More unique IPs than any network on the planet, in the world's largest residential pool. That's Shifter.More unique IPs than any network, in the world's largest pool. That's Shifter.

See the numbers

How Much Web Traffic Is Bots? Why the Big Reports Disagree

Bot traffic estimates range from 10% to 53%. We lined up Thales, Cloudflare, Akamai, Fastly, F5 and Wikimedia to show what each counts and which to trust.

Data Science · Updated November 18, 2026 · 6 min read
share of traffic from botsThales53%Fastly49%Cloudflare, HTML pages48%Akamai42%Wikimedia35%Cloudflare, all requests31%F510%
One question, seven answers. Each amber bar is one report, and each counts something different.
Knowledge
On this page
  1. The numbers side by side
  2. Why they disagree
  3. Almost everyone measuring is selling
  4. How much of it is AI
  5. Which number to use
  6. What this means for data teams
  7. The bottom line
  8. Sources and references

“Bots now outnumber humans on the internet” is one of the most repeated statistics in tech. It is also one of the least precise. Depending on which report you read, bots make up 10%, 31%, 42%, 49% or 53% of web traffic. None of these numbers is wrong. They are answers to different questions, and the difference matters for anyone who builds or defends against bot walls.

We lined up the main published figures, checked each against its source, and noted exactly what it counts.

The numbers side by side

SourceBot shareWhat it countsPeriodWhose traffic
Thales (Imperva)53% (40% bad, 13% good)“Observed web traffic”2025Its customers’ sites
CloudflareAbout 48% (44% non-AI bots, 4.2% AI bots)Requests for HTML pages2025Its network
Cloudflare31.2%All application requestsApr 2023 to Mar 2024Its network
Fastly49%RequestsJanuary 2026Its security customers
Akamai42%“Overall web traffic”Published June 2024Its platform
WikimediaAbout 35%PageviewsEarly 2025Wikipedia and sister sites
F5 Labs10.2%Web and API transactionsNov 2023 to Sep 2024Customers already using its bot defence

Why they disagree

They count different things. Requests for HTML pages, all requests, pageviews and transactions are different units. A human visit to one page can trigger dozens of requests for images, scripts and API calls; a scraper may fetch only the HTML. Cloudflare shows how big the gap is within one company: about half of HTML requests, but 31% of all application traffic. Fastly found bots made 47% of requests served from cache and 60% of requests that reached the origin.

They measure their own customers. Each vendor sees the sites that pay it, which are not a sample of the web. HUMAN Security says plainly that its data “does not represent the totality of internet traffic”. F5’s 10.2% comes from customers already running bot defence; F5 notes that sites without it often see far more.

They define bots differently. Thales counts AI crawlers as benign automation alongside search engines. Cloudflare reports AI bots and other bots separately. Fastly decided “not to reclassify general-purpose search engine crawlers as AI bots”. The same request can land in a different bucket in each report.

They only count what they catch. Thales notes its analysis is “based on detectable traffic, meaning systems that identify themselves or trigger security controls”. Wikimedia shows what that misses: in 2025 it found bots built to evade detection had been counted as people, and once it reclassified them, human pageviews were down about 8% on the year before.

The numbers move quickly. Fastly reported bots at roughly 29% of requests in the third quarter of 2025, and 49% in January 2026. Changes in customers, detection and real bot activity all feed into a single headline figure.

Almost everyone measuring is selling

Thales, Cloudflare, Akamai, Fastly, F5 and HUMAN all sell bot management. That does not make their data wrong, and their networks see more traffic than anyone else’s. But they have a reason to publish big numbers, and none releases enough method to reproduce the result. Wikimedia is the main exception: a non-profit reporting on its own sites, which found bots made about 35% of pageviews but at least 65% of its most resource-heavy traffic.

How much of it is AI

Less than the headlines suggest, but growing fastest.

