How Many Websites Are There on the Internet — and How Many Search Queries Never Reach Them?
Data brief · September 2026
How Many Websites Are There on the Internet — and How Many Search Queries Never Reach Them?
A 2026 accounting of hostnames, active sites, Google search volume, AI Overviews, LLM bots, zero-click behavior, and the traffic that no longer lands on the open web.
The short version
There is no official census of the internet. What exists instead is a stack of measurements that people collapse into one number — and then treat as if it described a living web.
As of mid-to-late 2026, crawlers can find about 1.5 billion hostnames that answer an HTTP request. Only about 200–220 million of those look like maintained websites. Roughly 400 million domain names are registered. Attention, meanwhile, concentrates on a few dozen destinations.
On the demand side, Google processes on the order of 13–14 billion searches a day. A large and growing share of those queries — plus billions more prompts sent to ChatGPT, Gemini, Perplexity, Claude, Grok, and Copilot — now end in a synthesized answer rather than a click to an independent site. Publisher networks report Google Search and Discover referrals down roughly a third to 40 percent year over year. Combined referral traffic from standalone AI chat products is still a rounding error next to Google. The disintermediation is happening inside Google first, and in chat windows second.
How many websites are there?
Three different questions, three different numbers
“How many websites are there?” is actually three questions. Mixing them is how 1.5 billion, 400 million, and 200 million all circulate as “the” answer.
| What is counted | Latest figure | As of | Source |
|---|---|---|---|
| Hostnames answering HTTP | 1,494,915,628 | July 2026 | Netcraft Web Server Survey |
| Domains seen in that crawl | 305,348,459 | July 2026 | Netcraft |
| Web-facing computers | 14,772,048 | July 2026 | Netcraft |
| Active / maintained sites | ~216.7 million | July 2026 | Netcraft active-site cut |
| Registered domain names | 401.6 million | Q2 2026 | Verisign DNIB |
Netcraft’s monthly survey is the figure most journalists quote. In July 2026 it received responses from 1,494,915,628 sites across 305,348,459 domains and 14.8 million web-facing computers — up 5.5 million sites on the prior month. Earlier in the year the same survey sat at 1.43 billion (March) and 1.49 billion (June). Growth of hundreds of millions of hostnames a year is real. It is also mostly noise.
Netcraft itself separates raw hostnames from active sites: pages with real, distinct content and signs of maintenance rather than a parked page or bulk-hosting placeholder. That cut is about 216.7 million in July 2026 — roughly 14–15 percent of everything the crawler can see. Siteefy and related recaps of the same survey put the March 2026 active total near 208 million. The gap of more than 1.2 billion hostnames is parked domains, placeholders, redirects, and abandoned experiments. Domain registration got cheap. Building and keeping a site did not.
Verisign counts something else again: paid registrations. Its Domain Name Industry Brief for Q2 2026 reported 401.6 million domain names across all top-level domains, up 9.1 million on the quarter and 29.9 million (8.1 percent) year over year. Verisign’s number is higher than Netcraft’s domain count because it includes names that are registered but not serving a live site.
Active does not mean visible
Even the 217 million “active” sites are not a democracy of attention. DomainsProject’s September 2026 crawl of websites with resolvable A records found 289.7 million sites answering from only 12.7 million IPv4 addresses. Half of those websites answered from just 1,025 addresses — down from 3,908 addresses in April 2023. The median website now sits on an address that serves more than 18,000 other sites. The web grew 71 percent in that window; the number of addresses behind it grew 16 percent. Consolidation is the story under the growth chart.
On the human side, a handful of properties absorb almost all visits. Semrush traffic rankings in summer 2026 put google.com at about 112 billion monthly visits, YouTube at 54 billion, Facebook at 11 billion, Instagram at 7 billion, and chatgpt.com already inside the global top five at about 6.2 billion. Reddit, Wikipedia, X, Amazon, TikTok, and the Microsoft/Google utility cluster fill out the rest of the first page. Over a billion hostnames exist. A few dozen destinations are the internet most people actually use.
