Google sent 87.52% of all search referral traffic observed by Cloudflare Radar in the 28 days ending April 20, 2026. Bing sent 3.46%. TikTok sent 3.19%. Yandex 2.31%. DuckDuckGo 1.34%. Every AI chatbot put together (ChatGPT, Gemini, Claude, Perplexity) sent 0.27%. Meanwhile, Anthropic's ClaudeBot crawled 13,528 pages for every 1 human visit Anthropic sent back. OpenAI's crawl-to-refer ratio is 1,252:1. Google's is 5:1.

I spent the last 18 months at Microsoft Bing and then on the AI team before co-founding TechnologyChecker, so search engine market share is a chart I've stared at for a long time. The version of that chart I grew up with is a pie slice labeled "Google" next to slivers labeled "Bing," "Yandex," and "DuckDuckGo." Almost every SEO blog still publishes that version. In 2026, it isn't enough.

The real story in this year's search data isn't whether AI is "taking share" from Google. It isn't. The real story is that AI platforms invented an entirely new web traffic pattern: enormous outbound crawling with near-zero inbound referrals. The asymmetry is so extreme it has changed what "search engine market share" even means in practice.

This post is the full breakdown for April 2026, pulled live from the Cloudflare Radar API across 28 days of global HTTP traffic. Who sends traffic, who takes it, and what that means if you run a website, do SEO, or build AI.

The topline: Google still owns 87.5% of search referrals in April 2026

The first chart is the one most people came for. It answers the question every marketing dashboard tries to answer. When someone lands on a website from a search engine, which search engine did they come from?

According to Cloudflare Radar's get_bots_crawlers_data endpoint (REFERER dimension, 28-day window ending April 20, 2026), the split is:

Search Source	Referral Share	Category
Google	87.52%	Traditional search
Bing	3.46%	Traditional search
TikTok	3.19%	Social search
Yandex	2.31%	Traditional search
Baidu (mobile + desktop)	1.63%	Traditional search
DuckDuckGo	1.34%	Privacy search
ChatGPT	0.20%	AI search
Bing China	0.17%	Traditional search
Google Scholar	0.10%	Vertical search
Gemini	0.035%	AI search
Claude.ai	0.019%	AI search
Perplexity.ai	0.018%	AI search

Source: Cloudflare Radar get_bots_crawlers_data endpoint, REFERER dimension, March 23 to April 20, 2026.

Two things jump out.

Google's 87.52% is one of the most concentrated market shares in any major internet metric. For context, global mobile OS splits 72/27 Android to iOS, and cloud infrastructure splits 31/25/11 AWS/Azure/Google. A single company sending 87.52% of the traffic-sending layer of the web is historically rare.

The second thing is the AI chatbot column. Combining every major model (ChatGPT at 0.20%, Gemini at 0.035%, Claude at 0.019%, Perplexity at 0.018%) gets you to 0.272% of search referrals. That's one-third of what DuckDuckGo sends, one-twelfth of what Bing sends, and one 322nd of what Google sends. The "AI is replacing search" narrative has been loud since 2023. In the Cloudflare referral data for April 2026, it isn't yet real.

TikTok is now a bigger search engine than DuckDuckGo and Yandex

The single most interesting row in that table is TikTok at 3.19%. That isn't page-load traffic or in-app traffic. It's traffic where someone tapped a link inside TikTok and ended up on an external website. That's the classic definition of a search referral.

Three years ago, TikTok didn't meaningfully exist in the search referrer table. In April 2026, it sends more traffic to the rest of the web than Yandex (2.31%), DuckDuckGo (1.34%), and Baidu (1.63%) send individually. It sits just behind Bing.

The mechanism is "social search." Young users open TikTok and type "best Italian restaurant Brooklyn" into the app's search bar the way their parents opened Google. Prabhakar Raghavan, Google's then-head of Search, said publicly at Fortune's Brainstorm Tech in July 2022 that nearly 40% of 18-to-24-year-olds in Google's internal studies now use TikTok or Instagram for local discovery instead of Google Maps. Cloudflare's 2026 referral data suggests that behavior now drives roughly 3% of the web's search-driven traffic.

If you run SEO for a consumer brand, TikTok isn't an awareness channel anymore. It's a search engine.

How has search market share moved over the last 12 months?

To check whether the April 2026 topline is a snapshot or a trend, I pulled the REFERER dimension across a 52-week window and sampled the first full ISO week of every month. Here's the 13-month trajectory.

