State of AI Search 2026
Discovery is moving from ranked links to synthesized answers. Here are the 2026 numbers on assistant adoption, zero-click search, citation behavior, AI referrals and the crawler economy — compiled from the primary research, with what it means for brands.
From the AI Search Optimization toolset.
For twenty-five years, digital discovery had one shape: a query, ten blue links, a click. That shape is dissolving. By mid-2026, ChatGPT alone serves roughly 900 million people a week, Google has pushed AI-generated answers in front of billions of searches, and the share of US Google searches that end without any click at all has climbed to 68%. The question brands used to ask — where do we rank — is giving way to a harder one: are we in the answer.
The economics underneath are shifting just as fast. AI platforms now crawl hundreds to thousands of pages for every visitor they send back, publishers are blocking and metering crawlers at the network layer, and the small stream of traffic AI assistants do refer converts dramatically better than the average organic visit. Visibility, traffic and value have come apart, and each now needs its own strategy.
This report compiles the current evidence — platform disclosures, Pew Research, Ahrefs, Semrush, SparkToro, Adobe Analytics, Cloudflare and the citation-behavior studies — into one picture of where AI search stands in 2026 and where it is heading. Every number is sourced. Where trackers disagree, we show the spread and say why.
The shift in discovery
Assistant adoption has moved past the early-adopter phase and into the mainstream. OpenAI announced 800 million weekly active ChatGPT users at DevDay in October 2025 and has reported roughly 900 million weekly actives through 2026 — about double the figure from a year earlier. Perplexity, the largest of the pure answer engines, disclosed 780 million queries in May 2025 with growth above 20% month over month, and most trackers place it near a billion queries a month by mid-2026.
Google answered by making search itself generative. AI Overviews — the synthesized answer block above traditional results — reached 2 billion monthly users by mid-2025 on Google’s own disclosure, and the fully conversational AI Mode passed 1 billion monthly users by mid-2026. The Gemini app roughly doubled year over year to around 900 million monthly actives. Gartner’s much-debated February 2024 call that traditional search volume would fall 25% by 2026 got the mechanism half right: rather than shrinking, search absorbed the assistant and changed form from the inside.
How much of search is already AI-mediated depends on who you ask, and the spread is instructive. Semrush measured AI Overviews on roughly 16% of US queries in late 2025 after a mid-year peak near 25%, while BrightEdge — which tracks commercial verticals — recorded presence on about 48% of monitored queries by early 2026. Both can be true at once: coverage concentrates on informational and consideration-stage queries, exactly the territory where brands earn their pipeline.
- ChatGPT hit 800M weekly active users in October 2025 and roughly 900M through 2026 — about double the prior year. — OpenAI / TechCrunch, 2025–2026
- Perplexity processed 780M queries in May 2025, growing more than 20% month over month per its CEO; most trackers place it near 1B monthly queries by mid-2026. — Perplexity / TechCrunch, 2025
- Google AI Overviews reached 2B monthly users by mid-2025; AI Mode passed 1B monthly users by mid-2026. — Google, 2025–2026
- The Gemini app roughly doubled year over year to about 900M monthly active users. — Google I/O, 2026
- AI Overviews coverage estimates range from ~16% of US queries (Semrush, late 2025) to ~48% of tracked commercial queries (BrightEdge, early 2026) — the gap reflects keyword mix, and both point the same direction. — Semrush; BrightEdge, 2025–2026
Reported user base of AI answer surfaces, mid-2026 (millions)
Company-reported figures; cadences differ (weekly vs. monthly actives). AI Overviews shown muted because it reaches users inside classic Google results rather than as a destination people choose.
The zero-click squeeze
The clearest evidence on what AI answers do to clicks comes from Pew Research Center, which analyzed roughly 68,000 real Google searches from about 900 US adults in March 2025. When an AI Overview appeared, users clicked a traditional result 8% of the time, versus 15% when it was absent — click probability nearly cut in half. Links inside the AI Overview itself were clicked on about 1% of views, and users ended their browsing session outright 26% of the time with an AI summary present, versus 16% without.