How fast the inventory grows
Netcraft’s monthly net change has recently run in a wide band — on the order of 8–17 million additional hostnames in some months, with occasional net declines when parked inventory falls off. That implies something like 270,000–570,000 newly recorded hostnames per day, which popular explainers flatten into “546,000 new websites a day” or “six per second.” Domain registration growth is slower: Verisign’s net new domains in the year to early 2026 imply on the order of tens of thousands of new paid names per day, not hundreds of thousands of living publications.
Google does not publish a current website count. Antitrust testimony put the search index near 400 billion documents around 2020, drawn from trillions of crawled URLs, and Google has said a larger index is not automatically a better one. Gary Illyes has noted that Google checks robots.txt on about 4 billion hostnames daily — a ceiling on how many “sites” the company even bothers to consider, not a count of indexed properties. Indexing itself has gotten choosier. Multiple 2026 reports document a rise in “crawled — currently not indexed” in Search Console, especially for commodity informational pages.
How search volume actually looks in 2026
Google is still the pipe
Google disclosed in 2025 — its first official volume figure since 2016 — that it handles more than 5 trillion searches a year. That is about 13–14 billion searches a day, or roughly 158,000 per second, across Search and adjacent surfaces (Maps, Images, Shopping, and the rest). Older “8.5 billion a day” estimates from Internet Live Stats and similar panels almost certainly undercount those verticals. Some secondary trackers push the annual figure toward 5.9 trillion (about 16 billion a day). The safe working range is 8.5–14 billion Google searches per day, with Google’s own 5-trillion-plus annual disclosure as the floor.
StatCounter put Google at 90.5 percent of global search share across devices in May 2026 (about 85.5 percent in the United States). Bing sits around 4–5 percent globally and higher on U.S. desktops. Google.com itself drew on the order of 87.5 billion visits in a recent measured month. About 3.9 billion people use Google Search on a given day, averaging a few queries each. Google has long said that about 15 percent of daily queries have never been seen before — more than a billion novel strings every 24 hours.
Importantly, total Google query volume has not collapsed. Comscore counted 77 billion U.S. desktop searches in Q2 2026, up 8 percent from 71 billion in Q2 2024. Google has said search queries hit an all-time high even after AI features rolled out. The shift is composition, not extinction: longer, more conversational queries; more time spent on Google’s own surfaces; fewer outbound clicks.
| Pipe | Working volume | What it measures |
|---|---|---|
| Google Search (all surfaces) | ~13–14B / day (5T+ / year) | Official 2025 disclosure; range 8.5–16B in secondary sources |
| U.S. desktop search (Comscore) | 77B in Q2 2026 | Up 8% vs Q2 2024 |
| ChatGPT prompts | ~2–2.5B / day | OpenAI-linked disclosures; not all are “search” |
| Perplexity queries | Tens of millions / day | 780M in May 2025 last hard company month; growing |
| Combined AI assistants (one compilation) | ~4–5B prompts / day | Chat + Gemini + Claude + others; mixed intent |
The disintermediation: AI Overviews, chatbots, and zero-click
“Disintermediated” here means a question that used to send a person to someone else’s website now ends in a synthesized answer — on a Google results page, in AI Mode, or in a chatbot — with no click, or with a click so rare it does not replace the old referral.
Google AI Overviews are the main event
Coverage estimates disagree because panels disagree. Informational and commercial keyword sets light up AI Overviews far more often than broad or transactional sets. Treat any single percentage as a function of the sample.
| Tracker | AIO presence | Date / sample |
|---|---|---|
| Similarweb | ~43% of Google searches | Mid-2026; up from ~15% a year earlier |
| Comscore | 39.4% of U.S. desktop searches | June 2026; up from 25.8% in July 2025 |
| BrightEdge | ~48% of tracked queries | February 2026; ~31% a year earlier |
| Xponent21 | ~60.3% U.S. | April 2026 |
| Conductor | ~25.1% | Q1 2026, 21.9M queries |
| Semrush | ~15.7% | Late 2025, 10M-keyword panel |
Verticals are even more uneven. BrightEdge’s February 2026 industry cut put healthcare at 88 percent, education at 83 percent, B2B technology at 82 percent, and restaurants at 78 percent, against an all-industry aggregate of 48 percent. Entertainment sits much lower. Ahrefs and others find “why” and question queries in the 50–60 percent range, how-to lower, and transactional queries in the single digits to low teens in some panels. HasData’s 2026 re-run of an intent panel found AI Overviews on 67 percent of its queries overall and 88 percent or higher on symptom, troubleshooting, informational, travel, and comparison buckets — with local and recipes still relatively protected.
