Here’s the uncomfortable truth: the three-click rule is a catchy slogan, not a usability law. You’ve probably heard it in a kickoff: “Everything should be reachable in three clicks or the user bails.” Neat. Also how you end up with bloated mega-menus, vague labels, and a homepage that reads like a yard sale. Users don’t count clicks; they count progress.

Let’s bust some myths.

• Many UX pros now treat “three clicks” as a dated, baseless buzzword. The data agrees.
• Users don’t bail at click three; they bail when they feel lost. Measure success, time, and path clarity, not raw clicks.
• Use deeper-but-clear structures, hub pages, breadcrumbs, and solid search-fast paths beat short ones.

What is the 3-click approach?

If you work in UI/UX, you’ve heard it: users should reach anything in ≤3 clicks or they’ll leave. The idea was popularized in 2001 by Jeffrey Zeldman, who framed it as a suggestion tied to users’ desire for quick gratification- not a law and not backed by evidence. It stuck because it’s catchy, not because it’s true.

Many UX designers now call the rule what it is-outdated shorthand that collapses a complex problem into a magic number. Let’s see how it fares in the real world.

What the data actually says?

Joshua Porter’s UIE study (44 users, 620 tasks, 8,000+ clicks) found no correlation between clicks and success. No higher quit rate at 3 vs. 12 clicks; many tasks ran well past 3; satisfaction roughly flat (unsatisfied ~46–61%) across click lengths. Myth busted.

NN/g’s conclusion is identical: the rule is arbitrary and not supported by published studies. In one e-commerce test, making products 4 clicks from the homepage (instead of 3) increased findability dramatically (reported as ~ 600% in summaries). The lesson: click quality beats click quantity.

Bonus: on mobile-news tasks, click/tap count wasn’t tied to failure-even long paths didn’t trigger mass abandonment-though more steps can feel harder, which is why clarity and speed still matter.

Translation: it’s not the odometer; it’s the map.

Why click count is the wrong KPI

Because it counts steps, not effort. Here’s what actually moves the needle-and why a raw step count misleads.

• Clicks ignore interaction cost. A click can be a 50 ms toggle or a 3-second load plus a tough choice. Users feel effort (reading, deciding, waiting, typing), not integers.
• Clicks ignore information scent. Clear, predictive labels (“Financial Aid,” “Program Requirements”) make an extra step trivial; vague labels make even one step risky.
• Clicks reward shallow, punish clarity. Chasing “≤3” flattens IA, bloats menus, and spikes cognitive load. Depth with clear labels beats shallow sprawl.
• Clicks assume everyone starts at Home. They don’t. People land deep from search or social; orientation and wayfinding matter more than distance from the homepage.
• Clicks don’t capture performance. Five snappy transitions feel better than three sluggish ones. Speed and responsiveness change perceived effort more than shaving a step.
• Clicks miss comprehension errors. Backtracks, pogo-sticking, and dead-ends are the real red flags-instrument and fix those.
• Clicks are a vanity metric; outcomes aren’t. Track task success, time to first meaningful content, error/backtrack rate, and quick satisfaction pulses.

Count friction, not clicks. Prioritize scent, orientation, and speed; let the structure be as deep as it needs-so long as every step is obvious and fast.

Pros & cons (if you insist on using it)

Use it if you must, but treat “three clicks” like linting, not law. It’s a quick way to spot detours and dead weight-then you ignore it the second it fights clarity, speed, or trust. With that frame, here are the “pros”… with caveats.

The “pros” people cite (with caveats)

Cleaner interfaces. Aiming for fewer steps can expose filler pages and redundant detours. (Great instinct-just don’t worship the number.)

Fast gratification. Users value speed. But “3” isn’t special; page performance and clear choices move the needle more than step count.

Shorter paths can be nice. When they reduce real friction (network calls, re-auths, form errors), not just clicks on paper.

The very real cons

Bad tradeoffs. Flattening IA to hit “3” bloats top-level menus and increases cognitive load. NN/g flags “3 clicks” vs “no more than 7 items” as two competing myths that force dumb decisions.

Irrelevant to modern entry paths. The homepage often isn’t the starting point anymore; users land deep via search or social. Your IA and wayfinding must work regardless of entry.

Not universal. Context matters. A therapy platform that shoves “Book Now” in 3 clicks without intake will feel sketchy; a short questionnaire builds trust (yes, more clicks, but better UX).

Patterns that work in 2025 (instead of counting)

Stop counting steps and start reducing thinking. The patterns below focus on clarity, orientation, and speed-the stuff users actually feel. Most are copy-and-ship fixes, not “burn it down and redesign” projects.

Write labels with teeth (strong information scent)

Users follow scent, not slogans. Use the NN/g “4-S” test for link text: Specific, Sincere, Substantial, Succinct. Kill filler like “Learn more”-it hides meaning and hurts accessibility.

Use hub/section landing pages for complex areas

Don’t cram everything into the top nav to “save a click.” Give each major section a hub with short blurbs and grouped links-clear paths beat shallow sprawl. You can add clarity without a full redesign; iterate your IA and labels first.

Add breadcrumbs for wayfinding

They anchor users in the hierarchy and support quick back-ups-especially on deep sites. Keep them concise, tappable, and visible; don’t let them wrap on mobile.

Prefer mega menus (on desktop) over hover mazes

Mega menus expose options in one roomy panel, improving scannability and reducing precision errors from cascading flyouts. Don’t let them cover the entire screen; keep them readable and lightweight.

Pair navigation with competent search (don’t outsource IA to a box)

Most people try nav first, and site search is frequently weaker than you think. Treat them as a synergy: clear categories for exploration, good search for specificity.

Use faceted navigation and filters for large catalogs

Facets let users narrow by multiple dimensions (brand, size, price) and understand the content space as they go. Make filter categories jargon-free and show applied filters so users don’t get lost.

Apply progressive disclosure in flows and dense UIs

Reveal complexity gradually (wizards, accordions, conditional fields). This lowers cognitive load without obsessing over step count.

Optimize for speed; five snappy steps beat three sluggish ones

Mind the classic response-time limits (≈100 ms feels instant; ~1 s keeps flow; ~10 s tests patience). Performance is UX, full stop.

Design for real entry points (not just “Home”)

Users land deep via search and social. Ensure every page shows where it sits, what’s nearby, and how to move forward — via clear headings, subnav, breadcrumbs, and in-context links. (See hubs + breadcrumbs above.)

Make copy do the heavy lifting

Good nav is good writing: front-load keywords, name things users actually say, and keep link text self-sufficient. The label should answer “What happens if I click this?” without reading the surrounding text.

Mobile specifics that actually help

Bigger targets, shorter labels, and tray/overlay patterns for filters (with a clear “Apply” and “Reset”). Keep the path visible; don’t bury wayfinding in a hamburger forever.