Month (first full week)	Google	TikTok	Bing	Yandex	DuckDuckGo
Apr 2025	83.61%	8.66%	2.75%	2.43%	1.04%
May 2025	81.51%	9.92%	2.53%	2.50%	1.01%
Jun 2025	81.46%	9.31%	2.69%	2.60%	1.06%
Jul 2025	80.31%	12.22%	2.41%	1.47%	1.09%
Aug 2025	81.48%	10.85%	2.38%	1.50%	1.03%
Sep 2025	81.86%	9.77%	2.97%	1.59%	1.13%
Oct 2025	80.21%	11.16%	2.79%	1.67%	1.08%
Nov 2025	80.94%	10.59%	2.88%	1.64%	1.14%
Dec 2025	82.50%	9.62%	2.76%	1.69%	1.13%
Jan 2026	80.21%	12.17%	2.36%	1.63%	1.25%
Feb 2026	81.14%	11.08%	2.84%	1.61%	1.15%
Mar 2026	86.69%	3.95%	3.91%	1.74%	1.31%
Apr 2026	86.96%	3.56%	3.45%	2.46%	1.39%

Source: Cloudflare Radar get_bots_crawlers_data endpoint, REFERER dimension, weekly timeseries_groups across 52 weeks ending April 20, 2026.

For ten straight months, Google held between 80.21% and 82.50%. TikTok held between 9.31% and 12.22%. Bing, Yandex, and DuckDuckGo barely moved at all. Then, in the reporting week of February 16, 2026, something happened. Google jumped roughly 6 points to 86-88%. TikTok collapsed from 11.08% to 3.95% in a single window.

That kind of 7-point monthly swing in a traffic share number of this scale doesn't happen organically. It's almost certainly the result of Cloudflare reclassifying a block of TikTok link-out referrals that the platform used to bucket as "search." No public Cloudflare changelog confirms the change as of publication, but the signature is unmistakable. The TikTok loss almost exactly offsets the Google gain (TikTok dropped 7.1 points, Google gained 5.6 points, with the remainder redistributing to Bing and DuckDuckGo). The timing is synchronous. And the rest of the world didn't move.

The practical takeaway: if you've been tracking your "TikTok search share" in any analytics tool that sources from Cloudflare Radar, your number before February 2026 is probably overstated, and your number after is more conservative. Treat the two halves of the year as two different methodologies.

Month-over-month is effectively flat

Zooming into the last two comparable 28-day windows (March 23 to April 20 vs February 23 to March 23, both post-reclassification), the changes are tiny:

• Google: 87.61% to 87.52%, down 0.09 points
• Bing: 3.58% to 3.46%, down 0.11 points
• TikTok: 3.51% to 3.19%, down 0.32 points
• Yandex: 1.85% to 2.31%, up 0.45 points
• DuckDuckGo: 1.41% to 1.34%, down 0.07 points
• ChatGPT: 0.21% to 0.20%, down 0.007 points

Google moved by less than a tenth of a point. ChatGPT's share actually fell by a rounding error. The biggest gainer is Yandex, which is a geopolitical story about Russian internet isolation, not an AI search story.

When someone shows you a slide predicting 20% AI-search-referral share by 2027, this is the chart to hold it up against. The referral data is moving glacially. Whatever AI is doing to how users consume information, it isn't yet changing how traffic lands on websites.

The other half of the story: the crawl-to-refer ratio flips everything

Here's where the data gets uncomfortable.

I pulled the CRAWL_REFER_RATIO dimension from the same get_bots_crawlers_data endpoint. That metric measures, per AI operator, how many HTML pages that operator's bots crawl for every 1 human visit the operator sends back. For the week of April 13 to 20, 2026:

Operator	Pages Crawled per 1 Referral	What This Means
Anthropic	13,528 : 1	Essentially pure extraction
OpenAI	1,252 : 1	Crawling 1,252x more than it returns
Perplexity	95 : 1	The best of the AI-native operators
Mistral	53 : 1	Small operator, still mostly extractive
Microsoft	29 : 1	Copilot + Bing blended
Yandex	18 : 1	Traditional regional search operator
ByteDance	10 : 1	TikTok + Bytespider training
Baidu	10 : 1	Traditional Chinese search
Google	5 : 1	Traditional search, the baseline
DuckDuckGo	1.5 : 1	The healthiest ratio of any major operator

Source: Cloudflare Radar get_bots_crawlers_data endpoint, CRAWL_REFER_RATIO dimension, April 13 to April 20, 2026.