People Also Ask has been absorbed by the same machinery. AlsoAsked’s sample of 19.2 million English queries found 97 percent of PAA answers were AI Overviews in the first week of September 2026, up from 86 percent in August and from about 12 percent fourteen months earlier. Separate miners have called the rate effectively 100 percent.
AI Mode — the conversational continuation after an Overview — is a second surface on the same property. Similarweb tracked AI Mode visits rising from 126 million in June 2025 to 279 million in May 2026. Google has put AI Mode past a billion monthly users and AI Overviews in front of on the order of 1.7–2.5 billion people a month, depending on the briefing.
What happens to the click
The click is where the open web gets paid, ranked, and argued over. It is also the number that is falling.
| Finding | Magnitude | Source |
|---|---|---|
| Organic CTR with AIO vs without | 1.76% → 0.61% (−61%) | Seer Interactive, Sept 2025 |
| Organic clicks per 1,000 impressions | 87 → 36 (−59%) | First Page Sage, 2026 |
| Zero-click rate with AIO vs without | 82.4% vs 62.7% | First Page Sage, 2026 |
| Outbound organic clicks when AIO present | −39.8% | Randomized AIO-hidden experiment, 2026 |
| Click a traditional result when AIO shows | 8% vs 15% without | Pew Research, 2025 |
| Clicks that come from links inside the AIO | ~7% of remaining outbound clicks | Experimental / industry recaps |
| U.S. zero-click overall | ~68% | 2026 compilations; ~49% in 2019 |
A few caveats belong next to those drops. Seer later reported a partial CTR rebound on AIO queries in early 2026 (from a 1.3 percent floor in December 2025 to about 2.4 percent in February). Visitors who still click after an Overview sometimes stay longer and convert better — First Page Sage saw session duration and lead rate up even as raw clicks fell. Brands cited inside an Overview can outperform uncited competitors on the same page. None of that restores the missing volume. Among outbound clicks that remain when an Overview is present, only a small fraction come from the AI block itself; the rest still come from classic blue links on a smaller pile of traffic.
AI Mode is more closed still. Some trackers put zero-click on AI Mode queries in the low-to-mid 90s. That product is still a sliver of all Google searches, but it is the direction of the interface.
Standalone LLM bots
ChatGPT, Perplexity, Gemini, Claude, Grok, and Copilot are a different disintermediation path. The user never loads a results page. The model either answers from weights or runs its own hidden search (“fan-out” queries), reads pages, and returns a synthesis.
Volume is large and still not Google. ChatGPT crossed roughly 900 million weekly active users in February 2026 and about 1 billion weekly users by August 2026, with daily prompt counts cited around 2–2.5 billion. That is not 2.5 billion web searches. A large share is drafting, coding, tutoring, roleplay, and multi-turn follow-ups. A growing share is search-shaped: how-to, product comparison, current events, “best X in 2026.” Perplexity last published 780 million queries in a single month (May 2025, about 30 million a day) and was compounding quickly; later industry estimates put monthly query volume in the low billions. Gemini’s consumer app and Google’s own AI Mode blur the line between “search engine with an answer box” and “chatbot with a search index.”
One Q1 2026 compilation put combined AI-assistant prompt volume near 4.7 billion a day against about 8.5 billion Google searches in that author’s series — a ratio of roughly 1.8-to-1, versus about 8-to-1 two years earlier. Treat the point estimate as directional. The trend is not: chat is no longer a rounding error next to search. It is a parallel information market, especially for long-form research, coding, how-to, and comparison queries, where AI-assistant share has jumped double-digit points since 2024.
Those systems do not browse the way a human does. Studies of LLM search engines find fewer sources per answer (often three to five domains), heavy reuse of Wikipedia and a short list of encyclopedic or news domains, and large non-overlap with Google’s top 10. Grok and Gemini in some academic samples cited no external website in a substantial share of responses. ChatGPT Search is widely understood to lean on Bing’s index, which means a site invisible to Bing is invisible to that product regardless of its Google rank. Perplexity runs its own snippet-level index. The “search” happening inside the bot is real. The user still does not visit the page.