Instrument outcomes, not clicks

Track task success, time to first meaningful content, backtracks, dead-ends, and step-wise drop-offs. Those tell you where the path breaks; “≤3 clicks” doesn’t. (Then iterate labels, hubs, and filters accordingly.)

Wrapping up

The three-click rule is catchy, not credible. It’s folklore that collapses a complex problem into a lucky number. Real users don’t bail at “three”; they bail when labels are vague, pages are slow, or wayfinding is missing. Count friction, not clicks.

Here’s the replacement rulebook:

• Scent over shortcuts. Write labels that predict the next screen, not brand poetry.
• Depth with dignity. Use hubs, subnav, and breadcrumbs; shallow sprawl is harder to scan than one more obvious step.
• Speed is UX. Five quick transitions beat three sluggish ones.
• Search complements IA. Good categories first; smart search as backup.
• Design for where people land. Every page should show “where am I?” and “what’s next?”
• Build for everyone. Targets, focus states, ARIA, no hover-only traps.

Measure what matters (set a baseline, then set targets):

• Top-task success ≥ 90%
• Median time to first meaningful content ≤ 60s
• Backtrack rate < 15% on key paths
• Dead-end rate < 5% and trending down
• Step-wise drop-offs decreasing after each iteration

Keep the number of steps secondary to clarity, orientation, and speed. If a step earns its keep — by reducing confusion, improving accuracy, or building trust — it stays. If it doesn’t, cut it. Simple as that.

Originally published at https://icons8.com on September 22, 2025.

Before I started working at Icons8, I was a system administrator. I was repairing PC’s, doing backups, and restoring deleted emails that had suddenly become very important. But sometimes I get really weird requests. Like fixing a light bulb in a women’s bathroom (I hope that wasn’t a euphemism) or checking on a microwave. One day, I was even asked if I was good with blenders.

That’s the way people tend to see tech people. Yes, we know stuff: our stuff (in most cases). Yet not all tech-related stuff is our stuff. Ironically, even we often think we know stuff we’ve never had any experience with. Because we too see ourselves as “tech” people.

Thus, a myth was formed: tech people are tech-savvy. Which easily extrapolates into tech people being savvy in everything.

Well, if that’s true, then tech people can easily repair, say… a bicycle.

Hold on. Bicycles? Yes, bicycles. And before you get scared that I’m jumping like Spider-Man from IT to bicycles, let me explain myself.

If you want to repair a bicycle, you have to know how it works. And if you know how it works, you can easily draw a bicycle. Most people are confident they know how a bicycle works. And yet, if you ask them to draw a bicycle, their drawings may look like this:

Scientists call this an illusion of knowledge. Our brain more readily convinces us that we know something than it lets us admit that we don’t. National Geographic made a whole episode about this phenomenon, where 90% of participants drew bicycles unrealistically.

In his famous project Velocipedia, Gianluca Gimini pushed it even further. He asked different people to draw a bicycle and then created 3D models, based on their sketches:

So, 90% of people don’t know how a bicycle operates. And yet…

In both aforementioned examples (National Geographic and Velocipedia project), participants were people with general knowledge (many were students). No specific criteria. This is where I got lucky.

Recently, we’ve been training a neural net, whose sole purpose was to recognize icons, sketched by people. We sent an email to our user database asking our customers to draw all kinds of things: cars, houses, trash boxes… and bicycles.

And our audience consists of:

Developers. Half of the people we asked to draw a bicycle were developers. Now the question stands: will 90% of bicycles still be unrealistic, or will there be any improvement given the fact that half of the audience are developers?

Overview

I asked my friend, an engineer and avid biker (told you I’m lucky) to help me analyze 200 drawings of bicycles that we then put into 4 different categories:

Not rideable: it’s usually a very primitive drawing of two wheels and a frame, attached in a way that prevents wheels from rolling at all.
Ride by rolling: these bicycles can roll, but can’t turn. Or, sometimes, be sat on. So these bikes are for very straightforward people who have no time to sit.
Rideable (more or less): these have small issues like no pedals/chain or redundant structures of the frame.
Totally rideable: people really knew what they were drawing there.

Overall, 76% drawings are unrealistic, not 90%.

Developers draw bicycles that are actually rideable slightly more often.

Before we jump to conclusions, however, there are a few important factors that could affect this number.

Constructed Drawing vs. Prepared

Before asking my engineer friend for help with analyzing the drawings, I asked him to draw one. Unprepared and receiving no hints. Here’s how it went:

Take a look at how he doesn’t use some prepared mental shortcuts for elements, but constructs the bicycle. It is a constructed drawing.

The lines may be irregular and proportions may be messed up, but that’s not important. You’ll see that these bicycles were also constructed:

On the other side, there are quite a few designers among our audience (~30%). Their drawings are very accurate.

These are prepared drawings. Fortunately, only a very small portion of all drawings seemed like prepared ones. So I could conclude two things:

• It does not affect overall numbers that much
• 30% of our audience is designers, yet only a small portion of drawings were prepared. I suppose not many designers are able to draw a bicycle from memory. By the way, judging by this experiment, even if you give them a reference, it may not be enough.

Other factors

There are few more factors at play here:

Uncontrolled experiment

Drawings were not restricted by time, and there were no observers — people could Google for a reference. However, judging by the number of unusual bicycles, not many of them did that. Even when given a chance, people still prefer to invent:

“A single designer could not invent this many new bike designs in 100 lifetimes. And this is why I look at this collection in such awe.” — Gianluca Gimini, Velocipedia

Country

In some countries, bikes are more popular than in others. However, only one of our top 5 traffic countries (Japan) is on the list. I’d love to see how good people are at drawing bicycles in the Netherlands, where there are more bicycles per soul than horses in the Mongol army.

Conclusion

With all the factors mentioned, plus statistical faults etc. the number of people, who draw bicycles unrealistically is 5–10% lower among developers than in general. But that is a big stretch, for there are so many factors in play: general popularity growth of bikes during the last years, gender specifics (92% of developers are men) etc. Yes, I’d like to think techs are savvy; however, the numbers are not that dramatic to be really sure that’s because of their job.

So I’ll end this article just as I started it with my own example.

I’ve repaired hundreds of PC’s, configured many network devices. I’ve never repaired a single bike in my life.

So, after watching hundreds of drawings of bicycles, I’ve decided to draw one myself. Not to copy it, but to construct it in my head, from scratch. Not using references and mental shortcuts:

Though I’ve had half a dozen of them, I’ve never had a chance to repair one, and never had the motivation. I messed up the wheeling frame. My bicycle can’t turn.

I’m not saying tech people can’t fix things. I do believe tech people are savvy, and repairing a bike is easier than migrating from one framework to another, while covering every bit of code with tests, in a train, using a phone… You got it. Everything is possible if you have the motivation.

I’m saying that it shouldn’t be expected of them to fix anything. So, if you really want me to repair your bike, try giving me some motivation. A cupcake is a good start.

