Anti-Bot Technology Compared: What Each Vendor Does
Most large websites now run bot protection, and the products work in very different ways. We read each vendor's own documentation, plus independent research, to compare how 15 products detect bots, what they do about them, and where each falls short.
Two vendors dominate. On the 179 major US sites we tested, Akamai appeared on 44 and Cloudflare on 41, far ahead of DataDome, Imperva and HUMAN[13].
The products differ most in where they sit. Some run at a CDN edge (Cloudflare, Akamai, Fastly), some rely on JavaScript in the browser (HUMAN, DataDome, Kasada, F5), and some watch only server or network traffic (Netacea, Cequence).
CAPTCHAs are fading. Kasada and F5 use none[20][22], Cloudflare's Turnstile replaces puzzles with invisible checks[3], and research shows AI can solve classic image CAPTCHAs[29].
AI crawlers now get their own rules. Cloudflare and AWS block AI bots by default and can charge them[6][24]. Akamai splits AI traffic into training, search and agent categories[14].
Verified identity is the new front door. Kasada and AWS support Web Bot Auth, a way for legitimate bots to prove who they are with cryptographic signatures[21][24].
Most enterprise vendors don't publish prices. Only Cloudflare's plans, AWS, Google reCAPTCHA, hCaptcha, Friendly Captcha and Azure list prices publicly.
How bot detection works
Most products combine several of these layers:
Network reputation: where the request comes from, such as a data center, a known proxy or a residential connection.
Connection fingerprints: how the client sets up its encrypted connection (TLS fingerprints such as JA3/JA4) and formats its requests. Browsers and scripts tend to look different here.
Browser checks: JavaScript that inspects the device and browser to see whether it's real or automated.
Behavior: how a visitor moves through the site, and how fast and how regularly.
Machine-learning scores that combine the signals into one risk score.
Responses: allow, block, rate-limit, show a challenge or CAPTCHA, slow the connection down (tarpit), or serve decoy content.
Verified identity: good bots and AI agents prove who they are, through published IP lists or cryptographic signatures.
Who's most common
Anti-bot vendors detected on 179 major US websites
Marquee Data survey, October 2026. Some sites run more than one[13].
Scores every request from 1 (almost certainly automated) to 99 (almost certainly human)[1]. A machine-learning engine makes most detections, and a JavaScript engine spots headless browsers[2]. Turnstile is its CAPTCHA alternative, using invisible checks such as proof-of-work and proof-of-space[3].
Strengths
Free and low-cost tiers: Bot Fight Mode is free, and Super Bot Fight Mode comes with Pro, Business and Enterprise plans[4][5]
Blocks AI crawlers by default since July 2025, and can charge them through Pay Per Crawl[6][7]
AI Labyrinth feeds AI-generated decoy pages to unwanted crawlers[8]
Limits
TLS fingerprints (JA3/JA4) are available only to Enterprise Bot Management customers[9]
Bot Fight Mode can't be adjusted with custom firewall rules[4]
A bad Bot Management file caused the November 18, 2025 outage, during which some traffic got a bot score of zero[10]
Akamai
CDN edge · Bot Manager · Content Protector
Bot Manager uses AI models for behavior analysis and browser fingerprinting and acts at Akamai's edge. Its behavior detection works by injecting a script into each monitored page[11]. Content Protector targets scrapers specifically, with headless-browser detection and responses including monitor-and-deny, tarpits and challenges[12].
Strengths
The most common vendor in our survey of major sites: 44 of 179[13]
Separate AI categories since September 2026: training crawlers, search crawlers, and fetchers and agents, so each can get its own policy[14]
Tarpits slow scrapers down instead of only blocking them
Limits
No public pricing found
Behavior detection depends on a script injected into pages
HUMAN (formerly PerimeterX)
JavaScript sensor + cloud detector + enforcer
A JavaScript sensor in the browser sends data to a cloud detector that scores it with machine learning and behavioral analytics. An enforcer installed on the app, load balancer or CDN then allows, blocks or challenges[15].
Strengths
Enforcer can sit on the app, load balancer or CDN, so it isn't tied to one network
Needs both a browser sensor and a server-side enforcer
No public pricing found
DataDome
JavaScript device check · server modules
Device Check runs JavaScript in the browser to collect hundreds of signals, working like a CAPTCHA without showing one. When it needs more evidence, it escalates to a visible CAPTCHA[16]. It also scores AI agents with a trust score[17].
An academic study sent half a million requests from 20 commercial bot services at a test site protected by DataDome, and found an average evasion rate of 52.93%[18]
No public pricing found
Imperva (Thales)
Advanced Bot Protection
Combines client checks, behavior analysis, machine learning and connection characteristics[19].
Strengths
Says legitimate users don't see a CAPTCHA on 99.999% of requests[19] (vendor claim)
Client sensors · edge, proxy or backend integration
Hundreds of hidden sensors look for traces of automation in the browser, and its code runs in an obfuscated virtual machine to prove it really executed. Kasada says it uses no CAPTCHA and classifies sessions in under 2 milliseconds[20].
Strengths
No CAPTCHAs and no rules for customers to manage[20]
AI Agent Trust supports the Web Bot Auth standard for verified agents[21]
Limits
Speed and detection figures are vendor claims
No public pricing found
F5 Distributed Cloud Bot Defense (formerly Shape)
JavaScript + mobile SDK
Collects telemetry with JavaScript in browsers and a native SDK in mobile apps, instead of using CAPTCHAs[22].
Strengths
Covers mobile apps through its SDK
No CAPTCHAs for users
Limits
Mobile protection requires building the SDK into the app
No public pricing found
AWS WAF Bot Control
AWS WAF managed rules · CloudFront and regional
"Common" level catches bots that identify themselves; "Targeted" adds browser interrogation, fingerprinting and behavior heuristics for bots that don't[23]. A dedicated rule blocks AI bots by default, it accepts cryptographic identity from legitimate AI agents (Web Bot Auth), and it can answer AI bots with HTTP 402 Payment Required[23][24].
Strengths
Published pricing: Common includes 10 million requests a month free, Targeted 1 million[25]
AWS documents false positives, for example mobile apps with non-browser user agents[26]
Targeted protection needs AWS's JavaScript or mobile SDKs to work well
Google reCAPTCHA
Script on the page · score-based (v3) or challenge (v2)
An invisible, score-based check[27]. reCAPTCHA v3 returns a score where 1.0 is very likely a person and 0.0 very likely a bot, and the site decides what to do[28].
Strengths
Free up to 10,000 assessments a month, then $8 flat up to 100,000[27]
Researchers at ETH Zurich solved 100% of reCAPTCHA v2 image challenges with a fine-tuned vision model[29]
A 13-month study of over 3,600 users concluded reCAPTCHA v2 has "an immense cost and no security"[30]
hCaptcha
Script on the page · CAPTCHA and enterprise risk scoring
A CAPTCHA service with free, Pro and Enterprise tiers[31].
Strengths
Published pricing: Pro at $139 a month (or $99 billed yearly) with 100,000 evaluations, then $0.99 per 1,000[31]
Limits
Claims WCAG 2.2 AA accessibility but tells sites to evaluate their own deployment[32]
Fastly Bot Management
CDN edge
Uses TLS client fingerprints (JA3/JA4), edge detection and optional JavaScript checks for headless browsers[33], plus a "defensive maze" that misleads bots[34].
Strengths
Runs at the edge with no page changes needed for basic detection
Deception options besides blocking
Limits
Pricing through sales only
Netacea
Server-side, no client code
Analyzes traffic on the server, with no code in the browser: "Never deploy agents again."[35]
Strengths
Nothing for visitors to download or run
Works across websites, apps and APIs
Limits
Without browser signals it relies on traffic behavior alone
No public pricing found
Cequence
Network-based, no client code
Detects bots from network traffic with "no client-side JavaScript or SDK integration". Responses include blocking, rate limiting, header injection and deception[36].
Strengths
No changes to web pages or apps
Covers APIs as well as websites
Limits
No public list pricing found
Friendly Captcha
Proof-of-work CAPTCHA
Has the visitor's device solve a cryptographic puzzle in the background, sets no cookies, and starts at €9 a month[37].
Strengths
No puzzles for people and no tracking cookies
Low published prices
Limits
Proof-of-work raises the cost of automation but doesn't identify who is behind it
Microsoft Azure WAF (Front Door)
Managed bot rules on Front Door Premium
Sorts bots into bad, good and unknown; by default bad bots are blocked, good bots allowed and unknown bots logged[38].
Strengths
Simple defaults managed by Microsoft
Limits
Its CAPTCHA doesn't work for AJAX or API calls[39]
We couldn't read Radware's or Arkose Labs' product pages reliably, so they aren't included.
The CAPTCHA problem
AI solves them: a fine-tuned vision model solved 100% of reCAPTCHA v2 image challenges, up from 68–71% in earlier research[29].
They cost people time: a 13-month study of over 3,600 users found reCAPTCHA v2 has "an immense cost and no security"[30].
They block disabled users: WebAIM's survey of screen-reader users ranks CAPTCHA as the most problematic thing on the web "by a notable margin"[42].
The replacement is invisible checks: proof-of-work puzzles, device signals and behavior scoring that most people never see[3][37].
The AI shift
Blocking by default: Cloudflare has blocked AI crawlers by default since July 1, 2025, unless they pay[6]. AWS has a rule that blocks AI bots by default[23].
Charging for access: both Cloudflare and AWS can answer AI bots with HTTP 402 Payment Required[7][24].
Telling AI bots apart: Akamai now separates training crawlers, search crawlers, and fetchers and agents[14]. DataDome scores individual AI agents[17].
Deception: Cloudflare's AI Labyrinth and Fastly's "defensive maze" waste unwanted bots' time with decoy content[8][34].
What it means if you need web data
Every protected site is different. The same vendor can be set up strictly on one site and loosely on another, so the only reliable test is the actual target.
Identity and permission matter more each year. Default AI blocking, verified-bot programs and paid access mean collection increasingly runs on agreed terms.
Legal risk tracks how barriers are handled. Our report on how AI companies collect web data covers recent rulings on CAPTCHAs and other access controls.
Methodology
We gathered these facts on November 15, 2026 from each vendor's own product pages and documentation, published pricing, peer-reviewed and preprint research, and our own survey of 200 major websites. Strengths and limits come from documented features, published research and vendor statements, which we label as vendor claims. We didn't test any product ourselves for this article. Products change often, so check the vendor's current documentation.
Marquee Data has built and managed custom web data collection since 2017 — Fortune 500 and hedge-fund clients, 100% US-based senior engineers, fixed fees. You don't pay until you're satisfied with the data.