For a demand gen team, “where does our audience look for answers?” is not a philosophical question. It’s budget. It’s headcount. It’s what gets built next.
So when SparkToro said it could show how much an audience uses Google vs. ChatGPT (and Perplexity vs. Yahoo, Claude vs. Bing, and plenty more), that landed as a practical promise: finally, a clean way to see whether discovery is moving from search to AI.
In March 2026, SparkToro effectively admitted the first version of that promise was… messy. Not malicious. Just methodologically wrong for the question most marketers thought they were asking.
The upgrade wasn’t a new chart. It was a new definition of “use.”
Rand Fishkin explained that SparkToro’s prior reporting relied on clickstream visitation—where audiences went—rather than where they actually searched or prompted. Those aren’t the same thing. (Source: SparkToro post by Rand Fishkin, March 8, 2026.)
And the difference isn’t academic. Fishkin pointed to SparkToro’s research with Datos showing that while 97% of visitors to Google search, only 56% of visitors to ChatGPT and 35% of visitors to Yahoo do. (Source: SparkToro post by Rand Fishkin, March 8, 2026.)
That’s the pattern interrupt. The “AI vs. Google” story most teams are telling themselves is about behavior. But a lot of the data floating around the industry is about traffic. Those two can diverge fast.
So SparkToro changed its Search & AI Tools graph to be search/prompt-centric rather than visit-centric, specifically because visits include plenty of non-search activity—like checking API credits or consuming shared links. (Source: Research Brief: SparkToro methodological correction, March 2026.)
Why this matters right now: the verification loop is the new funnel
The timing is sharp because buyer behavior is splitting. Multiple sources summarized in the Research Brief suggest that 37% of consumers now start searches with AI tools instead of Google—up from negligible levels two years earlier—while Google still holds 78–93% global market share of digital queries depending on the source. (Source: Research Brief statistics.)
That looks contradictory until the second piece clicks into place: AI often complements search instead of replacing it. The Research Brief cites that 85% of AI users verify answers elsewhere, often on Google, and only 6% report high trust in AI summaries. (Source: Research Brief statistics.)
In other words, a lot of “AI discovery” doesn’t end in an AI-only decision. It kicks off a loop: prompt, skim, doubt, verify, compare, repeat. Demand gen teams feel that loop in pipeline conversations even when analytics doesn’t show it cleanly.
And there’s one more pressure point. Zero-click search is already normal: the Research Brief cites 58–60% of Google searches ending without a click, with a figure of 83% cited in connection with Google AI Overviews. (Source: Research Brief statistics.)
Traffic capture is getting harder. Influence is getting easier to miss.
What SparkToro changed besides the methodology (and why it’s a tell)
SparkToro didn’t just fix measurement. It also changed the product surface area around the question “what should a marketer do with this?”
Early in 2026, SparkToro upgraded keyword reporting to prioritize relevance over comprehensiveness, filtering out off-topic terms that aren’t useful for marketing decisions. (Source: Research Brief: SparkToro keyword reporting upgrade.) That’s a quiet philosophical shift: better to show fewer terms that drive action than to flood a report with noise that looks impressive.
Then it removed underused features—SERP features, trending keywords, related questions—because fewer than 1 in 1,000 users accessed them. (Source: Research Brief: SparkToro removed low-usage features.) That’s not a feature cut. It’s an editorial stance about what matters in day-to-day work.
Most important for 2026 behavior: SparkToro now presents traditional search keywords and AI prompt data in a unified view, covering what audiences search on Google/Bing and what they prompt tools like ChatGPT, Claude, and Perplexity to do. (Source: Research Brief: SparkToro unified view.)
That unification matters because it matches how research actually happens now. A buyer might Google a category term, then ask an AI tool to compare options, then return to search to validate a claim. Separating “SEO” from “AI visibility” creates organizational comfort. It doesn’t reflect the journey.
A practical read for demand gen: stop planning around visits, start planning around intent
Here’s the uncomfortable part: lots of teams have been using “visits to AI tools” as a proxy for “AI-first intent.” SparkToro’s update is a reminder that proxies can lie.
Fishkin’s explanation for ChatGPT is especially clarifying: many people experience AI tools through shared prompt/response URLs rather than prompting themselves. (Source: SparkToro post by Rand Fishkin, March 8, 2026.) That means a clickstream panel might record a “ChatGPT visit” even when the user performed zero prompting behavior—no question, no intent signal, no query to map to messaging.
For Yahoo and other properties with content attached, the same confusion shows up in reverse. People visit for weather, sports, finance, or news; search is only a slice of the session. (Source: SparkToro post by Rand Fishkin, March 8, 2026.) Counting visits as searches inflates the importance of the search box.
So what changes in execution?
First, audience research needs a new baseline: search share and prompt share are the numbers that should drive content bets, not raw platform visitation. If the question is “where is demand forming?”, only search/prompt behavior gets close to an answer.
Second, keyword strategy can’t stay keyword-only. The Research Brief describes a world where AI handles complex conversational queries efficiently, while Google remains dominant in volume—and many users combine both. (Source: Research Brief expert perspectives.) A content plan that ignores prompts misses the language buyers use when they ask for synthesis, comparison, and evaluation.
Third, measurement needs to accept that clicks are no longer the only proof of impact. When 58–60% of searches end without a click—and higher rates are being cited in the AI Overview era—demand gen teams have to get comfortable measuring visibility and downstream outcomes, not just sessions. (Source: Research Brief statistics.)
The short version: the funnel didn’t disappear. It just stopped being linear.
SparkToro’s upgrade is valuable because it reduces one category of self-deception. It forces a cleaner read on what audiences do—not merely where they show up. In 2026, with AI-first habits rising and trust still lagging, that distinction is the difference between chasing noise and tracking intent.
