How many searches now end without a click.
SparkToro, working with clickstream data, has tracked this measure for several years. In 2024 it found that 60.45% of Google searches in the United States ended without a click to any website. That is the figure most people know.
The same research now puts it at 68.01% for the first four months of 2026. Expressed the way that matters to a business: for every 1,000 US Google searches, the clicks reaching the open web fell from 374 to 276 in two years.
Eight points in two years is the useful part. This is not a number that can be learned once and relied on; it is a moving one, which is why it is worth checking the date on any figure in this field before planning around it.
What an AI Overview costs the top result.
Ahrefs analysed 300,000 keywords, comparing Search Console click-through rates before and after the rollout. Where an AI Overview appears, they measure position one falling from a 3.7% click-through rate to 1.6%. A single headline percentage loses the more useful detail, which is how sharply the effect varies by position.
| Result position | CTR reduction where an AI Overview appears |
|---|---|
| Position 1 | 58% |
| Position 2 | 50.8% |
| Position 10 | 19.4% |
Two things in that table matter. The effect is concentrated at the top, so the sites with the most to lose are the ones that were winning. And Ahrefs' own earlier study, ten months previously, measured the position-one decline at 34.5%. It has roughly doubled inside a year, which is the clearest signal available that this is still moving.
Who is losing the most.
This is the part a small company should read first, and it is usually reported as a single average. Chartbeat data reported by Press Gazette shows Google referral traffic to publishers globally down by roughly a third in the year to November. Split by size, it is not one story.
| Publisher size | Search referral decline |
|---|---|
| Small — under 10,000 daily page views | 60% |
| Medium — up to 100,000 daily page views | 47% |
| Large | 22% |
It is tempting to assume large sites absorb the worst of this while smaller ones stay insulated. On this evidence the gradient runs the other way, and the reason is straightforward: the large sites have direct traffic, apps and subscriber lists to fall back on. The small ones had search.
One caution we would apply to our own argument: this is publisher data, and a publisher’s entire product is informational pages — exactly what an AI answer replaces most cleanly. A company page answering a commercial question about a service is not in the same position. The size gradient is real. Assuming it transfers unchanged to your business is not.
What the conversion figures do and do not show.
The encouraging finding in this area is that visitors arriving from AI assistants convert around 4.4 times better than ordinary organic visitors. That is real research, from Semrush, and it is genuinely good news about the visitors who do arrive. It is not evidence that the lost traffic has been replaced.
It is a rate, and the mechanism its own authors describe explains why: the assistant has already answered the questions, so the people who click are the ones who had already decided. That makes them valuable and few. Ahrefs puts ChatGPT at roughly 12% of Google’s search volume, while the referral traffic it passes on to websites is a small fraction of that share.
Our own site is a modest illustration. Between June and early September 2026 we recorded four sessions from ChatGPT against 51 from Google organic search. A conversion rate several times better on four sessions is not a channel. It is a rounding error with good manners.
The measurement problem underneath all of it.
In April 2026, Cloudflare recorded automated traffic passing human traffic across its network for the first time: 57.4% bots against 42.6% people. That is not a rhetorical flourish about the state of the web. It is a practical warning about your own analytics.
The clearest illustration comes from Wikipedia. The Wikimedia Foundation reported human page views down 8% year on year for March to August 2025 — but only found it after updating its bot-detection logic and reclassifying a wave of automated traffic that had been counted as people. Until the detection improved, the decline was invisible. The traffic looked healthy because machines were making up the difference.
Wikipedia is among the most heavily cited sources in AI answers and is losing human visitors because of it. Being the source a machine reads and being the site a person visits have become different things.
What we cannot tell you.
None of this is Malaysian data. The zero-click and click-through figures are United States measurements; the publisher figures are global but weighted towards English-language Western media. The mechanism is unquestionably the same here — the same search engine, the same feature, the same assistants — but the magnitude in this market has not been measured, by us or by anyone we can find. If someone quotes you a Malaysian percentage, ask where it was measured.
These are also panels and samples rather than censuses. They are credible, they are the best available, and they are estimates. That is worth saying plainly in a field where the same figure is rounded up at every retelling.
What we would do with it is narrow. Look at your own analytics rather than anyone’s average: the AI Assistant channel is already there and will tell you what actually reaches you. Publish the answers to the questions buyers ask, because a summary can only repeat what you have written down. And treat the parts of your pipeline that never depended on search as the ones worth strengthening. We have set out the argument behind that separately.
Retrieved from https://amkatechnologies.com/insights/how-much-traffic-are-websites-losing-to-ai
Published 2 September 2026. AMKA Technologies Sdn Bhd, SSM 202301041763 (1535682-T).