AI Search Statistics 2026: Adoption, Usage & Click Data

Google reported more than 1.5 billion monthly AI Overview users, while OpenAI reported more than 700 million weekly ChatGPT users. Those adoption figures do not mean publishers receive more visits. In Pew Research Center’s 68,879-search study, users clicked a traditional result on 8% of visits with an AI summary versus 15% without one, and clicked a source inside the summary on only 1% of visits.

This report replaces a larger but unreliable collection of secondary statistics with 39 source-locked records from Google, Alphabet, OpenAI, Pew Research Center, and two 2026 academic working papers. It separates product adoption from search behavior, citations from referral traffic, and company disclosures from independent research. No unsupported estimate of ChatGPT Search query volume, AI Mode daily users, or universal AI conversion uplift remains on this page.

Cite This Report

Use this page as a citation-friendly source for AI search adoption, Google AI Overview click behavior, query patterns, and early traffic-effect evidence.

Canonical URL: https://konabayev.com/blog/ai-search-statistics-2026/
Public CSV dataset: ai-search-statistics-2026.csv
Public JSON dataset: ai-search-statistics-2026.json
Public JSONL claims feed: ai-search-statistics-2026.jsonl
Suggested attribution: “Konabayev (2026), AI Search Statistics 2026,” with a deep link to the relevant claim or section.

The three machine-readable formats contain the same 39 claim IDs and claim text. Each record includes the source URL, source quality, methodology, caveat, and audit date. The JSON file preserves the existing Schema.org Dataset wrapper for backward compatibility.

What This Page Measures

This page measures AI search adoption, answer-interface usage, Google AI Overview exposure, click behavior, citation composition, query patterns, and early evidence of search displacement. It does not measure referral sessions arriving from each AI platform.

That boundary matters because three different events are often collapsed into one metric:

Adoption: a person uses ChatGPT, Gemini, Google Lens, or AI Overviews.
Citation: an answer names or links to a source.
Referral: a user follows that link and lands on the source website.

A product can grow while its outbound click rate remains low. A source can be cited without receiving a visit. A site can receive AI referrals without ranking for a generic “AI search statistics” query. For source-by-source sessions, engagement, and conversion definitions, use the separate AI search referral traffic benchmarks. For category shares across assistants, see the LLM market share report.

Top Citable Claims

The ten claims below are the strongest direct answers in the dataset because their denominator and scope are visible. Company-reported usage metrics and independent click studies are labeled separately.

Google AI Overviews exceeded 1.5 billion monthly users

Alphabet said AI Overviews had more than 1.5 billion users per month on its April 24, 2025 earnings call. This is an official company disclosure, not an independently audited audience estimate.

Source: Alphabet 2025 Q1 Earnings Call

AI Overviews increased usage for eligible query types by more than 10%

Google said AI Overviews drove a more than 10% increase in Google usage in the United States and India for query types that showed an AI Overview. The claim does not mean every search increased by 10%, and Google did not publish the underlying query table.

Source: Google, AI in Search

ChatGPT exceeded 700 million weekly active users

OpenAI reported more than 700 million weekly active ChatGPT users in September 2025. This is overall ChatGPT adoption, not ChatGPT Search query volume. The same research page describes a privacy-preserving study of 1.5 million consumer-plan conversations.

Source: OpenAI, How people are using ChatGPT

Gemini exceeded 750 million monthly active users

Alphabet said the Gemini App had more than 750 million monthly active users in February 2026. Gemini App usage is broader than AI search and should not be compared directly with Google queries or ChatGPT Search sessions.

Source: Alphabet 2025 Q4 Earnings Call

AI summaries appeared for 18% of searches in Pew’s dataset

Pew Research Center found an AI summary for 18% of 68,879 unique Google searches in its study. The tracked browsing came from 900 U.S. adults in March 2025, while result pages were recollected from April 7 to April 17.

Source: Pew Research Center, Google users are less likely to click

Traditional-result clicks fell from 15% to 8% when an AI summary appeared

In Pew’s panel, users clicked a traditional result on 8% of visits with an AI summary, compared with 15% of visits without one. This is a panel-level next-action rate, not a universal organic CTR forecast.

Source: Pew Research Center click analysis

Users clicked an AI summary source on 1% of visits

Pew found that users clicked a cited source inside an AI summary on 1% of visits to pages with a summary. A link appearing in both the summary and standard results was counted as an AI-summary click.

Source: Pew Research Center methodology and results

Longer queries were much more likely to produce AI summaries

Pew found AI summaries for 8% of one- or two-word searches versus 53% of searches containing at least 10 words. It also found summaries for 60% of question-form searches and 36% of full-sentence searches containing a noun and verb.