About the author
Andrew started at Icons8 as a usability specialist, conducting interviews and usability surveys. He desperately wanted to share his findings with our professional community and started writing insightful and funny (sometimes both) stories for our blog.

We’ve completely revamped our AI illustration generator. These aren’t just random updates-they’re built to solve those specific creative roadblocks that make you question your life choices at 2 AM.

First Liquid Glass icon generator

Apple just dropped its Liquid Glass UI, and suddenly every designer on Earth is scrambling to recreate that translucent aesthetic.

While everyone else is still figuring out how to fake it in Figma, we built the first-ever Liquid Glass icon generator.

Now you can create those dreamy, blurred-background mockups in 5 minutes instead of spending your entire weekend manually tweaking transparency layers on your icons.

53 curated illustration and icon styles

Our AI learned from OUR artwork, not random internet scraping. While other platforms are out here training on whatever they found lying around Reddit, we’re using actual professional illustrations created by real humans with taste.

• Your graphics actually work together harmoniously (revolutionary concept, we know)
• Copyright issues? Not your problem
• Premium look without the premium budget stress
• High-quality illustrations for any project
• 26 SVG icon styles, 3 new SVG illustration styles

Infinite scalability without pixelated nightmares. Your designs stay crisp from favicon to billboard.

Community inspiration section

Added a user generation feed because sometimes you need to see what others are creating to spark your own ideas. It’s like Pinterest but actually relevant to your current project.

AI prompt assistant (for when your brain goes blank)

You know that moment when you’re staring at the prompt box thinking “make it… good?” Yeah, we’ve all been there. Our AI assistant helps you write actually useful prompts instead of watching you generate 47 versions of “business person doing business things” because you couldn’t think of anything more specific.

New AI model that actually gets anatomy right

Remember the dark days of AI-generated people with alien fingers and proportions that defied physics? Yeah, we killed that energy. Our new AI model is not only faster than before, but it’s finally learned what human hands are supposed to look like.

What’s actually improved:

• Speed boost. Generate images 3x faster without sacrificing quality
• Proper human anatomy. No more nightmare fuel with 7-finger hands and impossible arm positions
• Better proportions. People actually look like people, not funhouse mirror reflections
• Enhanced detail quality. Crisp, professional-grade output that doesn’t look AI-generated
• Consistent facial features. Characters maintain their identity across multiple generations

This isn’t just a minor update-it’s the difference between “this could work” and “this is exactly what I needed.” The model understands context better, follows prompts more accurately, and produces results that look professionally illustrated rather than algorithmically assembled.

Solving real problems

Whether you’re a freelancer juggling multiple clients or an in-house designer drowning in requests, this AI illustration generator handles the heavy lifting:

• Client says “something like this but different” → Generate variations instantly
• Need icons that match your illustration style → Same AI, consistent output
• Deadline is tomorrow, starting today → Fast generation, immediate results
• Budget screams stock, taste demands custom → Get premium vibes without premium prices

Start generating

Originally published at https://icons8.com on June 30, 2025.

Many entrepreneurs strive for revenue and focus on core features, often neglecting tiny things in UX that might be essential for an app to be successful in the long run.

These small details aren’t just pretty decorations — they’re what makes users remember and share your product. They’re what turns “just another app” into something people actually enjoy using.

In this digest, we’ll go through such things, from small to huge apps, that spent time making their interfaces pleasant to use.

Power of tooltips

Not only did sketchplanations make the underlying a little bit curvy, but they also showed a pop-up with a preview of what to expect from the next page.

This might be quite engaging if done properly, since this trick can increase the probability that users will click on the link. It’s like a sneak peek of what you will see if you click.

Another amazing example is done by Christopher Kindl:

In this case, it serves as an extra help when you need to choose the right option.

So the difference between these two examples is piquing curiosity versus providing help.

The downside of this idea is that it will not work on touch screens, since there is no hover event. But on the other hand, nothing will be broken too.

Showing care through details

Google Maps highlight if the place is owned by a woman by showing a nice heart icon.

Apart from that, they also show such things as whether the place is suitable for people with disabilities or not.

Actually, we often don’t pay attention to small improvements that add up very quickly. Of interest, you may check some sites via https://web.archive.org/ to see how they looked 5–10 years ago and how they look now.

Error prevention tips

It’s not a secret that people make mistakes, and that’s okay. However, it’s always better to prevent making mistakes, and there are a lot of ways to do so.

For example, in ConvertKit they show you an alert if you forgot to change the email subject. Sounds reasonable, since you don’t want to send a newsletter with the default subject, right?

Also, you may indirectly help users, which means that you may have a feature that helps working with your app, and this lowers the number of mistakes.

For example, when you paste text in markdown format, it’ll ask whether it should be formatted like markdown or not. Which is helpful: you might write your content in a markdown editor and then copy it to different apps, and if they recognize the markdown and automatically format text accordingly, fewer actions are needed from users. Fewer actions might lower the number of potential mistakes.

Another example. When you write a newsletter issue, you can enter into the focus mode, which becomes a must-have for text editors. Then, it’s possible to preview the text on different resolutions and, most importantly, preview the text as a subscriber. It comes in handy to check how it looks if you have some conditional logic.

Advanced validation tooltips

Most of the time, the validation does not end with a couple of required words. A good example where validation rules might be complicated is Typefully.

For those who are not familiar with it, it is a social scheduling tool. There is a typical problem for such apps: each social network has its own rules, such as message length restrictions. But users typically want to make a post for all social networks at once.

For example, if you make a long thread, on X and Bluesky the limit for posts is different. If it turns out that your thread is correct for X but some of your posts are too long for Bluesky, Typefully handles this case in a user-friendly way, properly showing a validation message.

Daring to try new patterns

There is a Jakob’s law in UX:

Users spend most of their time on other sites. This means that users prefer your site to work the same way as all the other sites they already know.

And sometimes it’s hard to deviate from common practices, since they are well-established and proven over time. But nothing stays in the same state forever, and UX is not an exception.

Sometimes you may try something new. It might not be new, though, but rarely used. An example of trying an unconventional pattern again comes from Typefully.

The idea is that for non-critical destructive actions, instead of showing the confirmation popup, they suggest you click once again to confirm the action.

While there might be downsides to such an approach, it certainly deserves consideration, since it may turn out that pros outweigh cons in your particular case.

Unexpected effects that matter

Mercury has the hidden mask effect, which only reveals itself when you hover over certain areas.

Combining it with an overall clean design and a plethora of nice UX solutions, we can expect that users will share it. And they do; there are tweets where people posted photos of the Mercury physical credit card, since it also looks good.

That’s the case when quality turns into word of mouth and people share your product. There is even a law related to this fact called the “ Aesthetic-Usability Effect “.

Asana allows you to set up two whimsical settings.

The first one is “Extra delight”, which makes nice gradient animations even nicer (they become colorful).

So when you complete a task, the gradient animation becomes colorful.