Keyword-level studies show the squeeze intensifying as the feature matures. Ahrefs’ March 2025 analysis of 300,000 keywords found a 34.5% lower click-through rate for the top-ranking page when an AI Overview was present; its December 2025 update on the same design measured the reduction at 58%. The aggregate result showed up in SparkToro and Similarweb’s clickstream data: 68% of US Google searches ended without a click to the open web in the first four months of 2026, up from about 60% in 2024 — the fastest two-year acceleration since tracking began.
The strategic consequence is that rankings and traffic have decoupled. A page can hold position one, feed the answer that satisfies the searcher, and watch its clicks fall by half. Impressions, citations and share-of-answer are becoming the honest measures of search visibility, and the clicks that remain skew toward higher-intent visitors who wanted more than a summary.
- Click-through on searches with an AI Overview: 8%, versus 15% without — links inside the Overview drew clicks on roughly 1% of views. — Pew Research Center, 2025
- Sessions ended entirely after 26% of searches with an AI Overview, versus 16% of traditional searches. — Pew Research Center, 2025
- Position-one CTR fell 34.5% with an AI Overview present in March 2025; the December 2025 re-run measured a 58% reduction. — Ahrefs, 2025
- 68.0% of US Google searches ended with zero clicks to the open web in January–April 2026, up from 60.4% in 2024. — SparkToro × Similarweb, 2026
The major AI Overviews click-impact studies
| Study | Sample | Core finding | Published |
|---|---|---|---|
| Pew Research Center | ~68K searches, ~900 US adults | CTR 8% with AI Overview vs. 15% without | Jul 2025 |
| Ahrefs (initial) | 300K keywords | −34.5% CTR for the #1 result with an AI Overview | Mar 2025 |
| Ahrefs (update) | 300K keywords | −58% CTR for the #1 result as rollout matured | Dec 2025 |
| SparkToro × Similarweb | US clickstream panel | 68% of Google searches end without a click | 2026 |
Methodologies differ — panel observation vs. keyword-cohort comparison — yet every credible study points the same direction, and the effect grows as AI answers cover more queries.
Report search performance on visibility and share-of-answer alongside clicks. A traffic decline with stable rankings is now an expected pattern; the teams that instrument citations and impressions early will read the shift accurately while competitors misdiagnose it as a content problem.
Who gets cited
AI answers concentrate their trust in a remarkably short list of domains. Semrush’s study of 150,000 LLM citations across ChatGPT, Perplexity, Gemini and Google AI Overviews found Reddit appearing in 40.1% of cited sources, Wikipedia in 26.3% and YouTube in 23.5% — and no other single domain cracked 5%. Community discussion, reference material and video carry the citation graph; the long tail of brand websites competes for what remains.
That mix is platform-specific and volatile. Semrush’s multi-platform follow-up — 230,000 prompts and over 100 million citations tracked across 13 weeks in 2025 — watched ChatGPT’s Reddit citation rate swing from roughly 60% of responses to roughly 10% after a September 2025 retuning, an overnight reshuffle of visibility that no site involved did anything to cause. Google’s AI Mode leans heavily on LinkedIn, YouTube, Reddit and Google’s own properties. Profound’s tracking recorded LinkedIn climbing from around #11 to #5 among ChatGPT’s most-cited domains between November 2025 and February 2026 — the largest authority shift it observed all year.
Format matters as much as domain. Wix Studio’s AI Search Lab analysis of citation behavior found listicles drawing 21.9% of AI citations, articles 16.7% and product pages 13.7% — three formats accounting for 52% of everything cited. The bias has a neutrality condition: in professional services, third-party listicles captured 80.9% of citations against 19.1% for self-promotional lists. Models reward content that compares options honestly, states tradeoffs and is structured for extraction — which is precisely the content most brands have historically left for affiliates to write.
- Reddit appeared in 40.1% of AI citations, Wikipedia 26.3%, YouTube 23.5% — no other single domain exceeded 5%. — Semrush, 150K citations, 2025
- A September 2025 ChatGPT retuning cut Reddit’s citation appearance from ~60% of responses to ~10% — model updates now reshuffle visibility the way algorithm updates once did. — Semrush multi-platform study, 2025
- Google AI Mode’s most-cited domains skew to LinkedIn, YouTube, Reddit and Google-owned properties. — Semrush, 2025
- Listicles drew 21.9% of AI citations, articles 16.7% and product pages 13.7% — 52% of all citations across three formats. — Wix Studio AI Search Lab, 2026
- Third-party listicles took 80.9% of professional-services citations versus 19.1% for self-promotional lists — models prefer editorial neutrality. — Wix Studio AI Search Lab, 2026
- LinkedIn rose from ~#11 to #5 among ChatGPT’s most-cited domains between November 2025 and February 2026. — Profound, 2026
Share of AI citations by domain (150K citations, four AI engines)
Semrush, June 2025 — share of AI answers in which each domain appeared as a cited source across ChatGPT, Perplexity, Gemini and Google AI Overviews. Shares overlap because answers cite multiple sources.
AI referral traffic: small share, outsized intent
The traffic AI assistants send back remains a small slice of the whole — roughly 1% of total site visits on average across the industries Conductor studied in early 2026 — but it is growing at rates search has never produced. Adobe Analytics, drawing on more than a trillion visits to 130-plus top North American retailers, measured AI-referred traffic to US retail up 393% year over year in Q1 2026.
Quality is the real story. Adobe’s early-2025 data showed AI-referred visitors converting below site average; by the 2025 holiday season they converted 31% better than non-AI traffic, and by March 2026 the advantage had widened to 42%. Semrush’s cohort study puts the average AI-search visitor at 4.4 times the value of a traditional organic visitor on conversion. The mechanism is straightforward: the assistant absorbs the research phase, so the person who clicks through arrives compared, qualified and closer to a decision.
Projections suggest the channel graduates from rounding error this cycle. Semrush projects that for the digital-marketing topic set it studied, AI search could drive more visits than traditional organic search by early 2028. Even at today’s ~1% share, a channel converting at multiples of organic deserves its own measurement, attribution and content strategy — most analytics setups still bury it inside referral noise.
- AI-referred traffic to US retail grew 393% year over year in Q1 2026, measured across 1T+ visits to 130+ top North American retailers. — Adobe Analytics, 2026
- AI referrals converted 31% better than non-AI traffic during the 2025 holiday season, and 42% better by March 2026 — a reversal from below-average conversion a year earlier. — Adobe Analytics, 2025–2026
- The average AI-search visitor is worth 4.4x a traditional organic visitor on conversion rate. — Semrush, 2025
- AI referrals average roughly 1% of total site traffic across studied industries — small share, steep curve. — Conductor, 2026
- For the digital-marketing topics Semrush tracked, AI search is projected to drive more visits than traditional search by early 2028. — Semrush, 2025
AI-referred visitor quality — the current evidence
| Signal | Reading | Source |
|---|---|---|
| YoY growth, AI referrals to US retail | +393% | Adobe Analytics, Q1 2026 |
| Conversion vs. non-AI traffic (US retail) | +42% | Adobe Analytics, Mar 2026 |
| Holiday-season conversion lift | +31% | Adobe Analytics, 2025 |
| Value per visitor vs. traditional organic | 4.4x | Semrush, 2025 |
| Share of total site traffic (cross-industry avg.) | ≈1% | Conductor, 2026 |
Retail figures reflect Adobe’s panel of top North American retailers; cross-industry share and value multiples come from separate cohort studies. Treat the direction as robust and the exact multiples as category-dependent.
The crawler economy
The old bargain of the web — let crawlers in, get visitors back — is being renegotiated in public. Cloudflare’s network data puts numbers on the imbalance: Google’s traditional search crawls on the order of five pages for every visitor it refers, while Perplexity crawls roughly 190, OpenAI roughly 900, and Anthropic’s crawler runs into the thousands of pages per referral. The exact ratios swing between measurement windows; the ordering and the orders of magnitude hold steady.
Purpose data sharpens the picture. In Cloudflare’s May 2026 measurements, under 10% of AI crawler requests were search-purpose — fetching content to answer a live query with a link back — while roughly 90% served training or mixed purposes. Infrastructure providers are responding with defaults: Cloudflare began blocking AI crawlers for new domains in July 2025, launched pay-per-crawl so publishers can charge for access, and announced that from September 15, 2026 it will block mixed-use crawlers on ad-carrying pages by default.
Meanwhile the llms.txt proposal — a manifest telling language models what to read — is spreading faster than it is working. Adoption in the top 10,000 domains rose from about 1% in July 2025 to 5.6% by June 2026, and reached 8.7% of the top 1,000 sites. Yet Ahrefs’ server-log analysis across 137,000 domains found 97% of llms.txt files received zero requests in May 2026. Publishing one costs an hour and may pay off if platforms adopt the standard; the enforceable controls today live in robots.txt, WAF rules and the network edge.
- Crawl-to-refer ratios, mid-2026: Google’s traditional search near 5:1, Perplexity ~190:1, OpenAI ~900:1, Anthropic in the thousands-to-one range. — Cloudflare Radar, 2026
- Under 10% of AI crawler requests were search-purpose in May 2026; roughly 90% served training or mixed purposes. — Cloudflare, 2026
- Cloudflare blocks AI crawlers by default for new domains (July 2025) and will block mixed-use crawlers on ad-supported pages by default from September 15, 2026. — Cloudflare / TechCrunch, 2025–2026
- llms.txt adoption in the top 10K domains grew from ~1% (July 2025) to 5.6% (June 2026); 8.7% of the top 1,000 sites now publish one. — Casey Burridge; Rankability, 2026
- 97% of llms.txt files received zero requests in May 2026 across 137K domains’ server logs — publish it cheaply, expect payoff only if platforms adopt the standard. — Ahrefs, 2026
Pages crawled per visitor referred (crawl-to-refer ratio)
| Platform | Pages crawled : 1 referral | Window |
|---|---|---|
| Google (traditional search) | ≈5 : 1 | Cloudflare, Q1 2026 |
| Perplexity | ≈190 : 1 | Cloudflare Radar, mid-2026 |
| OpenAI (GPTBot) | ≈900 : 1 | Cloudflare Radar, mid-2026 |
| Anthropic (ClaudeBot) | ≈10,000 : 1 | Cloudflare Radar, mid-2026 |
Ratios vary considerably between measurement windows as crawl volumes and referral behavior change; the ranking and the scale of the gap have been stable across every published window.
Blocking every AI crawler protects content and forfeits citations; allowing everything donates your corpus for a trickle of referrals. The defensible position is deliberate: allow search-purpose and user-query crawlers that cite and refer, meter or block pure training crawlers, and revisit quarterly as the paid-crawl market matures.
What winning brands do
A discipline has formed around all of this — answer engine optimization (AEO), generative engine optimization (GEO), AI search optimization — and in 2026 it moved from experiment to line item. Gartner, which predicted the 25% search-volume shift in early 2024, now projects that brands’ organic search traffic will fall by 50% or more by 2028 as consumers embrace generative search, and advises shifting investment toward the earned and owned assets AI systems actually consume. The channel’s economics justify the attention: visitors worth 4.4x organic, on a curve growing several hundred percent a year.
The winning pattern across the research is consistent. First, brands treat crawler access as policy rather than default, and instrument AI visibility — share of answer, citation counts by platform — with the same rigor they once applied to rank tracking. Second, they build content shaped like what gets cited: answer-first pages, honest comparisons, original data and benchmarks that models quote because independent quoting already happens. Third, they earn presence on the domains the citation graph runs through — Reddit threads, YouTube, LinkedIn, third-party listicles and reviews — because 90% of citations flow through territory brands influence through participation and PR rather than own outright.
The teams doing this well also plan for volatility. The September 2025 ChatGPT retuning that moved Reddit from 60% of citations to 10% is the new shape of an algorithm update: sudden, platform-specific and unannounced. Diversified citation presence across platforms and source types is the hedge, and quarterly measurement is the minimum cadence for a channel that reshapes itself this quickly.
- Gartner projects brands’ organic search traffic will decline 50% or more by 2028 as consumers embrace AI-powered search. — Gartner, 2025
- Gartner’s February 2024 prediction of a 25% traditional-search decline by 2026 played out as transformation: search absorbed generative answers and click behavior changed instead of volume. — Gartner, 2024
- The citation graph rewards neutral, structured, comparative content — third-party listicles drew 4x the citations of self-promotional equivalents in professional services. — Wix Studio AI Search Lab, 2026
- Original research is the most durable citation asset: models and the publications they cite both quote primary data, compounding visibility across the graph.