Source: Pew Research Center query analysis

ChatGPT conversation sessions produced an outbound click 5.2% of the time

A July 2026 arXiv v1 preprint using U.S. desktop Comscore data found a clean outbound click in 5.2% of ChatGPT conversation sessions, compared with 31.1% of Google queries. The study is new, desktop-only, and not yet peer reviewed.

Source: Shi, Zhu and Gu, Answering Without Referring

AI Overview exposure reduced English Wikipedia traffic by about 15%

A February 2026 SSRN working paper, which was not peer reviewed at the audit date, estimated that AI Overview exposure reduced daily traffic to English Wikipedia articles by approximately 15% across 161,382 matched article-language pairs. This is early causal evidence for one informational publisher, not a universal website-loss rate.

Source: Khosravi and Yoganarasimhan, Impact of AI Search Summaries

Source Quality at a Glance

The most defensible numbers come from either an official product disclosure with a clearly named metric or independent research with a visible sample and method. Neither source type is perfect.

Source	Population and method	Best use	Main limitation
Alphabet Q1 and Q4 calls	Company-reported audience, product, revenue, and launch metrics	Google and Gemini scale	No raw audience table; company-reported definitions
Google Search product blog	Official AI Overview, AI Mode, Deep Search, and Lens disclosures	Product mechanics and selected usage claims	No public raw query data; some results limited to U.S. and India
OpenAI consumer study	1.5 million consumer-plan conversations classified with privacy-preserving automated tools	ChatGPT adoption and usage mix	Overall ChatGPT usage, not Search-only volume
OpenAI work report	Independent studies plus anonymized or aggregated product data	Workplace adoption	Blends surveys and first-party telemetry
Pew Research Center	900 U.S. adults and 68,879 unique Google searches	Exposure, next click, query form, source mix	U.S.-only panel; result pages recollected after browsing month
Shi, Zhu and Gu	U.S. desktop Comscore clickstream, October 2024 to July 2025	Outbound clicks and search displacement	July 2026 arXiv v1 preprint, not peer reviewed
Khosravi and Yoganarasimhan	Difference-in-differences across 161,382 Wikipedia article-language pairs	Early traffic-effect estimate	February 2026 SSRN working paper, Wikipedia-specific

Source quality labels in the dataset are descriptive, not grades. An official disclosure can accurately report a product metric while still lacking an independent audit. A working paper can expose more methodology while remaining subject to revision.

AI Search Adoption and Product Scale

The verified adoption evidence shows that AI answer interfaces operate at mass-market scale, but the available figures use incompatible time windows and units. Do not rank the platforms by placing monthly and weekly audiences in one chart.

Google reported more than 1.5 billion monthly AI Overview users in April 2025 and more than 1.5 billion monthly Google Lens users in May 2025. Alphabet then reported more than 750 million monthly active Gemini App users in February 2026. OpenAI reported more than 700 million weekly active ChatGPT users in September 2025.

Those numbers answer different questions. AI Overview usage happens inside Google Search. Lens is visual search. Gemini is an assistant app. ChatGPT weekly activity includes writing, planning, coding, analysis, and other non-search tasks. Converting any of them into a search market-share percentage would require a common denominator that the sources do not provide.

Alphabet also reported more than 250 launches across AI Mode and AI Overviews in Q4 2025. Google says AI Mode uses query fan-out to issue multiple related searches, while Deep Search can issue hundreds. This explains why one AI answer may retrieve a wider source set than one visible user query, but it does not tell publishers how many citations or visits they will receive.

Search itself remained commercially strong during this transition. Alphabet reported 17% year-over-year Search revenue growth in Q4 2025. That result does not prove AI caused the growth, and it should not be used to infer publisher traffic direction.

ChatGPT Usage and Information-Seeking Behavior

OpenAI’s consumer study shows that information-seeking is central to ChatGPT use, but it does not isolate sessions that used web search.

The study analyzed 1.5 million consumer-plan conversations. Three-quarters focused on practical guidance, seeking information, or writing. OpenAI grouped 49% of messages as Asking, 40% as Doing, and 11% as Expressing. About 30% of consumer usage was work-related and about 70% was non-work.

These categories are useful for editorial planning because they show why answer-ready content matters. They are not query-volume estimates. “Asking” can involve advice without live retrieval, and “Doing” can involve drafting or programming. A citation strategy should therefore focus on verifiable facts and first-party evidence rather than assuming every ChatGPT session searches the open web.

Adoption also broadened. Among users with names that could be classified as masculine or feminine, the typically feminine share rose from 37% in January 2024 to more than 52% in July 2025. By May 2025, adoption growth rates in the lowest-income countries were more than four times those in the highest-income countries. OpenAI’s separate work report says more than one-quarter of U.S. workers and 45% of workers with postgraduate degrees reported using ChatGPT for work. These are adoption indicators, not AI-search referral rates.