DuckDuckGo's 1.5:1 shows what a healthy relationship looks like. DuckDuckGo crawls your site, indexes it, and for roughly every 1.5 pages it crawls, it sends back 1 human visitor. Google's 5:1 is also fine. For every page Googlebot fetches, Google Search sends back 1 referral for every 5 crawls. That ratio has held roughly steady for a decade. It's why publishers tolerated Googlebot even as it became the largest single source of bot traffic on the web.

Now look at Anthropic's 13,528:1.

For every 1 human visit Claude.ai sends back to a publisher, Anthropic's ClaudeBot has already fetched 13,528 pages from that publisher. That isn't a search relationship. It isn't even a bad search relationship. It's a data extraction pipeline with no compensating traffic flow.

OpenAI at 1,252:1 is 10x better than Anthropic but still 250x worse than Google. OpenAI runs GPTBot (training) plus OAI-SearchBot (live retrieval for ChatGPT), and the blended ratio is still extractive by an order of magnitude.

This is the structural reason publishers started blocking AI crawlers through robots.txt in 2024 and 2025. The economic bargain that let Googlebot access the open web (you crawl my content, you send me visitors) simply doesn't exist for most AI operators.

Humans are now a minority of web traffic

One last Cloudflare cut to set the context. When Cloudflare classifies every HTTP request by client type, the April 2026 split is:

• Non-AI Bots (search crawlers, SEO tools, monitoring, archivers, feed readers): 46.19%
• Humans: 45.20%
• AI Bots (training + agent crawlers): 5.58%
• Mixed-Purpose Bots: 3.03%

Source: Cloudflare Radar get_bots_crawlers_data endpoint, CLIENT_TYPE dimension, March 23 to April 20, 2026.

54.80% of HTTP requests observed by Cloudflare in April 2026 are non-human. Humans are the minority.

AI bots at 5.58% are still small relative to non-AI bots, but the composition of that 5.58% is what matters. Non-AI bots are the familiar ecosystem. Googlebot, Bingbot, SEO tools like Ahrefs and Semrush, monitoring services like Pingdom, RSS readers, archivers like the Wayback Machine. They mostly operate with the 5:1 or better ratio Google models. AI bots operate at 100:1, 1,000:1, or 13,000:1. The two categories aren't comparable, and combining them into "bot traffic" obscures the real economic shift.

Which AI crawlers are growing fastest?

Inside that 5.58% AI bot slice, the 28-day user agent breakdown looks like this:

User Agent	Share of AI Bot Requests	Operator	Primary Purpose
Googlebot	31.41%	Google	Search indexing + Gemini training
Meta-ExternalAgent	15.43%	Meta	Llama training
ClaudeBot	11.55%	Anthropic	Claude training
GPTBot	9.62%	OpenAI	GPT training
Applebot	9.50%	Apple	Apple Intelligence training
Bingbot	7.94%	Microsoft	Search + Copilot grounding
Bytespider	4.49%	ByteDance	Doubao training
Amazonbot	4.31%	Amazon	Alexa + Nova training
OAI-SearchBot	1.79%	OpenAI	ChatGPT live web retrieval
Other	3.96%	Various	—

Source: Cloudflare Radar get_ai_data endpoint, bots/summary/user_agent dimension, March 23 to April 20, 2026.

Googlebot at 31.41% shows the dual-use problem. Cloudflare classifies Googlebot as AI-adjacent because Google uses the same crawler output to power both Google Search and Gemini training. You can't block Googlebot-for-AI without blocking Googlebot-for-search. They're the same crawler. Google-Extended exists for exactly this reason, but it's opt-out and covers only a subset of training usage.

Applebot tripled share in 90 days. When we pulled the same data on a 90-day timeseries, Applebot's share rose from around 3% in late January to 9.50% by mid-April. That's the signature of Apple Intelligence scaling up its pre-training data collection. Nothing else on the chart has moved that fast.

Meta-ExternalAgent has no consumer search product behind it. Meta doesn't operate a search engine. 100% of Meta-ExternalAgent's crawl volume is Llama model training. By volume, Meta is now the second-largest AI crawler on the web despite having no referral mechanism to give back.

What are AI crawlers actually doing?

The final Cloudflare cut is the most structurally revealing. AI bots declare a crawl_purpose header. Summed across all AI crawler traffic for the 28-day window:

• Training: 50.69%
• Mixed Purpose: 39.35%
• Search: 7.28%
• User Action: 2.28%
• Undeclared: 0.40%

Source: Cloudflare Radar get_ai_data endpoint, bots/summary/crawl_purpose dimension.