The second one is showing occasional celebrations, which in fact are flying unicorns and other creatures when you complete a task.

Sounds crazy? Well, apparently people loved that, because the feature has been there for quite a while and became even more flexible.

It all started with flying unicorns; now there are more creatures. Here is how it looks.

Inspiration for your own micro-interactions

While we’ve looked at small UX gems in products, sometimes the best inspiration comes from creative individuals. Josh Comeau ‘s blog showcases numerous micro-interactions that could be adapted for product interfaces. And you can be sure that it’s something doable, not just your fantasy.

It would take a few hours to show all the different non-standard features, but let’s look at a few micro-interactions that could inspire your next product gem.

Rainbow generator

Not only is it an animated, nicely designed component. It is also configurable, and notice that the settings show how many users are on the site, and the changes you made will be reflected to every visitor.

Let’s see it in action.

Now, the thing is, every element is custom (e.g., sliders or buttons). And of course, it correctly changes when turning on the dark mode. Note that even Josh’s figure changes when the mode is switched!

Micro animations

The whole site is full of micro animations. It’s quite uncommon to see them in most apps, starting from small ones and ending with operating systems.

But that definitely makes an impression.

All skills together — the “about me” page

Now, imagine you’re looking for a skilled React developer. We’ll leave alone the biography of Josh, but look at his “about me” page.

1. Usage of maps, detecting how far you are from Josh
2. An interactive tutorial on drawing the Pride flag
3. Animated figure of his height
4. Drawn and interactive generative art machine
5. Cat, reacting to clicks
6. List of conferences
7. Number of students
8. An article about how he managed to code without a keyboard and mouse for 6 months
9. Name pronunciation

And most of the interactive elements have sound effects.

This is how a CV should look.

Small things, big impact

All these diverse examples are meant to convey one simple idea. People are not fools; they can see when a product has high quality and when it doesn’t. So adding nice transitions in a to-do app or having the dark mode for a reading application is not a bad idea. In the long term it will pay off.

These small UX gems often become the signature elements that users remember and share. Whether it’s Asana’s flying unicorns that bring unexpected joy, or Mercury’s subtle mask effects that feel magical, — these details transform functional products into memorable ones.

Next time you’re tempted to skip that “extra” animation or helpful tooltip because it’s not “essential,” remember that the line between a product people use and a product people love is often drawn with these seemingly small details. The best part? You don’t need to implement everything at once. Start with one delightful element and watch how users respond.

For more UX tips and tricks, check out these articles:

About the author
Victor Ponamariov, a web developer turned UI/UX enthusiast who shares design insights through his blog. He published a book with 100 practical UI/UX tips where he shares his experience.

Originally published at https://icons8.com on June 10, 2025.

Have you ever perfectly aligned elements in your design tool only to feel something still looks ‘off’? You’re not imagining things. The mathematical precision of design software often conflicts with how our brains actually perceive visual information.

The difference between technically correct and visually harmonious design often comes down to a single pixel - a subtle shift that your users will feel without knowing why. Master these principles, and you’ll create interfaces that not only look balanced but feel intuitively right.

Visual weight

Visual weight is the perceived “heaviness” or “importance” of an element in a composition. It depends on such factors as:

Shape

Dense, compact shapes (such as squares and circles) appear heavier than irregular, loose shapes. Goodluca used a nice badge on top of the phone that certainly draws attention.

Some examples of using shape and density are:

1. Marking certain items with a badge or a star (for example, if you want to highlight the best goods in your catalog)
2. We are placing a small, yet saturated dot next to the notification icon to inform users that there are new unread notifications.

Larger elements tend to appear “heavier.” On Gumroad, the avatar is tiny, but the dashboard button is huge (plus has a solid background), indicating its heavy weight.

This is the most trivial way to make an element more prominent-just increase its size.

One of the most common mistakes is not utilizing whitespace for text and CTA buttons. As they say:

“Use space, it’s free”

Saturated or darker colors have more weight than light, de-saturated colors.

For example, on ConvertKit’s landing page it’s obvious which buttons stand out more (= are heavier).

This method works fine, but keep in mind that every such element takes some attention, and don’t overuse it. Otherwise, all elements will be “heavy”, and none of them will stand out.

Actually, when we talk about visual weight, the word “weight” is a bit misleading, since we talk about what makes an element stand out amongst others. Which can be color but this is only one of the options.

Position

Elements near the design’s edges or corners feel heavier than those in the center. A common example is the scroll-to-top button placed at the bottom-right corner.

However, be careful with these points. Even though they are indeed considered to be so-called “anchor points”, due to their distance, users might not notice the element if it doesn’t have additional contrast.

For example, on the Threads web app, they have the add new post button in the bottom-right corner, but since it has a white background color, it doesn’t stand out enough and might not be noticed by users.

Another example of using position is moving elements at the top. In the example below, the first four cards will be more noticeable than others.

You can make it even more prominent by adding some white space between the first row and others.

Contrast

High-contrast elements draw more attention and carry more weight.

For icons accompanied by labels, the combination of saturation, contrast, and size is crucial, as we will see further in the article.

Contrast and color are a little bit different tricks. It’s easier to demonstrate usage of contrast in the following example.

This idea is often used on landing pages that rely on typography. When the secondary text is grayed out, the keywords have bigger saturation, making them “heavy” and easy to scan.

Direction

Elements that imply movement or direction (such as arrows) can affect weight perception.

Some examples:

1. The scroll-down animated icon implies that there is something below. So it feels a little bit like it pushes you down when it’s animated
2. Scroll up does the contrary
3. Right arrows (often called chevrons) in buttons convey the direction and work as a signifier. Additionally, they provide a slight visual weight.

The squint test

There is a related thing called a squint test that may sometimes come in handy to check your whole interface. To do this, take a screenshot of a page and blur it.

You will see that some elements stand out more than others. For example, let’s take X’s home page and blur it.

We can obviously see white buttons, and they stand out. In this case, they are “heavier” due to their size. Also, there is a menu of post actions that slightly stand out due to their color.

Now, let’s do the same with Bluesky.

Note the difference. They use not only size but color as well, which is more common.

This technique may help you to quickly identify what stands out more on your page.

Of course, in both cases, I didn’t take into account the feed because it takes up most of the space and apparently draws the most attention. However, the focus might shift if the feed solely consists of text messages, as it occasionally does.

In different types of applications, the test might be more useful than in others, so use it wisely when needed.

Icons and labels

There is a crucial thing that every designer should be aware of, since it is directly related to visual weight.

Consider the following example.

The color of text and icons is pure black. If we blur the menu, we’ll see black spots that carry more weight than the labels. This fact creates an imbalance.

This happens because filled (not outlined) icons, in particular, have a higher pixel density than a stretched, wide label.

To compensate for the balance, the icons should be lighter. In the example below, we made the labels lighter because the pure black color can be a bit harsh on our eyes.

When you hover over the items, you can increase the saturation of both the text and icons. But for icons, you need to increase it less than for text, since they are already dense and “heavy”.

Let’s take a look at a real application and see what it would look like if the balance between labels and icons' colors were wrong. Note that in the example before, I made it too obvious for demonstration purposes. But this is how the subtle difference would look in a real app.

While visual weight affects how elements compete for attention, optical balance addresses how our eyes perceive alignment and proportions, often in ways that contradict what our design tools tell us.

Optical balance

There are numerous examples of visual illusions that happen when our eyes perceive information inaccurately. The most common one is comparing a square with a circle.

In this example, the circle seems smaller, but it is exactly the same height.

This is important because we always work with rectangles. Even if you have a sophisticated curve in Figma, when you try to align it, it will have some width and height. Occasionally, a technically aligned curve may appear visually odd.

Physics and what it has to do with alignment

In physics, there is such a concept called the center of mass:

The center of mass of a distribution of mass in space (sometimes referred to as the barycenter or balance point) is the unique point at any given time where the weighted relative position of the distributed mass sums to zero.

Sounds scary, but here is a helpful illustration of how it works with icons. Consider this triangle. Imagine it has some physical weight. On the left side, there is more weight than on the right side.

But in Figma or any other tool, the triangle boundaries will be a square:

Now, imagine we need to make a play button. So we make a circle and a triangle and align them vertically and horizontally:

The alignment is correct since the boundaries of the icon are aligned. However, there is an optical shift of the triangle to the left side.

That’s when physics and geometry come into play. In our case we need to find the centroid of the triangle, which is the geometric center.

In the case of a triangle, the process is pretty straightforward:

Now let’s compare the geometrical center (which can also be the center of mass if the triangle is homogeneous, which means that the mass is distributed evenly) and the center of the circle.

As you can see, they are not the same. That’s why the triangle looks slightly off. In order to fix it, we need to center these two dots.

And now we have a properly centered triangle. Compare the first one (centered technically) and the second one (centered optically).

When one pixel makes a difference

When you use icons, they always have rectangular boundaries. However, sometimes optical alignment is not correct, e.g., when the icon is not symmetrical.

In this example, the left icon is a little bit misaligned; literally one pixel makes a difference.

Trust your eyes

Simply aligning items in Figma does not guarantee that they are indeed aligned. And even if they are aligned, our eyes may perceive visual information incorrectly. For example, one figure may seem to be larger than another one, while in fact they are equal.

The gap between technical precision and visual harmony is where great design happens. While your tools might tell you everything is perfectly aligned, your eyes may reveal a different truth.

Remember these principles when crafting your next interface:

• Visual weight determines what commands attention, through size, color, contrast, and position
• Optical balance requires adjusting beyond mathematical center points
• A single pixel shift can transform a design from awkward to harmonious
• The squint test quickly reveals your true visual hierarchy

Even the most sophisticated design tools can’t replace your trained perception. Trust your eyes and design for how humans see, not how computers measure. Your users will never articulate why your interface feels so right, but they’ll certainly notice the difference.

About the author
Victor Ponamariov, a web developer fell in love with user interfaces. He published a book with 100 practical UI/UX tips where he shares his experience.

Originally published at https://icons8.com on May 26, 2025.

AI image generators exploded. Turns out, most people don’t just want to see themselves with new haircuts.

Over the past five years, AI image generation has surged — bigger models, sharper images, jaw-dropping realism. But while headlines talk about art, business headshots, and digital twins, there’s a quieter truth sitting in the backend statistics: a huge chunk of that growth is driven by something much simpler. Desire.

Adult content — or at least sexually charged imagery — has been one of the biggest forces behind the adoption and evolution of generative AI. It’s not a new pattern, either. VHS didn’t beat Betamax because of picture quality. Streaming services weren’t propelled forward by indie cinema alone. Throughout tech history, wherever there’s a screen, there’s been a race to fill it with something a little more provocative.

And generative AI? No exception.

Quick reality check: what people actually generate

Look under the hood of almost any popular AI generator, and you’ll find a pattern: requests lean heavily into idealized, sexualized, and fantasy-driven images. We’ve seen it firsthand (more on that later). In short, while AI content officially sells itself as “professional,” the user base often quietly asks for “beautiful,” “wealthy-looking,” “sexy.” Old needs, new tools.

And it’s not just isolated users quietly testing boundaries — whole communities have sprung up around it:

• Reddit communities dedicated to NSFW AI generations regularly outpace their safe-for-work counterparts in engagement.
• Sites like Unstable Diffusion (ironically named) launched explicitly to enable uncensored image generation after bigger platforms cracked down.
• Popular open-source models like Stable Diffusion have entire forks built solely for adult content generation.

Even platforms that officially ban explicit content have their prompt systems flooded with soft suggestions — “elegant,” “seductive,” “luxurious,” “model-like,” “dreamy.” Call it what you want, but the instinct is there.

The ethical and legal pushback

Of course, a new wave of tech brings new problems. With AI’s ability to produce hyper-realistic faces and bodies, ethical concerns quickly followed:

• Child protection and the absolute necessity of strict content moderation.
• Deepfake misuse and the creation of non-consensual imagery.
• Unrealistic beauty standards pushed even further beyond human reach.
• Potential disruptions in industries like sex work and modeling.

Governments and major platforms are already trying to regulate. App Store and Google Play policies forced mobile apps into heavy content filtering. OpenAI, Midjourney, and Stability AI are all discussing — or have implemented — tighter NSFW generation controls.

And yet, users keep finding workarounds. Demand doesn’t vanish just because moderation policies exist.

How the market is adapting

With mobile app restrictions tightening, much of the more “adventurous” AI generation has quietly moved to web platforms. Web services aren’t tied to the same App Store/Google Play rules and can offer more freedom, within reasonable ethical and legal limits.

At the same time, even the big players are inching toward reality. OpenAI recently hinted that allowing more adult-themed generations might be on the horizon, because pretending the demand isn’t there just doesn’t work forever.

And while regulation is absolutely necessary for protecting people, pretending that humans won’t use AI to chase beauty, fantasy, and connection would be…well, naive.

GenYOU story

When we first built GenYOU as a mobile app, we played by the rules: strict NSFW filters, business-friendly image templates, and a lot of caution. And it made sense — app stores have zero tolerance for mistakes in that area.

But the data spoke louder — 30 to 50% of queries triggered the NSFW filters.
We saw that users weren’t just looking for polished profile pictures. They wanted something more personal. More confident. More glamorous. Sometimes a little bit sexier.

Here’s how user interests break down across common query themes:

It’s not about creating explicit content for the sake of it. It’s about people wanting to see themselves as idealized, enhanced, dreamlike versions of themselves. That’s as old as mirrors themselves.

That’s why we built the web version of GenYOU — to give users a bit more breathing room. Without the heavy-handed restrictions of app stores, we can allow for tasteful, adult-leaning generations, while still keeping everything legal, ethical, and respectful.

