Christian Heilmann

Hot take: social platforms should disallow special character growth hacking

Sunday, July 7th, 2024 at 6:19 pm

Over the last few years something magical has happened without much fanfare: social media platforms and operating systems automatically translate text content for us. Having spent a lot of time traveling around conferences, I amassed a lot of people I follow who do not speak or write like me. And whilst it was fun following them, it was annoying not to know what they are talking about.

So, if a post is in another language or even in another glyph set, I can use inline translation or browser translation to make it understandable – excellent.

Things went awry though when people realised that UTF-8 has a lot of glyphs that look like other characters. Whilst this was a internet in-joke for a while – starting on IRC - it has become a “growth hacking” tactic. I repeatedly come across posts like these:

An example post on LinkedIn using lots of special characters and emoji to highlight parts of the text.

This, together with the over-use of emoji seems to be what a lot of growth experts tell people to stand out. Sure they look fancy, but in the first instance, they don’t automatically trigger the translation link offer.

If you put in the extra effort to copy and paste these messages, Google translate fails to give a good result. Bing does a good job with it but is limited to a certain amount of characters. Microsoft Edge’s in-built translation works OK.

Google translate failing to translate the headlines and garbling the long message in special characters

Bing translate recognising the special characters as the ones they should represent and giving a sensible translation

However, if you use a screen reader (and no, that’s not only fully blind people), you get absolute nonsense as shown in this video.

The text in this post gets spoken as:

Semicolon – that is a sentence for eternity from Elon Musk Smiling Face with Sunglasses. Is that thinking face? No, that is suprised face with open mouth! Speaker outputting high volume. Serious face with monocle. That seems to be a clear statement but the reality in Tesla is different.

Ironically, using an image instead of these UTF-8 shenanigans and providing a proper alt text would work for screen readers and even some translation software.

Ever since I started with the web people made the mistake of painting with text. The power of the web is that text can be converted, indexed and made accessible to others. When you garble your message to achieve a certain look or stand out in a busy timeline, you block out quite a lot of people who further down the line could become important multipliers or customers.

And it is not clever and comes across as trying too hard.

It also can’t be in the interest of the platform to support this, as it does make content harder to index and search. It’s a classic case of a “clever hack” that in the end makes the implementer look like a fool and the end users suffer.

Share on Mastodon (needs instance)

Share on Twitter

Newsletter

Check out the Dev Digest Newsletter I write every week for WeAreDevelopers. Latest issues:

Dev Digest 146: 🥱 React fatigue 📊 Query anything with SQL 🧠 AI News

Why it may not be needed to learn React, why Deepfake masks will be a big problem and your spirit animal in body fat! 

Dev Digest 147: Free Copilot! Panel: AI and devs! RTO is bad! Pi plays!

Free Copilot! Experts discuss what AI means for devs. Don't trust containers. Mandated RTO means brain drain. And Pi plays Pokemon!

Dev Digest 148: Behind the scenes of Dev Digest & end of the year reports.

In 50 editions of Dev Digest we gave you 2081 resources. Join us in looking back and learn about all the trends this year.

Dev Digest 149: Wordpress break, VW tracking leak, ChatGPT vs Google.

Slowly starting 2025 we look at ChatGPT vs Google, Copilot vs. Cursor and the state of AI crawlers to replace web search…

Dev Digest 150: Shifting manually to AI.

Manual coding is becoming less of a skill. How can we ensure the quality of generated code? Also, unpacking an APK can get you an AI model.

My other work: