You have seen it. Perhaps you did not register it consciously, however you’ve got seen it. That little widget asking you to confirm you are human. That full-page safety verify earlier than accessing an internet site. For those who’ve spent any time on the Web, you’ve got encountered Cloudflare’s Turnstile widget or Problem Pages — seemingly extra instances than you may rely.
The Turnstile widget – a well-known sight throughout tens of millions of internet sites
Once we say that a big portion of the Web sits behind Cloudflare, we imply it. Our Turnstile widget and Problem Pages are served 7.67 billion instances each single day. That is not a typo. Billions. This may simply be the most-seen consumer interface on the Web.
And that comes with huge duty.
Designing a product with billions of eyeballs on it is not simply difficult — it requires a basically totally different method. Each pixel, each phrase, each interplay has to work for somebody’s grandmother in rural Japan, a youngster in São Paulo, a visually impaired developer in Berlin, and a busy govt in Lagos. All on the similar time. In moments of frustration.
At this time we’re sharing the story of how we redesigned Turnstile and Problem Pages. It is a story advised in three components, by three of us: the design course of and analysis that formed our choices (Leo), the engineering problem of deploying modifications at unprecedented scale (Ana), and the measurable affect on billions of customers (Marina).
Let’s begin with how we approached the issue from a design perspective.
Half 1: The design course of
Let’s be trustworthy: no person likes being requested to show they’re human. You realize you are human. I do know I am human. The one one who would not appear satisfied is that little widget standing between you and the web site you are attempting to entry. At finest, it is a minor inconvenience. At worst? You have in all probability wished to throw your pc out the window in a match of rage. We have all been there. And nobody would blame you.
Turnstile built-in right into a login movement
Because the world warms as much as what seems to be an inevitable AI revolution, the necessity for safety verification is simply growing. At Cloudflare, we have seen a big rise in bot assaults — and in response, organizations are investing extra closely in safety measures. Meaning extra challenges being issued to extra finish customers, extra typically.
The numbers inform the story:
2023: 2.14B every day
2024: 3B every day
2025: 5.35B every day
That is a 58.1% common enhance in safety checks, 12 months over 12 months. Extra safety checks imply extra alternatives for finish consumer frustration. The extra corporations combine these verification programs to guard themselves and their prospects, the upper the prospect that somebody, someplace, goes to have a nasty expertise.
We knew it was time to take a tough have a look at our flagship merchandise and ask ourselves: Are we doing proper by the billions of people that encounter these experiences? Are we fulfilling our mission to construct a greater Web — not only a safer one, however a extra human one?
The reply, we found, was: we might do higher.
Earlier than redesigning something, we wanted to grasp what we had been working with. We began by conducting a complete audit of each state, each error message, and each interplay throughout each Turnstile and Problem Pages.
What we discovered wasn’t one of the best.
The state of inconsistency within the Turnstile widget. A number of states with no unified method
The inconsistencies had been evident. We had no unified method throughout the multitude of various error situations. Some messages had been overly verbose and technical (“Your device clock is set to a wrong time or this challenge page was accidentally cached by an intermediary and is no longer available”). Others had been too obscure to be useful (“Timed out”). The visible language assorted wildly — totally different layouts, totally different hierarchies, totally different tones of voice.
We additionally examined the suggestions we might obtained on-line. Social media, help tickets, group boards — we learn all of it. The frustration was palpable, and far of it was avoidable.
Take our suggestions mechanism, for instance. We supplied customers suggestions choices like “The widget sometimes fails” versus “The widget fails all the time.” However what is the distinction, actually? And the way had been they imagined to understand how typically it failed? We had been asking customers to interpret ambiguous choices throughout their most pissed off moments. The extra we left open to interpretation, the much less helpful the suggestions grew to become — and the extra frustration we noticed throughout social channels.
The earlier suggestions display screen: “The widget sometimes fails” vs “The widget fails all the time” — what is the distinction?
