ChatGPT or Google: Who’s the Real Click Killer

Ever since AI search became mainstream, publishers have been stuck asking the same question:

“What’s killing my clicks?”

Naturally, many blamed the usual suspect – the one engine that made AI search SEO an industry virtually overnight: ChatGPT. Others pointed to the company that has been defining web search for decades now: Google. 

Similarweb’s August 2026 report, however, seems to absolve both by offering a third answer: nothing has really changed – users are still using Google, they’re just adding AI on top of it.

Yet, clicks are still disappearing.

What did Similarweb’s research find?

In August 2026, Similarweb released a 2026 Generative AI Landscape summary post with a headline finding that seemingly kills the “AI is eating search” narrative: 95% of ChatGPT users also use Google, and that number hasn’t moved between September 2025 and May 2026 snapshots.

The finding has been framed and accepted as evidence that users are layering AI search atop traditional search, not replacing the latter altogether. However, the framing is doing more work than underlying data supports and, worse, shifts the attention away from the real threat.

Problem 1: What the data can’t see

When you exclude the population most likely to demonstrate an effect, you cannot demonstrate that effect. 

Similarweb’s methodology disclosure states that “This report does not contain gen AI API usage, desktop applications, or in-app AI features within mobile and web-applications,” and further hedges the figures as “estimations and extrapolations… not warranted for accuracy or completeness.”

Effectively, that framing excludes categories where heaviest professional AI use happens:

• Developers calling the AI APIs directly;
• ChatGPT and Claude desktop app users (i.e., not the browser version);
• AI coding tool users (Cursor, Copilot, Windsurf, Zed);
• Users relying on in-app AI features (Notion, Slack, Gmail, Linear, Superhuman);
• Employees using AI integrated into internal company tools.

That’s essentially every category of the population most likely to have substituted traditional search for AI search, which leaves only one subset: browser-based casual users – exactly the category least likely to substitute anything, because the only AI they touch is chatgpt.com in a tab next to google.com.

Problem 2: The overlap metric does not detect substitution

The “overlap” the report leads with is a binary measurement. It shows whether a user visited google.com at least once over a three-month window (between March and May 2026), but gives nothing about how often they visited.

As we noted in an earlier article, AI visibility metrics must be treated as probability distributions with wide error margins, and the same principles can (and should) apply here: a single vendor’s snapshot on a metric this loose is barely a floor, let alone a trend line.

A counterweight to Similarweb’s finding can be found in the Bocconi University’s experiment (as pointed out by SEJ), which found that households given access to ChatGPT Search reduced traditional search queries by 9.4%.

That’s a notable drop in a behavioral metric (query volume) that actually points toward substitution, not addition, whereas the Similarweb’s can only detect total abandonment. And let’s be real, almost nobody ever fully abandons Google – AIOs and AI Mode are designed to keep it that way.

Who’s the real click killer, ChatGPT or Google?

Viewed in isolation, Similarweb’s finding reconciles the two contradicting narratives about AI’s impact on search, painting the referral collapse as nobody’s fault – not ChatGPT’s and not Google’s. The problem with that reconciliation is that it does NOT exist in isolation, and the external evidence overwhelmingly points to Google:

• SEJ’s report shows forced AI Mode assignment reduced external clicks by 18.8 percentage points.
• AIOs now appear on ~43% (Similarweb) to ~48% (Semrush) of US searches.
• AI Mode visits more than doubled (from ~120M to ~280M monthly) over the same window Similarweb tracked.
• An independent study found AIOs cut publisher clicks by 39.8%.
• SparkToro’s 2026 zero-click study found that fewer than one in three Google searches now end in a click

Notably, Similarweb’s report documents the AIO/AI Mode expansion, but it doesn’t measure their click impact on organic results at all – and in doing so, it effectively lets Google off the hook for the referral collapse it’s actively driving.

Lily Evans is the Managing Director at ZeroClick Labs, bringing over 8 years of comprehensive experience in SEO, local SEO, and AI optimization to every project. She began her career in content writing, developing a strong sense for search intent and messaging clarity in the digital realm – skills that form an unshakable core of her leadership to this day.