- Platform retunings now move citation share the way core updates once moved rankings — diversification across engines and source types is the hedge. — Semrush multi-platform study, 2025
AI search rewards the fundamentals brands should have invested in anyway: technical accessibility, structured data, genuinely useful comparative content, primary research and earned authority on the platforms people trust. The brands winning citations in 2026 started their AEO programs in 2024. The cost of starting now is a quarter of setup; the cost of waiting is watching competitors become the default answer.
Outlook: five calls for the next 18 months
US zero-click share passes 70% by the end of 2027.
SparkToro × Similarweb measured a 7.6-point rise in two years to 68% — the fastest acceleration on record — while AI Overviews coverage is still expanding and AI Mode queries more than double every quarter. The forces driving the number upward all remain in motion.
AI referrals reach 3–5% of site traffic in content-heavy categories by 2028, converting above site average the whole way.
The channel sits near 1% of traffic growing 300–900% year over year (Adobe, Conductor), Semrush projects an early-2028 crossover for the topics it tracks, and every conversion study since late 2025 shows AI visitors outperforming organic.
Citation volatility becomes the new algorithm update, and share-of-answer becomes a board-level metric.
ChatGPT’s September 2025 retuning moved Reddit from ~60% to ~10% of citations overnight (Semrush), and LinkedIn jumped six places in three months (Profound). Enterprise AI-visibility tracking — Semrush, Profound, Peec and a fast-growing tool category — is standardizing the measurement.
Paid placement arrives inside assistant answers at meaningful scale.
Google announced ads in AI Mode at Marketing Live 2025, Perplexity has run sponsored follow-up questions since late 2024, and OpenAI is widely reported to be building an ads capability against a ~900M-weekly-user surface. Monetization pressure at that scale has one historical outcome.
Crawling gets priced: paid access and licensing become the norm for premium content.
Cloudflare’s pay-per-crawl, its July 2025 default block for new domains, and the September 2026 mixed-use crawler block create the enforcement layer, while the wave of publisher licensing deals since 2024 establishes the price precedent. Crawl-to-refer ratios in the hundreds-to-thousands make free access economically irrational for content owners.
Sources
- TechCrunch — Sam Altman says ChatGPT has hit 800M weekly active users (OpenAI DevDay) (2025)
- TechCrunch — Perplexity received 780M queries in May 2025, CEO says (2025)
- Google — I/O and earnings disclosures: AI Overviews, AI Mode and Gemini usage milestones (2025–2026)
- Pew Research Center — Google users are less likely to click on links when an AI summary appears in the results (2025)
- Ahrefs — AI Overviews reduce clicks by 34.5% (2025)
- Ahrefs — Update: AI Overviews reduce clicks by 58% (2025)
- SparkToro × Similarweb — In 2026, less than one third of Google searches still send a click (2026)
- Semrush — The most-cited domains in AI: a 3-month study (2025)
- Semrush — We studied the impact of AI search on SEO traffic (2025)
- Wix Studio AI Search Lab — The content types most cited by LLMs (2026)
- Peec AI — Top domains cited by AI search: analysis of 30M sources (2026)
- Profound — AI platform citation tracking (LinkedIn authority shift) (2026)
- Adobe Analytics — AI-referred traffic to US retail: growth and conversion (2025–2026)
- Conductor — AI referral share of website traffic across industries (2026)
- Cloudflare — The crawl before the fall of referrals: AI’s impact on content providers (2025)
- Cloudflare — The crawl-to-click gap: data on AI bots, training and referrals (2026)
- TechCrunch — Cloudflare’s new policy pushes AI companies to pay for publishers’ content (2026)
- Gartner — Predicts search engine volume will drop 25% by 2026 due to AI chatbots (2024)
- Gartner — Brands’ organic search traffic projected to fall 50%+ by 2028 as consumers embrace GenAI search (2025)
- Rankability — llms.txt adoption across the top 1,000 websites (2026)
- Casey Burridge — State of llms.txt adoption (multi-million-domain crawl) (2026)
- Ahrefs — llms.txt server-log study: 97% of files receive zero requests (via PPC Land) (2026)
Let's grow.
EGGKNITE© 2026 ALL RIGHTS RESERVED — A VERTEX INNOVATION COLLECTIVE COMPANY