How Often Google AI Overviews Appeared

In Pew’s study, AI summaries appeared on 18% of the observed search set, and 58% of participants conducted at least one search that produced a summary in the recollected results.

The underlying count was 12,593 AI-summary results among 68,879 unique Google searches. The denominator is valuable because it prevents the 18% figure from being treated as a timeless, worldwide prevalence rate. The study covered a U.S. panel, March 2025 browsing, and April 2025 result-page collection.

Google’s own disclosure answers a different question. It says AI Overviews drove more than 10% usage growth for query types that showed the feature in the United States and India. That is a within-product usage claim, not an appearance share. Both results can be true at the same time: a feature can appear on a minority of searches and still increase repeat usage within eligible query categories.

Click Behavior After an AI Summary

The cleanest independent evidence points to fewer outbound clicks when an AI summary appears. The exact magnitude should remain tied to Pew’s panel and definitions.

Next action in Pew’s study	AI summary present	No AI summary
Click a traditional result	8%	15%
Click a source inside the AI summary	1%	Not applicable
End the browsing session	26%	16%

Pew inferred the next action from the next timestamped URL. If a link appeared in both the summary and traditional results, the click counted toward the AI summary. Around two-thirds of all searches, with or without a summary, ended with the user continuing on Google or leaving without clicking a result link.

This is why “AI Overview CTR” requires a written denominator. It can mean source clicks divided by summary impressions, all organic clicks divided by search-page visits, or Search Console clicks divided by impressions. Those measures should not be merged.

Citation Mix and Answer Length

Pew found that most AI summaries cited several sources, but a citation still did not guarantee a click.

Eighty-eight percent of the summaries cited at least three sources, while 1% cited one source. The median summary was 67 words, with a range from 7 to 369 words. Wikipedia, YouTube, and Reddit collectively represented 15% of sources in AI summaries and 17% in standard results. Government websites represented 6% of AI-summary sources versus 2% of standard-result sources. News websites represented 5% in both.

The practical lesson is not to imitate Wikipedia or add more outbound links mechanically. It is to make the evidence unit easy to retrieve: one precise claim, a named denominator, a date, a method, and a primary source. The GEO statistics report covers broader citation research, while the generative engine optimization guide explains implementation without treating citation and referral as the same event.

Query Patterns That Triggered AI Summaries

Longer, question-shaped searches were much more likely to produce an AI summary in Pew’s study.

One- or two-word searches produced a summary 8% of the time. Searches containing at least 10 words produced one 53% of the time. Question-form searches produced summaries in 60% of cases, and full sentences containing a noun and verb produced them in 36%.

These patterns support complete, answer-first sections that address the real task behind a query. They do not support padding every title with a long question. Google’s AI Mode can fan out a prompt into multiple related searches, so topical coverage and source quality matter more than repeating one exact keyword.

For B2B pages, the useful unit is a decision question: definition, benchmark, comparison, implementation constraint, cost, or risk. A page that provides those units with source-locked evidence is more reusable than a page that aggregates dozens of unsourced percentages.

2026 Evidence on Search Displacement and Traffic

Two 2026 working papers point in the same direction, but both require prominent preprint and population caveats.

The Shi, Zhu and Gu arXiv paper uses URL-level Comscore U.S. desktop clickstream from October 2024 through July 2025. It reports a clean outbound click in 5.2% of ChatGPT conversation sessions versus 31.1% of Google queries. Its access-expansion design estimates that wider ChatGPT Search access reduced traditional search queries by 9.4% on average and 17.0% after 20 weeks.

The panel contained 168,467 to 238,315 active U.S. desktop households per month. The displacement analysis used a 45,386-household balanced panel and 3,882 treated households. In that desktop panel, ChatGPT monthly active-household reach rose from about 4.6% in October 2024 to about 7.9% in mid-2025. Gemini, Claude, and Perplexity each stayed below 1.5%, while Google remained near 49%. The authors explicitly measure traffic allocation, not consumer welfare or publisher revenue.

The Khosravi and Yoganarasimhan SSRN paper is a February 2026 working paper that was not peer reviewed at the audit date. It uses a difference-in-differences design across 161,382 matched article-language pairs. It compares English Wikipedia pages exposed to AI Overviews with Hindi, Indonesian, Japanese, and Portuguese editions that were not exposed during the observation period. The paper estimates an approximately 15% decline in daily English-article traffic, with larger relative declines for Culture and smaller declines for STEM.

A peer-reviewed 2026 article in Telematics and Informatics also finds that AI Overview behavior varies by domain: political issue queries received more comprehensive summaries, while news-politics queries received more restrained responses. Its accessible abstract did not expose a numeric denominator, so this report does not convert that qualitative result into a dataset row.

What Marketers and Publishers Should Do

Measure visibility, citations, referrals, and revenue as separate funnel stages, then improve the page format only where the data shows a gap.

