Emoji That Look Identical on One Phone but Not Another, Measured

Whether two similar emoji look the same depends on the phone that draws them. We measured 119 easily confused pairs on four emoji sets and found that 😀 and 😄 differ by just 1.8% of pixels in Apple's artwork but by 14.5% in Twemoji. Seven pairs sit under 5% on Apple yet clear 10% on Twemoji. Only two pairs stay under 5% on every set we tested.

So the face you pick carefully on your phone may arrive looking like its neighbour, or it may arrive looking more distinct than you expected. Below are the numbers, the pairs to watch on each platform, and the substitutes that hold up everywhere.

How the measurement works, in plain terms

The pairs come from the 21 groups of lookalike emoji on our emoji lookalikes page: 80 emoji in total, and every pair within each group, which gives 119 pairs. 63 of those pairs are faces.

On 19 October 2026 a script downloaded each emoji from four artwork sources: Apple's set as packaged by the open source emoji-datasource project, Google's Noto, Twemoji 14.0.2 (the open source set behind Twitter's emoji) and OpenMoji. Each image was placed on a white background and scaled to 64 by 64 pixels. The script then compared the two emoji in a pair pixel by pixel.

A pixel only counts as changed when its red, green and blue values move by more than 96 in total, out of a possible 765. That cut-off ignores the soft grey edges every image has and still catches a real stroke like an eyebrow. The figure we report is the share of the 4,096 pixels that changed. Lower means less to look at.

One honest caveat up front: this is a count of pixels, not a test of human eyes. More on what that misses further down.

The hardest pair to tell apart on each set

Each set has its own worst offender, and no two sets share the same one:

• Apple: 😀 grinning face versus 😄 grinning face with smiling eyes, 1.8%.
• Google: 😀 grinning face versus 😃 grinning face with big eyes, 2.6%.
• Twemoji 14: 😐 neutral face and 😶 face without mouth, 2.7%.
• OpenMoji: 🙁 slightly frowning face versus ☹️ frowning face, 0.3%, the closest result in the whole data set.

That OpenMoji figure means roughly a dozen pixels out of 4,096 separate the two frowns. Our lookalikes curation describes them as "Plain face, slight frown" and "Plain face, deeper frown". At 64 pixels, OpenMoji's line art leaves almost nothing of that difference.

The two frowns also have different ages. In Unicode's emoji list, ☹️ is Emoji 0.7 and 🙁 is Emoji 1.0, so the smaller frown arrived in a later emoji release than the deeper one.

Seven pairs that blur on Apple but not on Twemoji

This is the core finding. If the person you are messaging uses Apple's artwork, these seven pairs are each under 5% apart. On Twemoji 14 every one of them is at least 10% apart:

• 😀 and 😄: Apple 1.8%, Twemoji 14.5%
• 😀 and 😃: Apple 2.7%, Twemoji 11.9%
• 😯 hushed face and 😧 anguished face: Apple 2.0%, Twemoji 11.8%
• 😯 hushed face and 😦 frowning face with open mouth: Apple 4.5%, Twemoji 10.2%
• ☺️ smiling face and 😌 relieved face: Apple 4.1%, Twemoji 22.3%
• 🙂 slightly smiling face and 😊 smiling face with smiling eyes: Apple 4.6%, Twemoji 18.8%
• 😮 face with open mouth and 😲 astonished face: Apple 4.8%, Twemoji 11.4%

The ratio on the first pair is about eight to one. The ☺️ and 😌 pair is the one we would worry about most, because the meanings are not close: one is warm, the other is a sigh of relief. On Apple's drawing they share most of their pixels. On Twemoji they are clearly two different faces.

Google sits between the two on most of these. It keeps 😀 and 😃 just as close as Apple does (2.6%), but separates 😯 and 😧 better (5.5%).

Where Google or Twemoji is the one that blurs

Apple is not always the problem. A few pairs flip the other way, and they are mostly outside the faces.

• 🌚 new moon face against 🌑 new moon: Twemoji 8.3% and OpenMoji 5.4%, but Apple 30.2% and Google 64.4%. On Google's set the two moons look nothing alike. On Twemoji the face on the moon is a small detail on a large dark disc.
• ⭐ star against 🌟 glowing star: Google 6.5%, against 21.9% on Apple, 27.2% on Twemoji and 25.6% on OpenMoji. Google is the only set where the plain star and the glowing one are a near match.
• 💕 two hearts against 💞 revolving hearts: Twemoji 6.7% and OpenMoji 3.3%, but Apple 17.9% and Google 22.8%.
• 🌝 full moon face against 🌕 full moon: Twemoji 9.2% and Google 12.6%, against 24.4% on Apple.

So "Apple users get the worst lookalikes" is not right either. Apple compresses the grins and smiles. Google, in our measurement, compresses ⭐ and 🌟. Twemoji compresses the moons and the pink hearts.

The two pairs nobody draws apart

Only two of the 119 pairs stay below 5% on all four sets. These are the ones to avoid if you cannot afford to be misread, because no recipient's phone will rescue you.