Only 7.28% of AI crawler traffic is classified as search. That's a bot fetching a page in real time on behalf of a user prompt, the AI equivalent of "Googlebot indexing in response to a query." It's the only AI crawl category with any plausible referral mechanism behind it. The other 92.72% is training or mixed-purpose extraction with no direct path back to the publisher.

That's the crux of the 2026 search engine market share story. Traditional search engines run a value exchange. They crawl, they index, they send traffic back. AI operators, on aggregate, run at the 50:1 to 13,000:1 ratio because they aren't running a search engine at all. They're running a training pipeline that occasionally (7.28% of the time) does a little search.

What this means if you do SEO, run a website, or build AI

For SEOs, the data has two simple implications. Google's share isn't moving because of AI, and that's unlikely to change in the next year. Google still sends 87.52% of search referrals, and that number moved 0.09 points in a month. Separately, your TikTok traffic share matters in a way it didn't 24 months ago. Any consumer-facing brand without a TikTok SEO plan is leaving roughly 3% of search-driven traffic on the table. The AI chatbot referral channel (ChatGPT, Gemini, Claude, Perplexity combined) is 0.27% of search referrals. Worth monitoring. Not yet worth optimizing for the way Bing or Yandex are.

For content websites, the crawl-to-refer numbers are the story. You're paying origin bandwidth to serve ClaudeBot at 13,528 pages per 1 return visit. You're paying OpenAI's training pipeline at 1,252:1. Some publishers have done the math and started blocking. Cloudflare's own data shows roughly 40% of top websites now block at least one major AI crawler.

For AI builders, the asymmetry is also your problem in the medium term. Training data access is tightening. Every month another large publisher adds a new crawler to its robots.txt disallow list. 13,528:1 isn't sustainable. Either the industry builds genuine referral flows back to publishers, or the open web's willingness to be crawled keeps eroding and the training data pipeline narrows.

Methodology and source data

All data in this post was pulled from the Cloudflare Radar API via the Radar MCP server between April 20 and April 21, 2026.

• Search engine referral share (topline): get_bots_crawlers_data endpoint, REFERER dimension, 28-day window, March 23 to April 20, 2026. Control window: 28dcontrol (February 23 to March 23, 2026) for month-over-month.
• 12-month search share trend: get_bots_crawlers_data endpoint, REFERER dimension, timeseries_groups format, 52-week window, April 21, 2025 to April 20, 2026. One data point per first full ISO week of each month.
• AI platform crawl-to-refer ratios: get_bots_crawlers_data endpoint, CRAWL_REFER_RATIO dimension, 7-day window, April 13 to April 20, 2026. The longer windows returned an API error at query time, so the 7-day window is the operative one.
• Visitor type split: get_bots_crawlers_data endpoint, CLIENT_TYPE dimension, 28-day window.
• AI crawler volume: get_ai_data endpoint, bots/summary/user_agent dimension, 28-day window.
• AI crawler purpose: get_ai_data endpoint, bots/summary/crawl_purpose dimension, 28-day window.
• Applebot 90-day growth: get_ai_data endpoint, bots/timeseries_groups/user_agent dimension, 90-day window, January 20 to April 20, 2026, PERCENTAGE normalization.

Cloudflare Radar's sample covers traffic across Cloudflare's global network of 330 cities in 125+ countries, including sites that use Cloudflare as a reverse proxy, CDN, or authoritative DNS. The sample is large (hundreds of billions of requests per day) but tilts toward the Cloudflare customer base: Western web traffic, e-commerce, media, and SaaS. Search referrer data specifically only measures referrals to Cloudflare-protected origins, so regional search engines that send traffic exclusively to non-Cloudflare destinations are under-represented. The February 2026 discontinuity in the TikTok/Google trend line is consistent with a Cloudflare-side reclassification rather than a real behavior change. Read the trend section with that caveat in mind.

No referrer data was filtered by geography, vertical, or ASN. All numbers above are global, unfiltered totals.

---

Data source: Cloudflare Radar (radar.cloudflare.com). Endpoints get_bots_crawlers_data (REFERER, CRAWL_REFER_RATIO, CLIENT_TYPE, USER_AGENT dimensions) and get_ai_data (bots/summary/user_agent, bots/summary/crawl_purpose, bots/timeseries_groups/user_agent dimensions). Pulled April 20 to April 21, 2026.