SourceAI shareMeasured as
Cloudflare4.2%Average share of HTML requests, 2025
AkamaiNearly 1%Share of bot traffic on its platform
FastlyAbout 8%Share of wanted bot traffic, January 2026

The growth figures are steeper. Cloudflare saw AI crawling for user actions rise more than 21 times during 2025, and HUMAN measured automated traffic growing 23.51% year on year, roughly eight times faster than human traffic. Our piece on the bot wall AI agents face covers how sites are responding.

Which number to use

If your question isUse
What share of page loads are automatedCloudflare’s HTML share, about half
How much of all request volume is botsCloudflare’s 31% or Fastly’s monthly figure, with dates
How big is the security problemThales’ bad bot share, 40%, as a vendor estimate
What bots cost open infrastructureWikimedia’s 35% of pageviews and 65% of heavy traffic
What is happening on your siteYour own logs

Always quote the unit and the source with the number. “Bots are 53% of traffic” is a vendor’s figure for its customers’ observed traffic, not a census of the internet. Where a report gives no unit or period, as with Akamai’s 2024 release, treat it as the weakest figure in the set.

What this means for data teams

The reason these reports exist is the reason collection is getting harder: bot management is now everywhere, and it is tuned by the same companies publishing the numbers. Sites that can identify the anti-bot stack in front of a target can plan for it, and teams that identify themselves, respect robots.txt and keep request rates low are the ones least likely to be counted among the “bad bots”. Residential proxies help legitimate collection see pages as local visitors do, but they do not change the rules a site sets.

The bottom line

Bots are somewhere between a third and a half of web requests, depending on how you count, and AI crawlers are still a single-digit share of that, growing fast. The big reports disagree because they measure different units, different customers and different definitions, and most come from companies selling protection. Quote them with their units, treat them as direction rather than census, and trust your own logs for your own site.

Sources and references

  • Thales, 2026 Bad Bot Report (Imperva): imperva.com
  • Cloudflare, Radar 2025 Year in Review: blog.cloudflare.com
  • Cloudflare, Application Security report, 2024 update: blog.cloudflare.com
  • Fastly, Threat Insights Report, Q1 2026: learn.fastly.com
  • SiliconANGLE, Fastly report on bot traffic in Q3 2025 (18 December 2025): siliconangle.com
  • Akamai, bots compose 42% of web traffic (25 June 2024): akamai.com
  • Akamai, AI bots threaten web-based business models (4 November 2025): akamai.com
  • F5 Labs, 2025 Advanced Persistent Bots Report: f5.com
  • HUMAN Security, 2026 State of AI Traffic and Cyberthreat Benchmark: humansecurity.com
  • Wikimedia, How crawlers impact the operations of the Wikimedia projects (1 April 2025): diff.wikimedia.org
  • Wikimedia, New user trends on Wikipedia (17 October 2025): diff.wikimedia.org

Frequently asked questions

What percentage of internet traffic is bots?

It depends on what you count. Thales' 2026 Bad Bot Report puts automated traffic at 53% of observed web traffic in 2025. Cloudflare found bots made about half of requests for HTML pages in 2025, but earlier measured 31% of all application traffic. Fastly reported 49% of requests in January 2026. A defensible answer is that bots make up roughly a third to a half of requests, depending on the measure.

Why do bot traffic reports give different numbers?

They count different things: HTML page requests, all requests, pageviews or transactions. They measure their own customers' traffic, which differs by vendor. They classify AI crawlers and search engines differently. And they can only count the bots they detect, so evasive bots are counted as humans.

How much web traffic comes from AI crawlers?

Still a small share of the total. Cloudflare measured AI bots at an average of 4.2% of HTML requests in 2025, and Akamai said AI bots made up nearly 1% of the bot traffic on its platform. The growth is fast, though: Cloudflare reported AI crawling for user actions up more than 21 times during 2025.

Are bot traffic reports reliable?

They are useful but not neutral. Nearly every major report comes from a company that sells bot protection, each measures only its own customers, and none publishes enough method to be reproduced. Use them for direction and orders of magnitude, and use your own logs for decisions about your site.

Ready to get started?

Try Shifter's residential proxies, 205M+ IPs, 195+ countries, from $0.10/GB.

Get Started