What publishers actually receive
Referral data is the cold shower after the query-volume charts.
Cloudflare Radar snapshots in 2026 still attribute on the order of 87–88 percent of search referral traffic to Google, with ChatGPT, Gemini, Claude, and Perplexity combined around 0.3 percent. AI referrals to publisher networks grew by large multiples off a tiny base and then, in some 2026 Chartbeat cuts, flattened or declined across the major chat brands. ChatGPT remains the best of a small set.
The damage is on the Google side of the ledger. Chartbeat’s 2026 Publisher Playbook recorded a 40.2 percent year-over-year decline in Google Search referrals across its network from July 2025 to July 2026. Search fell from about 9 percent of pageviews in July 2024 to 5 percent in July 2026. Discover fell 28 percent. Direct, internal, and dark social took share. Press Gazette and related publisher surveys have put global Google Search traffic to news sites down about a third year over year, with U.S. figures worse, and a three-year expected decline north of 40 percent. First Page Sage’s 2026 site-level panel saw organic search’s share of sessions drop from 51.3 percent in January 2023 to 42.8 percent in July 2026 (−24 percent session volume), while “AI platforms” rose from 0.1 percent to 6.2 percent of sessions in that particular dataset — a reminder that “AI traffic” can mean citations, branded navigations after an answer, and misattributed direct visits as much as clean referrers.
Gartner’s widely repeated forecast that traditional search engine volume would drop 25 percent by 2026 as answer engines absorbed queries has not shown up as a collapse in Google’s top-line query count. It has shown up as a collapse in the share of those queries that graduate into an open-web session.
Putting the two pictures on one page
| Layer | Stock or flow | 2026 working number |
|---|---|---|
| Hostnames that respond | Stock | ~1.50 billion |
| Maintained websites | Stock | ~217 million |
| Registered domains | Stock | ~402 million |
| Google searches | Daily flow | ~13–14 billion |
| Google queries that trigger an AI Overview | Share of that flow | ~40% central estimate; 16–60% by panel |
| U.S. searches that end with no click | Share | ~68% |
| AIO pages that end with no click | Share | ~80% |
| ChatGPT-class prompts | Daily flow | ~2–2.5 billion |
| AI-chat referrals vs Google referrals | Share of referrals | ~0.3% vs ~87%+ |
| Publisher Google Search traffic | YoY change | About −33% to −40% |
A back-of-the-envelope that stays honest about uncertainty: if Google runs ~14 billion searches a day and AI Overviews appear on ~40 percent of them, that is on the order of 5–6 billion Google queries a day that already meet an answer box. Not all of those would have produced a click in 2019 either — zero-click was already near half — but the incremental suppression is large. Layer on knowledge panels, featured snippets, People Also Ask, and Maps, and most informational intent on Google never leaves Google. Add a few billion daily LLM prompts, only some of which are search substitutes, and the number of “questions about the world that used to start with a blue link” now answered in place is several billion per day.
That does not mean 217 million websites are obsolete. Navigational queries (“facebook,” “gmail,” “chase”), transactions, local intent, video, forums, and brand destinations still move people onto sites. AI Overviews themselves cite an average of about five sources. Being one of those five is a new ranking problem. It is not the same business as ranking number one on a clean ten-blue-links page. First Page Sage found that an Overview compresses the value gap between positions: position one is worth about 8 times position ten instead of 13 times. The prize shrank and flattened at the same time.
What this means — and what it does not
Three mistakes dominate the commentary. The first is treating 1.5 billion websites as a map of culture. Most of that number is unused inventory. The living web is two hundred million sites and change, and the attended web is two or three orders of magnitude smaller than that.
The second is declaring that “AI replaced Google.” Google’s query volume is at or near highs. Google’s own answer layer is the primary replacement for Google’s old results page. ChatGPT is a real second front, especially for work and research tasks, but it still sends almost no referral traffic relative to Search. The publisher recession is a Google-SERP recession first.
