As WeChat Tests AI Writing for Moments, I Want to Write My Own Words Even More

When I saw that WeChat was testing “AI Writing” and “AI Commentary” for Moments, my first thought was: oh no.

My partner had roughly the same reaction.

It is not that WeChat should not build AI features, or that other people should not use them. One thing I like about Moments is that when I see a sentence, I generally know who wrote it. Some people write long posts. Some search around for a caption, arrange their photos, and add only a few emoji. Even if it is only a few symbols, they usually still thought about them, chose them, and typed them themselves.

According to current reports, AI Writing can suggest several captions from the text and images in a draft. AI Commentary is also being tested with a small group. These features are not available to everyone, and reports from the current test indicate that the user may still need to confirm the final post. The details are covered by The Beijing News and IT Home in Chinese.

The default assumption was that a person wrote it

People have always copied captions. Anyone could also generate something elsewhere and paste it into Moments. The difference is not that this was impossible before. It just was not something people did by default.

When I read something written by a friend, I do not usually begin by asking whether a machine wrote it. I assume it is what they wanted to post that day. It may be awkward or unclear. It may be only a few emoji. The choice still belongs to them.

Moments also feels like a small space people arrange for themselves. Choosing a photo, writing a line, or removing one are minor actions. Later, seeing that photo again may remind someone why they posted it that way.

If AI writing becomes a standard feature and everyone knows a caption can be generated immediately, people may start reading with a new question: did my friend want to say this, or did AI fill it in?

I am not afraid of Moments suddenly filling up with smooth captions. I just know I would start wondering: did my friend actually write this? Copying was always possible, but suspicion did not have to come first.

I often use AI too

I am not against AI. I use it often while developing Inks—for research, coding help, and engineering work. It is genuinely useful. I also occasionally use it to polish release notes or other public-facing copy.

I treat those tasks as work: get the facts right, then let the tools make the rest faster.

I make a different choice when writing to myself, my family, or close friends. The thoughts I save in flomo are usually typed by me or dictated in my own words. I rarely send them through AI afterward.

I wrote more about what I actually save there in my flomo recording workflow. The everyday observations and product ideas in that article also stay in the form in which I originally entered them.

Some thoughts are incomplete, and some wording is clumsy. When I read it again later, I can still tell that it was me at that moment.

Inks will not write or polish a letter for you

This is not something I simply have not had time to build. Inks will not do it.

Inks preserves the finished letter, but it also preserves how words were added, deleted, and changed, along with pauses during writing. A letter opens to the finished result. Its writing process appears only when someone chooses to revisit it.

I wrote separately about why the result and the process stay distinct in Why Preserve the Writing Process.

What Inks tries to keep is how those words were actually written.

Generating a whole paragraph with AI would not be difficult. But the record would only show a block of text appearing all at once. It would not show how the writer reached the thought, why they changed their mind, or why they kept the final sentence.

For that reason, Inks will not add AI ghostwriting, continuation, or polishing. The writing itself is still left to the letter writer.

AI may help after the writing is finished, but it will not write for you

Could Inks ever include an AI-related feature? Possibly, but the boundary would be clear.

With the user’s explicit permission, it might one day look only at letters the user has already finished and prepare a short monthly or yearly review. It might help arrange past events into a timeline so the user can revisit what happened during a period of life.

These are not current features, and there is no settled implementation. If I ever build something like this, it would only use work the user had already finished and had explicitly agreed to share with AI. It would not change the original words, and I would first have to explain clearly how the content is handled.

I do not yet know whether I will build any of these ideas. At least for now, they are not the focus.

Some words do not need to be efficient

I use AI while coding because I want to work faster. But when I write a few words to myself, my family, or a friend, I do not really want to save those few minutes.

We already communicate quickly enough. Messages need quick replies, and writing is expected to stay short. If I also let AI smooth out the words meant for people close to me, I feel that something is missing.

Writing slowly or clumsily is fine. At least I wrote it myself.

6 min read · 2026-08-31 · 2026-08-31

By Wenpeng Wang, Developer of Inks.