Same easy process — just a little more freedom to be as gorgeous, bold, or extravagant as you want.

Some things don’t change

Technology moves fast. Human nature stays put.

Every new breakthrough, sooner or later, gets pulled into the same timeless themes: attraction, beauty, fantasy, curiosity. AI generation is just another tool in that long tradition — sharper, faster, and maybe a little more daring.

Some will celebrate it, some will worry about it. Both reactions are fair. But either way, it’s here, and people are already figuring out how to push its limits.

If you’re wondering what that looks like in practice, GenYOU web is one place to start.

Adult content built the internet. Now it’s building generative AI, too was originally published in Generated Photos on Medium, where people are continuing the conversation by highlighting and responding to this story.

In the context of UX, icons can be considered as signifiers — additional indicators that amplify the function of a button.

However, the absence of labels can lead to ambiguity. What’s interesting is that a lot of top companies misuse icons, and you don’t need to conduct usability testing to know that. Such UX fails cost users time, create frustration, and may damage product experiences.

Sometimes icons are the only option when the space is limited, for example, on mobile screens. But what makes us wonder the most is that the choice is absolutely nonconventional, let alone the fact that the icons are not supplied with labels.

Let’s try to play a game. We show you a set of icons; you try to think what they are used for.

X/Twitter

Here is an icon set. All icons are in the same navigation sidebar except the last one (which deserves attention as well).

Imagine you develop an app and need to make a menu with proper icons. So you came up with this.

Well, let’s try to guess what they mean.

Here is the answer.

What is wrong with them? The answer is: recognition and relevance.

Some of the icons are hard to recognize, especially in the context of the sidebar menu.

Mentioning Jakobs Law won’t hurt:

Users spend most of their time on other sites. This means that users prefer your site to work the same way as all the other sites they already know.

For example, the “Premium” item is in the middle of the sidebar, and it shows a modal with subscription plans.

However, very often you’ll find subscription options either in the profile settings, in the header, or in the footer of the side menu, but not in the middle of it, especially with such an icon.

A perfect illustration of redundant functions was made by Andy Allen.

In order to fix the navigation, it’d be better to start off with proper information architecture.

There is also an article comparing Bluesky and X, where a lot more details were explained.

Also, recently another unbelievable case appeared on X. Guess what the icon means? Let’s assume you are already aware that the icon represents “Grok,” an AI assistant.

The answer is…

Most of the time, AI assists in the creation, rephrasing, and generation of posts. Explaining a post that is difficult to understand is not necessarily a bad thing. But two things don’t add up:

The Grok icon. It’s impossible to guess that it means “explain this post”. Absolutely.

Explaining posts is probably not a frequent action. Does it deserve a place here? Just to remind you, there is a “more” action under each post with at least 8 items, such as “not interested in the post”, “mute

The Grok icon. It’s impossible to guess that it means “explain this post”. Absolutely.

Instagram

Here is an Instagram profile. There are three buttons:

1. Follow
2. Message
3. …?

It seems like the icon means to add the person to something, maybe to some kind of list. But in reality, it opens an additional panel below (violating the proximity law) suggesting similar accounts to me. Impossible to guess.

Actually, if there wasn’t a hint on hover, I bet users would be afraid of clicking on it, because maybe it will send the user some kind of notification, and you don’t want to do this (e.g., friend request or something).

Threads

There is a very subtle and interesting case of using icons. Here’s an example of an icon in use on Threads.

If you don’t use Threads, try to guess what it means. Most users that I asked thought that the icon meant alignment because it resembles aligning text or elements to the left.

The paradox lies in the icon’s ability to serve multiple purposes, as its appearance can convey multiple meanings. However, it has become common practice to use a slightly different version of the icon for the purpose described below.

Here is the context. Maybe now it’s easier to guess that it means “ Add a poll “.

Let’s see what icon is usually used in such cases.

Note that in most cases, the bars are vertical. Another pattern is horizontal bars with the same length and circles indicating options.

Twitter

Telegram

LinkedIn

Google Pictures

Note that in most cases the bars are vertical. Another pattern is horizontal bars with the same length and circles indicating options.

Big guys can screw up, but you don’t have to

Even tech giants sometimes choose icons that are really confusing in common situations. But you don’t have to-you can learn from their experience. Here’s how you can help users avoid solving puzzles while using your designs:

• Search for common, well-known metaphors. Check out other app interfaces and icon libraries. Google actions and features to see what images appear, and decide which icon metaphor will work better for your users.
• Use icons and labels together when possible. This increases the clickable area and decreases cognitive load by removing ambiguity.
• Select appropriate icons and add hover hints if space is limited. But be smart about it-if your top navigation has a widely recognized profile icon, you don’t need a tooltip. Instead, include the user’s name, or on mobile screens, just leave the recognizable icon as is.

Strike a balance between an interface cluttered with unnecessary hints and one filled with mysterious icons. Remember: if users need to guess what your icon means, you’ve probably chosen the wrong one.

If you want to dive into icons, here’s your full guide on everything you need to know about icons and learn how to design icons for killer UI.

About the author
Victor Ponamariov, a web developer, fell in love with user interfaces. He published a book with 100 practical UI/UX tips where he shares his experience.

Originally published at https://icons8.com on May 14, 2025.

AI website builders love to brag: “Generate a website in seconds!” “Launch your business with no code!” “Let AI do the work!” But I was skeptical. Do these tools actually deliver, or are they just slapping shiny templates on mediocrity?

I tested 11 different AI website builders. Some I found through Google while searching for “AI website builders,” others were recommended by fellow designers, and one showed up on Product Hunt. I was mainly looking for platforms that could quickly generate a simple, client-ready site without much manual setup or coding — ideally something a small local business could start using right away, no technical fuss.

For each platform, I used the same brief/prompt:

“Build a website for a local cleaning service called SparklePro with a homepage, service list, customer reviews, and a contact form. Use a bright, trustworthy design.”

The results? A mixed bag of impressive innovation, mediocre templates, and some genuine surprises. Here’s what I discovered.

TL;DR

Most AI builders are still just wrapping old-school templates in fancier packaging. But a few — Framer and Lovable.dev — feel like real evolution.

Framer: sleek, modern, almost too cool

Framer required two attempts-the first version was just a hero section with two buttons. But the second attempt delivered something genuinely impressive: a modern layout with responsive design, customizable fonts and colors, and solid section structure.

What stands out is how it feels less like a “website generator” and more like a proper web design tool that leverages AI. You can publish to a Framer subdomain instantly, but you’ll want to spend time fine-tuning the details.

Verdict: One of the best results. Looks like something you’d actually pay a designer for.

Pricing: Personal starts at $5/mo for 2 landing pages, Business from $75/mo

Lovable.dev: surprisingly decent design

Among the more visually impressive results, Lovable.dev generated a clean, scrollable site with working navigation and clear structure. The design aesthetic was noticeably better than most competitors.