Our Problem Pages — the full-page safety blocks that seem after we detect suspicious exercise or when web site homeowners have heightened safety settings — had comparable points. Some states had been complicated. Others used an excessive amount of technical jargon. Many failed to offer actionable steering when customers wanted it most.
The state of inconsistency on the Problem pages. A number of states with no unified method
The audit was humbling. Nevertheless it gave us a transparent image of the place we wanted to focus.
To design higher experiences, we first wanted to grasp each attainable path a consumer might take. What was the blissful path? Was there even one? And what had been the sad paths that led to escalating frustration?
Mapping the entire consumer journey — from preliminary encounter via error situations, with sentiment monitoring
This was a real cross-functional effort. We labored carefully with engineers like Ana who knew the technical ins and outs of each edge case, and with Marina on the product aspect who understood not simply how the product labored, however how customers felt about it — the love and the hate we might see on-line.
Now we have a few of the smartest folks engaged on bot safety at Cloudflare. However intelligence and readability aren’t the identical factor. There is a delicate stability between technical complexity and consumer simplicity. Solely when these two dance collectively efficiently can we talk info in a method that truly is sensible to folks.
And this is the factor: the messaging has to work for everybody. An individual of any age. Any psychological or bodily functionality. Any cultural background. Any stage of technical sophistication. That is what designing at scale actually means — you may’t ignore edge instances, since, at such scale, they’re not edge instances.
Establishing a unified info structure
Some of the influential books in UX design is Steve Krug’s Do not Make Me Suppose. The core precept is straightforward: each second a consumer spends attempting to interpret, perceive, or decode your interface is a second of friction. And friction, particularly in moments of frustration, results in abandonment.
Our audit revealed that we had been asking customers to assume far an excessive amount of. Totally different items of knowledge occupied the identical house within the UI throughout totally different states. There was no constant visible hierarchy. Customers encountering an error state in Turnstile would discover info in a very totally different place than they’d on a Problem Web page.
We made a basic determination: one info structure to rule all of them.
Visible diagram displaying a unified info structure with a constant construction throughout Turnstile widget and Problem pages
Each Turnstile and Problem Pages would now observe the identical structural sample. The identical visible hierarchy. The identical placement for actions, for explanatory textual content, for hyperlinks to documentation.
Did this constrain our design choices? Completely. We needed to say no to quite a lot of inventive concepts that did not match the framework. However constraints aren’t the enemy of excellent design — they’re typically its finest pal. By limiting our choices, we might go deeper on the small print that truly mattered.
For customers, the profit is profound: they needn’t re-learn what each bit of the UI means. Error states look constant. Assist hyperlinks are all the time in the identical place. When you perceive one state, you perceive all of them. That is cognitive load diminished to a minimal — precisely the place it must be throughout a safety verification.
What consumer analysis taught us
How do you retain your self accountable when redesigning one thing that billions of individuals see? You take a look at. Lots.
We recruited 8 members throughout 8 totally different international locations, intentionally looking for range in age, digital savviness, and cultural background. We weren’t on the lookout for tech-savvy early adopters — we wished to grasp how the redesign would work for everybody.
Our method was rigorous: members noticed each the present expertise and proposed modifications, with out figuring out which was “old” or “new.” We counterbalanced positioning to remove bias. And we didn’t simply take a look at our new concepts, but additionally challenged our assumptions about what wanted altering within the first place.
Two totally different variations of a Turnstile being examined in an A/B take a look at
Some issues didn’t want fixing
One speculation: ought to we align with rivals? Most CAPTCHA suppliers present “I am human” throughout all states. We use distinct content material — “Verify you are human,” then “Verifying…,” then “Success!”
Had been we overcomplicating issues? We examined it head-to-head.
Our method gained decisively. For the interactivity state, “Verify you are human” scored 5 out of 8 factors versus simply 3 for “I am human.” For the verifying state, it was much more dramatic — 7.5 versus 0.5. Customers wished to know what was occurring, not simply be advised what they had been.