The third is assuming the missing click is a missing reader in a simple sense. Some of those people are better served. A weather query, a definition, a dosage, a conversion, a “what time does the game start” does not need a homepage. The problem is the queries that did need original reporting, specialist documentation, or a living community — and now resolve to a fluent paragraph trained on that work, with a citation the user does not follow. The economic loop that paid for the next article, the next man page, the next forum answer is the thing that is thinning out. Chartbeat and Press Gazette are measuring that thinning. Academic work on “answer bubbles” is measuring the other cost: different systems, given the same question, retrieve different sources, hedge less than classic search snippets, and overweight a short list of canonical domains.
For anyone who still publishes on the open web, the operational read is narrow and unromantic. Indexing is pickier. Informational SERPs are answer-first. Citations inside models and Overviews matter more than a raw #1 pack. Bing visibility matters if ChatGPT Search is a channel you care about. Direct audience — email, podcasts, Discord, church bulletins, campus listservs — anything that does not have to win a lottery on a results page is insurance. The 1.5 billion websites will keep growing. The number that a human being, or a model acting for a human being, actually consults will not grow with them.
Notes on the numbers
- Netcraft counts hostnames that respond, not “websites a person would recognize.” Active-site methodology is Netcraft’s, as relayed by Siteefy, Sunny Patel, SQ Magazine, Copahost, and others. July 2026 is the latest cluster of matching figures (1,494,915,628 total / 216,702,733 active).
- Verisign DNIB Q2 2026: 401.6 million registrations. This is a registry total, not a content total.
- DomainsProject, September 22, 2026: 289.7 million websites on 12.7 million IPv4 addresses; 1,025 addresses serve half the corpus.
- Google volume: company disclosure of 5 trillion-plus annual searches (2025), echoed by Foursets, Affinco, FOSS Post, OmniBound. Secondary range 8.5–16.4 billion/day.
- AIO coverage: Similarweb / TechCrunch / The Keyword (~43%); Comscore / Search Engine Land (39.4% U.S. desktop, June 2026); BrightEdge (~48% Feb 2026); Xponent21, Conductor, Semrush, Ahrefs for the rest of the spread. Industry cuts from BrightEdge and HasData.
- CTR / zero-click: Seer Interactive; First Page Sage 2026 AI Impact report; Pew Research; randomized AIO-hidden studies summarized by WebProNews and Digital Content Next; SparkToro / Search Logistics compilations for the 68% U.S. zero-click figure.
- Publisher traffic: Chartbeat 2026 Publisher Playbook (Search Engine Journal, Sept 24, 2026); Press Gazette / Reuters Institute recaps; First Page Sage session-mix table.
- LLM volumes: OpenAI WAU and prompt-count disclosures via Axios/TechCrunch and later recaps; Perplexity CEO figures for May 2025; Similarweb chatbot traffic; SparkToro “search happens everywhere” desktop share work; academic LLM-SE citation studies (arXiv 2512.09483 and follow-ons).
- PAA absorption: Search Engine Roundtable, September 9, 2026, citing AlsoAsked and Allintitle.
Selected sources
Netcraft Web Server Survey via Siteefy, SQ Magazine, Sunny Patel, Terms and Conditions Template, Copahost, HostingGuider, MuseMind (March–August 2026 recaps). Verisign Domain Name Industry Brief, Q2 2026. DomainsProject, “How Many Computers Host the Web?,” September 22, 2026. Semrush / Search Engine Land most-visited sites, August 2026.
Google volume and share: Foursets “150+ Google Search Statistics 2026”; Affinco; FOSS Post; OmniBound; StatCounter via those recaps; Comscore Q2 2026 desktop search totals.
AI Overviews: Similarweb analysis reported by TechCrunch (July 27, 2026) and The Keyword; Comscore Q2 2026 AI Intelligence Report via Search Engine Land (September 22, 2026); BrightEdge one-year mark; SurferSEO 405,576-overview study; WordsAtScale CTR compilation; Seer Interactive; First Page Sage “AI Impact on Website Traffic: 2026 Report”; Pew Research click study; Search Engine Roundtable on PAA and on the 43–48 percent band.
Publisher impact: Chartbeat 2026 Publisher Playbook via Search Engine Journal (September 24, 2026); Digital Content Next on experimental AIO traffic effects (September 15, 2026); WebProNews on the same experiment family.
LLM search behavior: Peec AI fan-out analysis; Swfte 2026 query-construction note; Georgetown / related “beyond search” user studies; arXiv papers on LLM search engines and answer bubbles.
Comments
Post a Comment