The only significant drawback? The critical “Book Now” button required additional backend setup, which undermined the “automatic” promise and demanded technical knowledge.

Verdict: Visually solid design, great for those comfortable with some technical configuration.

Pricing: Free plan limited to 30 messages per month, Starter plan at $20/mo

Durable: fast, basic, surprisingly smart

Durable delivers on its promise of a complete site in seconds. While the design is fairly generic, each content block can be edited manually or with AI assistance. A standout feature: it automatically writes alt text for images-genuinely useful for SEO and accessibility.

You’ll need to pay to publish, but for quickly generating a draft for a small business site, it hits the mark.

Verdict: Good foundations, but expect to spend time customizing.

Pricing: Starts from $12/mo

Wix: still pushy after all these years

Wix has evolved its platform with AI capabilities, and it shows in their smart onboarding questions. However, the experience quickly pivots to upselling domains, stores, and booking services before you even see a draft.

The generated site was functional but sterile-like those generic business cards people make and immediately forget. On the upside, all the features work right out of the box without technical setup.

Verdict: Reliable but feels more like a business funnel than a creative tool.

Pricing: From $17/mo

Readdy.ai: cool idea, cursed images

Readdy AI is a recently-launched tool (which made a splash on Product Hunt) that delivers decent design and copy with an intuitive editing experience. Unfortunately, the AI-generated images look like they came straight from a 2022 GAN model-uncanny and distracting.

While you can export static HTML for free, everything else costs money, and the free export still requires hosting knowledge.

Verdict: Promising, but the visuals need serious clean-up.

Pricing: Starts from $20/mo

Squarespace: a little soulless

Squarespace offers a clean, customizable onboarding process where you select site type, tone (“friendly,” “bold,” etc.), and required pages. The resulting site is polished but feels indistinguishable from every other Squarespace template.

The benefit is complete editability and Squarespace’s reliable infrastructure.

Verdict: A solid option if you need minimal fuss and don’t mind looking like everyone else.

Pricing: Starts from $16/mo

SITE123: aggressively oka

SITE123 doesn’t attempt to be cutting-edge. It feels like a platform from 2011-but in a comfortingly functional way. The AI primarily generates text, and you select stock images from their gallery. Everything works without surprises.

One notable quirk: I received six marketing emails within six hours of signing up. Someone’s AI is definitely enthusiastic about follow-up.

Verdict: Solid. Predictable. Gets the job done.

Pricing: Starts from $12.80/mo

Bolt: lightning-fast but a bit too bare

Bolt instantly generated a functional site with mobile responsiveness included. The downside? It’s almost too minimalist, and after hitting Deploy, you’re thrown into technical integration with GitLab/GitHub and Netlify-confusing for non-technical users.

Verdict: Risky for clients who just want a website without technical knowledge.

Pricing: “Free plans have a 150,000 token daily limit and 1 million token monthly limit.” (Whatever that means.)

10Web: beautifully slick wizard, generic outcome

10Web starts strong with a slick onboarding flow and smooth UI, building anticipation… only to deliver disappointment. The generated site was bland and filled with low-quality images. Worse, you can’t edit anything without paying.

Verdict: All form, no function-like a fancy box with nothing inside.

Pricing: Starts from $10/mo

Jimdo: confusingly popular, confusingly bad

Despite asking detailed questions and building expectations, Jimdo delivered a site with a headline reading “Strategic Consulting & Innovation”-for a cleaning company. Nothing felt tailored to the brief, and the resulting site lacked personality entirely.

Verdict: Feels like legacy software pretending to be modern.

Pricing: Free for basic features, then from $11/mo

Brizi.io: promising, then broken

Brizi begins with a business-focused wizard that seems promising. The result had a standard modern layout but with strange floating blobs across the page.

When I tried to edit, I encountered errors that refreshing couldn’t fix-suggesting it’s still in beta.

Verdict: One to revisit later, perhaps, but not ready for prime time.

Pricing: $190/yr

Reality check: what AI website builders can and can’t do

After testing these platforms, a few truths became clear:

What they can do:

• Generate functional site structures quickly
• Produce decent generic copy for standard business types
• Save time on initial setup compared to starting from scratch
• Provide usable results for simple business needs

What they can’t do:

• Understand your brand’s unique voice and values
• Create truly custom designs that stand out from templates
• Replace the judgment of an experienced designer
• Deliver sophisticated functionality without human configuration

So… are AI website builders the future?

They’re definitely not replacing actual designers. But if you’re a small business owner who just wants to stop handing out business cards with no URL on them, they’re a lifeline.

AI won’t magically understand your brand, your tone, or your customers. But it can give you a functional starting point in under an hour. And sometimes, that’s all you need.

Are there free website builders?

Yes and no. Most let you build and publish a basic site for free — usually on an ugly subdomain. But the moment you want to edit the layout, connect your own domain, or add anything useful (like analytics or forms), you’ll hit a paywall.

Just remember: no matter how good the AI, it’ll still need a human to care. That’s you.

About the author
Alexander Dallow. UX writer who transforms confusing interfaces into conversations users actually want to have. Constantly testing AI chatbots for flaws, owns more drumsticks than pens, and firmly believes that user flows and workout routines have more in common than people realize.

Originally published at https://blog.icons8.com on April 28, 2025.

Let’s skip the fluff-you need mockups that don’t suck, and you need them yesterday. After years of hunting (and several rage-quits over bad shadow rendering), here’s my real-talk breakdown of mockup tools that actually deliver.

Quick picks for the impatient

• Best for newbies: Shots — Completely free, browser-based, with no software knowledge required. Like design training wheels, but make it stylish.
• Best for physical objects: Maneken — When you need that packaging to look so real, clients will try to pick it up off their screens.
• Best for real-world mockups: Fockups — Because in real life, your beautiful billboard design will eventually get rained on and graffitied. Embrace the chaos!

Maneken

✅ Maneken offers trendy, photo-realistic mockups with shadows that actually behave like physics exists. Browser-based editing means no layer-hunting nightmares, and the free collection is genuinely usable.
❌ If you’re hunting for obscure device mockups, prepare for disappointment.

Shots

✅ Shots is completely free (yes, actually free) with browser-based editing that doesn’t make me want to throw my laptop. Their browser/mobile mockups are a chef’s kiss.
❌ The laptop mockups have that “I believe I can fly” vibe-floating in space with zero grounding.

Pixeden

✅ Pixeden is a free mockup goldmine with packaging options that make clients drool.
❌ Device selection is meh, and your non-designer friends are locked out (Photoshop required).

Mockup Cloud

✅ Mockup Cloud has solid all-around quality with delightfully weird options. (Needed an air freshener mockup last Tuesday. They had it. Don’t ask.)
❌ Some mockups have that uncanny valley problem-close to reality but just off enough to trigger my designer spider-sense.