Consumer testing outcomes: customers strongly favored our method over the competitor-style design
This experiment did not ship as a function, however it was invaluable. It gave us confidence we weren’t simply being totally different for the sake of it. Some issues had been already proper.
However these wanted to alter
The analysis surfaced 4 areas the place we had been failing customers:
Assist, not paperwork. When customers encountered errors, we supplied “Send Feedback.” In testing, they had been baffled. “Who am I sending this to? The website? Cloudflare? My ISP?” Extra importantly, we found one thing basic: in the mean time of most frustration, folks do not wish to file a report — they wish to repair the issue. We changed “Send Feedback” with “Troubleshoot” — a single phrase that guarantees motion fairly than paperwork.
The problematic “Send Feedback” immediate: customers did not know who they had been sending suggestions to
Consideration, not alarm. We might used crimson backgrounds liberally for errors. The response in testing was visceral — members felt they’d failed, felt powerless. Even for easy points that may resolve with a retry, customers assumed the worst and gave up. Purple at full saturation wasn’t speaking “Here’s something to address.” It was speaking “You have failed, and there’s nothing you can do.” The repair: crimson just for icons, by no means for textual content or backgrounds.
The evolution: from the states with unclear error state description in crimson to a lot clearer and concise error communication in neutral-color textual content.
Scannable, not verbose. We might tried to be thorough, explaining errors in technical element. It backfired. Non-technical customers discovered it alienating. Technical customers did not want it. Everybody was attempting to learn it within the tiny actual property of a widget. The lesson: much less is extra, particularly in constrained areas throughout aggravating moments.
Accessible to everybody. Our audit revealed 10px fonts in some states. Gray textual content that technically met AA (not less than 4.5:1 for regular textual content and three:1 for giant textual content) compliance however was tough to learn in observe. “Technically compliant” is not adequate once you’re serving the whole Web.
We set a transparent purpose: to satisfy the WCAG 2.2 AAA commonplace— the best and most stringent stage of internet accessibility compliance, designed to make content material accessible to the broadest vary of customers, together with these with extreme disabilities. All through the redesign, when visible consistency conflicted with readability, readability gained. Each time.
This prolonged past imaginative and prescient. We designed for display screen reader customers, keyboard-only navigators, and folks with coloration imaginative and prescient variations — going past what automated compliance instruments can catch.
And accessibility is not nearly impairments — it is about language. What suits in English, overflows in German. What’s concise in Spanish is ambiguous in Japanese. Supporting over 40 languages pressured us to radically simplify. The identical “Unable to connect to website / Troubleshoot” sample now works throughout English, Bulgarian, Danish, German, Greek, Japanese, Indonesian, Russian, Slovak, Slovenian, Serbian, Filipino, and lots of extra.
The redesigned error state throughout 12 languages — constant structure regardless of various textual content lengths
So what did we really ship?
First, let’s discuss what we did not change. The blissful path — “Verify you are human” → “Verifying…” → “Success!” — examined exceptionally effectively. Customers understood what was occurring at every stage. The distinct content material for every state, which we might anxious may be overcomplicating issues, was really our aggressive benefit.
The blissful path: Confirm you’re human → Verifying → Success! These states examined effectively and remained largely unchanged
However for the states that wanted work, we made vital modifications guided by all the pieces we realized.
Simplified, scannable content material
We radically diminished the quantity of textual content in error states. As an alternative of verbose explanations like “Your device clock is set to a wrong time or this challenge page was accidentally cached by an intermediary and is no longer available,” we now present:
A transparent, easy state identify (e.g., “Incorrect device time”)
A distinguished “Troubleshoot” hyperlink
That is it. The detailed steering now lives in a devoted modal display screen that opens when customers want it — giving them room to truly learn and observe troubleshooting steps.
