AI Search Statistics, With Sources

This is the fastest-moving and worst-sourced topic in marketing right now.

Almost every AI search statistic in circulation comes from a company selling AI search visibility software. Some of those figures are sound. Several are estimates that acquired a decimal point on the way down the chain and now get quoted as findings.

There is one genuinely independent study with published methodology, and it is not the one everybody cites. It goes first.

A note on the review cycle: everything here is dated, and this page is set to be re-read every three months. A number from 2025 in this category is not necessarily still true.

The one independent study

Pew Research Center, July 2025.

Methodology, stated plainly on the page: browsing data from 900 US adults on Pew’s KnowledgePanel Digital, who agreed to install a tracking app. It covers URLs visited 1-31 March 2025, amounting to roughly 2.5 million webpage visits across 1.1 million unique URLs.

That is observed behaviour from a probability-based panel, not a self-report survey and not a vendor’s platform sample. In this category it is close to unique.

What it found:

• When an AI summary appeared, users clicked a traditional search result in 8% of visits
• When no AI summary appeared, they clicked in 15% of visits
• 58% of respondents ran at least one Google search in March 2025 that produced an AI summary
• The median AI summary was 67 words, ranging from 7 to 369

So the click rate roughly halves. It does not go to zero, which is the version you often hear. Nearly twice as likely to click without a summary is a serious finding on its own and does not need inflating.

What Google and OpenAI have said about scale

• Google AI Overviews reach 2 billion monthly users (Google, 2025)
• ChatGPT has around 700 million weekly active users (OpenAI, 2025)
• Google AI Mode reported 100 million users across the US and India (Google, 2025)

These are company disclosures, not independent audits. Treat them as directional and remember both companies have an interest in the number being large.

Semrush’s own research, separated from what it repeats

Semrush’s AI SEO statistics page is the most-linked resource on this topic. It is worth understanding what it actually is: mostly a roundup. Of the figures on it, a minority are Semrush’s own data and the rest are cited from Pew, Bain, Adobe, Seer Interactive, Momentic, Originality.ai and others.

Attributed to Semrush’s own analysis:

• 88% of AI Overview searches are informational
• 95% of AI Overview keywords carry no or low-CPC advertising
• 68% of AI Overview terms get 100 or fewer monthly searches
• 50% of ChatGPT citations point to business sites

That third one is the interesting one for a contractor. If most AI Overview terms are very low volume informational queries, the overlap with the handful of commercial searches that actually bring you work is smaller than the panic suggests.

The 4.4x figure, and what it actually says

You will see this everywhere as “AI search visitors convert 4.4x higher than organic.” Some sites have escalated it further, to 23x, or to per-platform conversion rates like ChatGPT 15.9% and Perplexity 10.5%.

What Semrush’s page states is that AI search visitors are worth 4.4x more than traditional organic visitors. Value, not conversion rate. Those are different claims, and the page does not publish the sample size, date range or method behind it.

We are not saying it is wrong. We are saying it is a vendor estimate of value with no disclosed methodology, and it is being quoted as a measured conversion multiplier. The per-platform percentages could not be traced to any study at all.

If someone is using that number to justify a retainer, ask them for the method.

What AI Overviews did to clicks on the top result

Ahrefs has run the best-documented study on this, and then updated it, which is worth knowing because both numbers are in circulation as if they were rivals.

Method, published on the page: 300,000 keywords from Ahrefs Keywords Explorer, split into 150,000 with an AI Overview present and 150,000 informational keywords without one. Average desktop CTR per keyword per month came from aggregated Google Search Console data, comparing March 2024 (before the US rollout) with March 2025 (after).

• The original study: AI Overviews correlated with a 34.5% lower CTR for the top-ranking page.
• The December 2025 update: that had widened to 58%.

Both are correct for their date. If you see the 34.5% quoted today it is not wrong, it is old.

The detail almost everyone drops. In the same study, informational keywords without an AI Overview also fell, from 0.056 to 0.031. AI Overview keywords fell from 0.073 to 0.026.

So the control group dropped by roughly 44% on its own. The whole informational SERP got less clickable over that year, and AI Overviews are one cause inside a larger shift. Attributing the entire decline to them overstates a case that is already strong enough.

What consumers say they do

From the BrightLocal Local Consumer Review Survey 2026 (1,002 US adults, SurveyMonkey panel, published 11 February 2026):

• 45% now use ChatGPT or similar to find local business recommendations, up from 6% the year before
• 40% say they trust AI platforms for business recommendations
• 42% trust AI recommendations as much as traditional reviews
• 82% read AI-generated review summaries
• 23% decide from the AI summary alone

Self-reported use, so read it as direction rather than precision. The full consumer picture around it is in the local SEO and review statistics.

Set against Pew’s observed 58% who encountered an AI summary in a single month, these two datasets tell a consistent story from different angles: exposure is now the majority case, and a meaningful minority are deciding inside the summary.

The background nobody mentions

Ahrefs studied roughly 14 billion pages in their Content Explorer index and found that 96.55% get zero traffic from Google. A further 1.94% get between one and ten visits a month.

Published December 2023, so it predates AI Overviews entirely. Ahrefs is also unusually honest about its own limits on that page: the sample is a portion of the web, traffic estimates come from around 651 million keywords, and long-tail traffic may be missed.

It belongs here because of what it implies. The great majority of pages were getting nothing long before AI search arrived. When traffic to a contractor site is flat, AI Overviews are the fashionable explanation and rarely the correct one. The far more common reason is the one in why your website doesn’t get calls: the page was never going to rank, or the term was never searched.

Where this leaves a contractor

Honestly: less exposed than a publisher, and less exposed than the coverage implies.

Someone with a burst pipe is not reading a 67-word AI summary and deciding not to hire anyone. They are looking for somebody nearby who answers the phone. The Map Pack, reviews and proximity still decide most of that, and none of the Pew data touches the local pack.

Where it does reach you is earlier, on the research questions. “How much does foundation repair cost”, “do I need a sump pump or drain tile” — those are informational queries, they are exactly the 88% Semrush describes, and they are increasingly answered without a click. If your traffic came from articles like that, expect it to fall.

The reasonable response is not to panic-buy an AI visibility tool. It is to be the business the summary names, which is mostly the same work as local SEO for contractors has always been: consistent information, real reviews, and a site that says clearly what you do and where.

What we could not verify

• “AI search traffic up 527% year over year” — attributed to Search Engine Land, 2025. Not traced to a published methodology.
• Per-platform conversion rates (ChatGPT 15.9%, Perplexity 10.5%, Claude 5.0%) — appear on vendor blogs, no study behind them.
• “AI visitors convert 23x higher” — an escalation of the 4.4x value estimate. No source.
• Semrush’s clickstream study methodology. The billion-lines-of-clickstream figure is widely repeated, but sample construction and dates are not published on the page we read.

Using these statistics

Pew publishes its data and method free, and is the one to cite if you only cite one. The Google and OpenAI figures are company statements and should be described that way. Semrush’s own numbers are worth using with the attribution kept on them.