Mr. Mockup

✅ Mr. Mockup offers trend-conscious, realistic mockups with a Figma plugin that’s saved me on deadline day more than once.
❌ Another Photoshop-dependent option. Sorry, subscription-dodgers.

Mockup-design

✅ Mockup-design has multiple angles per mockup (hallelujah!) and categorization that doesn’t feel like it was done by a drunk raccoon.
❌ Some mockups are aging worse than my first portfolio website. Plus, Photoshop.

Minimal Mockups

✅ Minimal Mockups is a free library for the budget-conscious.
❌ Photoshop required. Sensing a pattern?

Mockups Digital

✅ Mockups Digital-massive free collection.
❌ Classic aggregator issues-”free” sometimes means “surprise paywall!” And yes, Photoshop.

Mockuuups Studio

✅ Device selection at Mockuuups Studio that doesn’t quit, with browser-based editing that actually works.
❌ Their print mockup feature has been “coming soon” since I had a full head of hair.

Fockup

✅ The rebellious anti-mockup. Shows your designs in gloriously terrible real-world conditions-rain-soaked, graffiti-tagged, and victim to humanity’s infinite capacity for destroying nice things. Fockups were created from actual design disasters by Wytze Hoogslag’s Thansk studio.
❌ Your precious designs will look absolutely wrecked-but authentically so. Perfect for clients who need a reality check or brands with a sense of humor. Photoshop required (because beautiful destruction is still an art).

What is a mockup? The TL;DR version

A mockup visualizes how designs will appear in real-world contexts before implementation. It transforms abstract digital creations into contextual representations that help clients understand the final product.

Why designers (and clients) care

• Bridges the imagination gap. Converts abstract concepts into tangible assets
• Speeds up approval. Reduces revision cycles by showing practical applications
• Builds confidence. Helps clients see the design’s true potential
• Prevents expensive mistakes. Reveals potential issues before implementation

Think of it as test-driving a car versus reading its specs on paper-one experience simply sells better than the other.

Bottom line

The best mockup is the one that gets your client to stop asking questions and start signing checks. Pair any of these reality-bending tools with Icons8’s deliciously crisp icons and illustrations, and you’ve basically created a money printer. That sweet combo of contextual mockups and pixel-perfect graphics is the secret handshake to the “clients who actually pay on time” club. You’re welcome.

About the author
Anna Pearson. Self-taught design enthusiast who fell down the UI/UX rabbit hole. Applies the same organizing principles to her vinyl collection as her Figma files, captures photos with an obsessive eye for visual hierarchy, and mentally redesigns every confusing menu she encounters.

Originally published at https://icons8.com on April 16, 2025.

Everyone’s raving about OpenAI Operator — YouTube gurus hyping it as the next big thing, a productivity savior that’ll handle all your tedious tasks while you sip margaritas. “It’ll book you the cheapest flights! It’ll analyze tweets! It’ll simplify your work life!” they claim. Who wouldn’t throw down $200 to outsource all their mundane tasks to a robot?

Well, I did. And let me save you the suspense: I regret every cent.

So, what exactly is ChatGPT Operator?

If you’re wondering, “what is ChatGPT Operator?”, imagine hiring a virtual assistant who promises you the moon but delivers a small, slightly deflated balloon instead. In theory, this tool should automate your daily tasks-research, summaries, social media analysis, and more — saving you hours of mind-numbing work. Sounds perfect, right?

But after spending a week wrestling with this so-called game-changer, I’m convinced my productivity decreased. Dramatically.

My painfully honest ChatGPT Operator review

To be fair, I had high expectations, but here’s how it actually went.

1. Analyzing trending tweets about web design

The task seemed simple enough — ”Analyze trending posts about web design.” What I expected? Insightful highlights, popular discussions, actual trends. What I got? A messy list of tweets, half of which were duplicates or completely irrelevant. The operator proudly served up the same tweet three times — oh, and it was from two months ago. Hardly “trending.” It felt less like smart analysis and more like handing in an assignment five minutes before the deadline, hoping no one notices you copied Wikipedia word-for-word.

2. Summarizing the latest AI news

“Summarize the latest AI news.” Seems like a straightforward task, right? ChatGPT Operator’s approach: Find the first relevant Axios article, lazily summarize it, and call it a day. Did it check other sources? No. Did it provide comprehensive coverage? Nope. It felt like hiring someone who hands you the first Google result and proudly says, “Done!”

3. Listing popular AI-powered website builders and prices

I explicitly requested an independent analysis of AI-powered website builders, complete with pricing. ChatGPT Operator’s grand strategy? It took the first listicle it found on Bing and simply wrote out the first 10 items from there. When pressed for its methodology, it shamelessly admitted it copied everything straight from one source.

4. Summarizing multiple articles for a non-tech audience

My instruction was simple: “Summarize these articles for a non-tech crowd: [list of links to articles]”. ChatGPT Operator got to work… sort of. First, it summarized one article, then stopped and asked if it should continue. By the second time, it felt like dealing with a nervous intern on their first day, hovering awkwardly and asking, “Is this okay? Am I done now?” Then it spiraled into a bizarre loop of tab switching, as if it had completely lost the plot.

5. Taking website screenshots and adding them to a shared Google Doc

Initially, it confidently confirmed, “Sure, I’ll screenshot and upload everything to your doc.” Spoiler alert: It couldn’t take screenshots at all. Imagine telling your boss you’re fluent in French, then freezing when someone says “bonjour.”

6. Cherry on top — CAPTCHAs

Buried quietly on the OpenAI Operator’s landing page is a gem: it’s “trained” to hand tasks back to you whenever it hits logins, payments, or CAPTCHAs. That’s right — you pay premium prices for the luxury of solving those infuriating “select all images with traffic lights” challenges yourself. It’s like hiring a gardener who refuses to touch anything green.

ChatGPT Operator use cases (or lack thereof)

You might still think, it can’t be THAT bad. Unfortunately, its current real-world use cases seem limited to:

• Copy-pasting other people’s content (poorly).
• Annoying you with redundant questions.
• Failing miserably at tasks as simple as screenshotting.

How to use ChatGPT Operator (pro tip: don’t)

If you really must know how to use ChatGPT Operator, it’s straightforward:

• Open the app.
• Assign it a task.
• Immediately regret your choices.

Honestly, for now, it’s less a productivity booster and more an elaborate, expensive joke.

Verdict: Should you bother?

Here’s the harsh truth: ChatGPT Operator is basically an overhyped, underperforming intern. Instead of automating your workload, it’ll add more oversight, frustration, and, ironically, more work. Sure, tech hype is tempting, but in this case, you’re better off doing your own work — or hiring literally anyone else.

Maybe someday ChatGPT Operator will truly automate our mundane tasks flawlessly. Until then, save your $200. Spend it on coffee, therapy, or a puppy.

About the author:
Alexander Dallow, experienced content writer. Huge fan of AI, drums, and workout

Originally published at https://blog.icons8.com on March 27, 2025.