube video search, an Instagram search for a person, a Google informational query, and a ChatGPT prompt all as equivalent "searches." The article states this explicitly: "This report treats all searches and prompt sessions equally."

But are they equivalent? When someone searches "blue running shoes size 10" on Amazon, they are in a completely different mode than someone searching "why does my knee hurt when I run" on Google or asking ChatGPT "write me a training plan for a 5K." These are fundamentally different behaviours with different intents, different outcomes, and different implications for marketers.

I wrote about this distinction between search intent types in our piece on understanding search intent. Lumping them all together creates a number, but it's not clear what that number means for anyone trying to make practical decisions.

Analyzing The Claims

I went through SparkToro's key claims with the help of kimi-k2-thinking and fabric and these are some of what I found:

The 73.7% figure cannot be independently verified. It comes from Datos' proprietary panel data. Other market share reports from Statcounter and Comscore consistently put Google at 85-90% for traditional search. The difference is not because SparkToro is wrong. It is because they are measuring something different (all platform searches vs. search engine searches) and presenting it as though it replaces the traditional metric.

The claim that Google lost 3.5 percentage points of desktop search share during 2025 would be historically significant. Yet I think no other major analytics firm has reported such a drop (I could be wrong, of course). Single-source findings like this need corroboration before they should significantly change anyone's strategy.

The one claim that holds up well is the roughly 16% of Google SERPs showing AI Overviews. Semrush's own study of 1.5 million keywords found 16.2% in Q4 2025, and Search Engine Land reported similar numbers. That said, even this figure fluctuates significantly month to month, as I noted in Part 1 of our AJSA seminar report when discussing how Google's AI features are still experimental and volatile.

Why This Matters For Your Business

Just to be clear, none of this means SparkToro's research is useless. The underlying thesis, that search behaviour is broader than just Google, is correct and has been for years. I have been writing about this at kafkai.ai for a while now, from how generative AI is changing search to why the zero-click era demands a multi-channel approach.

But there is a difference between "search behaviour is diversifying" (true, and something you should pay attention to) and "Google only has 73.7% of search" (a number that depends entirely on how you define search and which platforms you include).

If you are a small business owner reading SparkToro's research and wondering whether to slash your Google SEO budget and pivot everything to Amazon and YouTube optimization, slow down. The research does not support that conclusion, and SparkToro is not actually saying you should do that either.

What you should do instead:

1. Understand where your specific audience searches. Not all audiences are the same. A B2B SaaS company and a fashion brand have completely different search footprints. As we covered in our article on AI curating buying decisions, the question is not "where does search happen globally" but "where do my potential customers look for information before making a decision." That also means understanding where your competitors are visible and where they are not. Competitive gap analysis (finding the keywords and channels your rivals have overlooked) is exactly the kind of specificity that turns generic data into actionable strategy, and it is what we built Kafkai to surface.

2. Read the methodology before sharing the headline. This applies to SparkToro's research and to every other piece of research you encounter. Desktop-only, 41 editorially chosen sites, proprietary panel data. These are not disqualifying factors, but they are important context.

3. Don't confuse market share with relevance. Even if Google "only" has 73.7% of desktop search (under SparkToro's definition), that is still an overwhelmingly dominant position. Diversifying your channels makes sense, but abandoning Google because the number went from 90% to 73% under a different measurement would be absurd.

A Note On Rand Fishkin And SparkToro

I want to be clear about something. I have a lot of respect for Rand Fishkin and what he does. He has been one of the most transparent voices in the SEO and marketing industry for over a decade, first at Moz and now at SparkToro. When he presents data, he includes the caveats, acknowledges its limitations and invites scrutiny. That is rare and admirable.

The fact that I am poking holes in the methodology does not mean the research is bad. It means the research is being used, as research often is, for conclusions it does not fully support. Rand's own conclusions in the article are actually quite measured. He says search is diversifying, that Google is losing share at the margins, and that marketers should pay attention to a wider set of platforms. All of that is reasonable.

The problem is what happens after publication. The headline gets shared and the caveats get dropped. "Google only has 73% of search!" becomes the takeaway, stripped of every qualifier that makes the number meaningful. That is not Rand's fault, but it is worth correcting.

The Real Takeaway

SparkToro's research is a useful conversation starter, not a strategic blueprint. The data is proprietary, the methodology has significant choices baked in, and the headline number is designed to be provocative (which, in fairness, is exactly what good marketing research should be).

The underlying message, that search is a behaviour that happens across many platforms and not just in a Google search box, is valid. But we have known this for years. The question has always been: what do you do about it?

In the next article in this series, I will dig into why treating an Amazon product search and a Google informational query as the same thing creates real problems for marketers trying to make sense of this data.