The troubleshooting modal: detailed steering when customers want it, with out cluttering the widget
The troubleshooting modal supplies context (“This error occurs when your device’s clock or calendar is inaccurate. To complete this website’s security verification process, your device must be set to the correct date and time in your time zone.”), numbered steps to attempt, hyperlinks to documentation, and — solely after the consumer has tried to resolve the problem — an choice to submit suggestions to Cloudflare. Assist first, suggestions second.
AAA accessibility compliance
Each state now meets WCAG 2.2 AAA requirements for distinction and readability. Font sizes have established minimums. Interactive parts are clearly focusable and correctly introduced by display screen readers.
Unified expertise throughout Turnstile and Problem pages
Whether or not customers encounter the compact Turnstile widget or a full Problem Web page, the knowledge structure is now constant. Identical hierarchy. Identical placement. Identical psychological mannequin.
Problem Pages now observe a clear construction: the web site identify and favicon on the prime, a transparent standing message (like “Verification successful” or “Your browser is out of date”), and actionable steering under. No extra partitions of orange or crimson textual content. No extra technical jargon with out context.
Re-designed Problem web page states with clear troubleshooting directions.
Validated throughout languages
Every bit of content material was examined in over 40 supported languages. Our course of concerned three layers of validation:
Preliminary design assessment by the design group
Skilled translation by our certified vendor
Remaining assessment by native-speaking Cloudflare workers
This wasn’t nearly translation accuracy — it was about making certain the visible design held up when content material size assorted dramatically between languages.
The result’s a safety verification expertise that is clearer, extra accessible, much less irritating, and — crucially — simply as safe. We did not compromise on safety to enhance the expertise. We proved that good design and robust safety aren’t in battle.
Re-designed Turnstile widgets on the left and a re-designed Problem web page on the proper
However designing the expertise was solely half the battle. Delivery it to billions of customers? That is the place Ana is available in.
Half 2: Delivery to billions
Some could say the toughest a part of being a Frontend Engineer is centering a div. In actuality, the actual problem typically lies a lot deeper, particularly when working near the platform primitives. Constructing a vital piece of Web infrastructure utilizing native APIs forces you to assume in another way about UI improvement, tradeoffs, and long-term maintainability.
In our case, we use Rust to deal with the UI for each the Turnstile widget and the Problem web page. This determination introduced clear advantages by way of security and consistency throughout platforms, however it additionally elevated frontend complexity. Many people are used to the ergonomics of contemporary frameworks like React, the place frequent UI interactions come nearly at no cost. Working with Rust meant reimplementing even easy interactions utilizing decrease stage constructs like doc.getElementById, createElement, and appendChild.
On prime of that, compile instances and strict checks naturally slowed down speedy UI iteration in comparison with JavaScript based mostly frameworks. Debugging was additionally extra concerned, because the tooling ecosystem continues to be evolving. These constraints pushed us to be extra deliberate, extra considerate, and finally extra disciplined in how we approached UI improvement.
Small visible modifications, huge international affect
What initially appeared like small visible tweaks akin to padding changes or alignment modifications rapidly revealed a a lot greater problem: internationalization.
As soon as translations had been obtainable, we had to make sure that content material remained readable and usable throughout 38 languages and 16 totally different UI states. Textual content size variability alone required cautious design choices. Some translations might be 30 to 300 p.c longer than English. A brief English string like “Stuck?” turns into “Tidak bisa melanjutkan?” in Indonesian or “Es geht nicht weiter?” in German, dramatically altering structure necessities.
Proper-to-left language help added one other layer of complexity. Supporting Arabic, Persian or Farsi, and Hebrew meant greater than flipping textual content course. Whole layouts needed to be mirrored, together with alignment, navigation patterns, directional icons, and animation flows. Many of those parts are implicitly designed with left-to-right assumptions, so we needed to revisit these choices and make them actually bidirectional.
Ordered lists additionally required particular care. Not each tradition makes use of the Western 1, 2, 3 numbering system, and hardcoding numeric sequences could make interfaces really feel overseas or incorrect. We leaned on locale-aware numbering and totally translatable listing codecs to make sure ordering felt pure and culturally acceptable in each language.