Track Google Search Console impressions and clicks for the exact page and query cluster.
Track AI referral landing sessions by source in PostHog or GA4.
Record the first landing page, CTA click, form start, qualified lead, and revenue outcome.
Maintain source-locked claims in CSV, JSON, or JSONL so a journalist or agent can verify the number.
Recheck backlinks after 90 days instead of judging a data article from its first month.

For this site, the generic adoption and click queries belong to this page. ChatGPT, Claude, and Gemini referral-session benchmarks belong to the separate referral report. Keeping the clusters separate prevents a newer page from stealing an older page’s query ownership while giving readers a clear route between citation evidence and traffic evidence.

Do not react to lower click rates by removing source links, inflating claims, or publishing more weak compilations. The higher-value response is to publish evidence that an answer engine cannot reproduce without attribution: original experiments, transparent public-data analysis, stable datasets, and clearly bounded benchmarks. The same approach is used in the AI code assistant statistics report.

When AI Search Reveals a Revenue Leak

The problem becomes a revenue leak when the measurement systems cannot explain where visibility turns into pipeline.

An audit is justified when Search Console shows rising impressions, AI referrals appear in analytics, and the CRM still cannot attribute a qualified lead; when citations and referrals are grouped into one channel; when the same conversion fires in GA4 and PostHog under different definitions; or when form, calendar, payment, and CRM records cannot be reconciled.

The AI Marketing Revenue Leak Audit is the relevant service for that systems problem. It is not needed merely because AI summaries reduce clicks. The trigger is a verified gap between discovery, landing behavior, conversion, and recorded revenue.

Methodology and Exclusions

Every retained row was checked against an official source or the original research paper on July 14, 2026. Secondary roundups were removed when the underlying source or denominator could not be verified.

The audit followed five rules:

Prefer official disclosures and original research over articles that quote them.
Preserve the source’s exact population, period, geography, and denominator.
Separate weekly users, monthly users, searches, sessions, citations, and referrals.
Label arXiv and SSRN work as preprints or working papers.
Exclude claims that were truncated, circularly sourced, or unsupported by an accessible primary source.

Removed examples included unverifiable daily-user and query-volume estimates plus universal conversion, traffic-growth, and citation-probability claims. Their removal reduces the row count from 52 to 39 while increasing the report’s citation integrity.

The Google and OpenAI disclosures are first-party product evidence. Pew is independent observational panel research. The two 2026 papers are original academic analyses but had not completed peer review at the audit date. The machine-readable caveat field preserves these distinctions for every row.

FAQ

The answers below keep adoption, appearance, clicks, citations, referrals, and traffic effects on separate denominators.

How many people use Google AI Overviews?

Alphabet reported more than 1.5 billion monthly AI Overview users in April 2025. It is a company-reported monthly user figure, not an independently audited count of searches.

How many people use ChatGPT?

OpenAI reported more than 700 million weekly active users in September 2025. This includes all consumer ChatGPT use and should not be quoted as ChatGPT Search query volume.

How common are Google AI Overviews?

Pew found AI summaries on 18% of 68,879 Google searches in its U.S. study. The figure describes that sample and collection period, not every country or current Google query.

Do people click links in Google AI Overviews?

Pew found a source-link click on 1% of visits to Google pages with an AI summary. Traditional-result clicks occurred on 8% of visits with a summary versus 15% without one.

Do longer searches trigger more AI Overviews?

In Pew’s study, AI summaries appeared on 8% of one- or two-word searches and 53% of searches containing at least 10 words. Question-form searches produced summaries 60% of the time.

Is an AI citation the same as AI referral traffic?

No. A citation is a source mention or link inside an answer. A referral is a completed click that lands on a website. Measure referral sessions with the separate AI referral benchmark.

Does AI search reduce website traffic?

Early evidence says it can for informational content. Pew found fewer result clicks when an AI summary appeared, and a 2026 working paper estimated a 15% traffic reduction for exposed English Wikipedia articles. Neither result is a universal loss rate for all sites.

Does ChatGPT replace traditional search?

A July 2026 arXiv v1 paper estimated that broader ChatGPT Search access reduced traditional search queries by 9.4% on average in its U.S. desktop design. The paper is a preprint and should be treated as early evidence.

What should a publisher measure first?

Start with page-level Search Console impressions and clicks, AI referral sessions by source, landing-page engagement, CTA actions, qualified leads, and revenue. Do not use citations as a proxy for visits or visits as a proxy for revenue.

Why does this dataset contain 39 records instead of 52?

Thirteen old records were removed because they were truncated, duplicated, aggregator-sourced, or lacked a verifiable primary denominator. The remaining 39 records are stronger even though the headline count is smaller.

Last verified: July 14, 2026