1. 😦 frowning face with open mouth and 😧 anguished face: Apple 2.5%, Google 3.8%, Twemoji 3.9%, OpenMoji 2.6%. Our spotting note says the second adds "worried brows" to the same frowning open mouth. Brows are a few pixels wide.
2. 😐 neutral face and 😑 expressionless face: Apple 2.6%, Google 2.9%, Twemoji 4.8%, OpenMoji 1.0%. The whole difference is dot eyes versus line eyes.

Both pairs follow the same pattern. The faces share a mouth and differ only in the eyes or brows, which are among the smallest features on the face.

Why OpenMoji wins almost every closeness contest

OpenMoji draws the closest version of 93 of the 119 pairs. Google is closest on 14, Apple on 9 and Twemoji on 3. That looks dramatic, and it mostly reflects style rather than carelessness.

OpenMoji uses flat colour with a dark outline and very little shading. A face like that has fewer pixels that can change: the yellow fill stays identical and only the thin lines move. Apple's set leans on gradients, highlights and shadows, so even a small change in a feature can shift the shading around it and add to the count.

It is not universal, though. On the hands group OpenMoji is the least alike of all four sets, averaging 54.5% against Google's 39.5%, and 🙌 against 🙏 hits 66.0%. On ❤️ red heart against ♥️ heart suit, OpenMoji is the most different at 30.0%, while Google is the closest at 11.6%. Where the shapes themselves differ, an outline style makes the difference stand out.

If you set OpenMoji aside and compare only Apple, Google and Twemoji, Google is closest on 55 pairs, Apple on 52 and Twemoji on 12.

All four sets ranked by how alike their lookalikes are

Averaging every pair gives each set a single closeness score. Lower averages mean the set leaves you less to tell apart:

• OpenMoji: average 19.2%, median 14.2%. 27 pairs under 5%, 51 under 10%.
• Google: average 29.2%, median 24.2%. 7 pairs under 5%, 25 under 10%.
• Apple: average 30.5%, median 26.4%. 14 pairs under 5%, 32 under 10%.
• Twemoji 14: average 35.2%, median 28.0%. 3 pairs under 5%, 15 under 10%.

Apple and Google have similar averages but very different tails. Apple has twice as many pairs under 5%, and all 14 are faces. Google's closest calls are spread more thinly. Twemoji keeps its lookalikes furthest apart on this measure, which may reflect its flat style and large features.

The group averages show the same story more sharply. Across the four grinning faces, the average pair differs by 4.2% on Apple, 8.2% on Google, 11.4% on Twemoji and 2.3% on OpenMoji. The check marks group sits at the other end: 91.9% on Apple, because ✅ is a filled green button and ✔️ is a bare tick.

For a broader look at how each company draws emoji, see our pages on Apple emoji, Google emoji and Twitter emoji.

What a pixel count cannot tell you

Pixel difference is a proxy. It is useful because it is objective and repeatable, but it is not the same as what people perceive. Three limits are worth spelling out.

Colour beats meaning. A change of colour across a whole face scores high even when the meaning barely moves. 😠 angry face and 😡 enraged face differ by 79.4% on Twemoji, largely because one is red. Meanwhile a tiny eyebrow that turns surprise into anguish scores under 4%. The numbers measure how much changed, not how much it matters.

Size changes everything. We compared at 64 by 64. In a chat bubble an emoji is often smaller than that, which makes small features even harder to see. On a large preview or a sticker the same pairs are easier.

Twemoji 14 is an older set. We used Twemoji 14.0.2 because it is the version our site renders. The X app may draw some emoji differently today, so treat the Twemoji column as a reference for that design style rather than a screenshot of the current app.

Finally, the Apple images come from a packaged copy of Apple's set, not from a device running the latest iOS. Apple occasionally redraws emoji, as our post on emoji redesigns that changed meaning shows, and any redraw could move these figures.

Picks that stay distinct on every phone

What you really want to know is which emoji you can send without worrying about the recipient's device. For each face group, this is the pair that stays furthest apart on its weakest set:

• Grins: 😃 and 😁 never drop below 3.9%, and sit at 7.7% or more on everything except OpenMoji.
• Laughing: 😂 and 🤣 never drop below 20.6%. Tears plus a tilted head is hard to miss.
• Sad: 😔 pensive face and 😟 worried face hold 13.1% or more everywhere except OpenMoji (6.8%).
• Blank faces: none are safe. Even the best pair, 😑 and 😶, drops to 1.9% on OpenMoji. If the difference matters, add a word.
• Tongue out: 😜 and 🤪 zany face keep at least 9.7%, while 😛 and 😝 drop to 1.7%.

Two rules come out of the data. Pick emoji that differ in the mouth or in colour, not just in the eyes. And when the exact shade of feeling matters, for an apology or bad news, back the emoji with a word, because no single glyph is guaranteed to survive the trip.

Checking what your recipient actually sees

You cannot see your friend's screen, but you can get close. Our side-by-side lookalike tables show the Apple, Google, Twemoji and OpenMoji drawing of every emoji in these 21 groups, so you can check a pair before you send it. Each emoji page, such as 😀 grinning face, also shows the vendor images together.

This measured, per-platform problem is different from the one covered in our post on why some emoji look identical in Unicode itself, where two separate characters such as cucumber and pickle exist for historical reasons. Here, the characters are clearly different in Unicode, and the confusion is purely in the drawing. For the wider question of why the art differs so much between companies at all, read why emoji look different on iPhone and Android.