Constructing confidence via testing
As we began itemizing motion factors in suggestions stories, correctness grew to become much more vital. Each motion wanted to render correctly, set off the proper movement, and behave persistently throughout states, languages, and edge instances.
To get there, we invested closely in testing. Unit exams helped us validate logic in isolation, whereas end-to-end exams ensured that new states and languages labored as anticipated in actual situations. This testing basis gave us confidence to iterate safely, prevented regressions, and ensured that suggestions stories remained dependable and actionable for customers.
What started as a set of technical constraints became a chance to construct a extra strong, inclusive, and well-tested UI system. Working with fewer abstractions and nearer to the browser primitives pressured us to rethink assumptions, enhance our internationalization technique, and lift the general high quality bar.
The outcome isn’t just an answer that works, however one we belief. And that belief is what permits us to maintain enhancing, even when centering a div seems to be the straightforward half.
Designing for billions of individuals is a duty we take severely. At this scale, it’s important to leverage measurable information to inform us the actual affect of our design decisions. As we put together to roll out these modifications, we’re specializing in 5 key metrics that may inform us if we’ve actually succeeded in making the Web’s most-seen UI extra human.
1. Problem Completion Price
Our main north star is the Problem Remedy Price: the proportion of issued challenges which are efficiently accomplished. By transferring away from technical jargon like “intermediary caching” and towards easy, actionable labels like “Incorrect device time,” we count on a big uptick in CSR. The next CSR doesn’t suggest we’re being simpler on bots; it means we’re eradicating the hurdles that had been by chance tripping up respectable human customers.
Each second a consumer spends on a problem web page is a second they don’t seem to be getting the knowledge that they want. Our analysis confirmed that customers had been typically paralyzed by alternative when seeing a wall of crimson textual content. With our new scannable, neutral-color design, we’re monitoring Time to Full to make sure customers can determine and resolve points in seconds fairly than minutes.
3. Abandonment Price Modifications
Previously, our liberal use of “saturated red” triggered a visceral response: customers felt they’d failed and easily gave up. By reserving crimson just for icons and utilizing a unified structure, we goal to scale back Abandonment Charges. We wish customers to really feel empowered to click on Troubleshoot fairly than feeling powerless and clicking away.
One of many greater shifts from a product perspective is our new Troubleshooting Modal. By offering clear, numbered steps immediately throughout the widget, we’re constructing self-service help into the UI. We count on this to end in a measurable lower in help ticket quantity for each our prospects and our personal inner groups.
We all know that safety challenges are not often liked, however they should not be hated as a result of they’re complicated. We’re monitoring Social Sentiment throughout group boards, suggestions stories, and social channels to see if the dialog shifts from “this widget is broken” to “I had an issue, but I fixed it”.
As a Product Supervisor, my purpose is usually invisible safety — one of the best problem is the one the consumer by no means sees. However when a problem should be seen, it must be an assistant, not a bouncer. This redesign proves that AAA accessibility and high-security requirements aren’t in competitors; they’re two sides of the identical coin. By unifying the structure of Turnstile and Problem Pages, we’ve constructed a basis that enables us to iterate sooner and shield the Web extra humanely than ever earlier than.
This redesign is a basis, not a end line.
We’re persevering with to observe how customers work together with the brand new expertise, and we’re dedicated to iterating based mostly on what we study. The suggestions mechanisms we have constructed into the brand new design — those that truly assist customers troubleshoot, fairly than simply asking them to report issues — will give us richer insights than we have ever had earlier than.
We’re additionally watching how the safety panorama evolves. As bot assaults develop extra subtle, and as AI continues to blur the road between human and automatic conduct, the problem of verification will solely get tougher. Our job is to remain forward — to maintain enhancing safety with out making the human expertise worse.
For those who encounter the brand new Turnstile or Problem Pages and have suggestions, we wish to hear it. Attain out via our group boards or use the suggestions mechanisms constructed into the expertise itself.



