WhatsApp’s new AI feature, which it calls “Message Summaries”, has now been released.

The feature was announced last month and has since been in testing. Now, it’s stable and ready to roll.

At this stage, it’s only available in the US and only in the English language, but it’s expected to come out in more countries, with support for more languages in the future.

“Message Summaries” allows you to tap on the unread message count in a chat and get Meta’s AI to summarise what you’ve missed. The summary will only be visible to you.

WhatsApp conceives that you’ll use the function to catch up on chats that have become overwhelming or where you need to get up to speed quickly.

“Message Summaries” will be switched off by default. Here’s why I don’t think you should switch it on.

AI often gets it wrong

The first, and biggest problem, is that AI isn’t yet accurate enough to do all the jobs it’s being asked to.

Do a quick Google search now, and there’s a good chance you’ll see a Gemini AI overview at the top of the page. But beneath the information it produces, you’ll see a note that reads: “AI responses may include mistakes”.

Foundry

That’s pretty incredible for a service whose sole job is to provide you with accurate information, isn’t it? It’s like buying a packet of cookies and reading on the label: “Some of these cookies may not actually be food.”   

I’m not saying that Meta’s AI is notably worse than rival large language models, such as Google Gemini or ChatGPT, but all models still have issues, in spite of their widespread adoption. ChatGPT’s o3 model was recently found to hallucinate – providing plausible-sounding but false information – in 33% of answers to questions about publicly available facts.

Meta’s AI is prone to exactly the same problems. One use of “Message Summaries” will likely be to précis work chats. Can I suggest you don’t? Your job is probably in a field in which you really want to have your facts straight.

The summary has no space for irony or context

Hallucinations aren’t the only problem, although they’re arguably the most serious one.

You might be tempted to use “Message Summaries” to get up to speed with group chats with friends and, if you do, prepare for miscommunications.

For a start, people don’t always say what they mean. Sometimes, in fact, they say the very opposite (a little thing we call sarcasm).

Now, if you ask ChatGPT: “What do I mean by this sentence? ‘Wow. I’m really looking forward to that party’”, it’s smart enough to reply that it “depends heavily on context, tone, and delivery”.

But there’s no space to get that kind of ambiguity across in a message summary. If you look at the example from Meta (“Dave and Frank can start earlier”, “Jasmine and Pilar will finish up any remaining work later”), it’s very much relying on people being straightforward and not using any irony or sarcasm when they text.

Meta

And that’s not exactly a staple of group chats.

Then there’s nuance. If you’re anything like me, you’ve suffered from the 3am replay of the awkward comment you made or the joke that didn’t quite land.

Texting – without the benefit of hearing someone’s voice, or reading their facial expressions and body language – already magnifies the chances of these kinds of miscommunications, as tone can so easily be lost. Adding an AI interpretation between writer and reader only creates another opportunity for things to be misunderstood.

Why bother to chat in the first place?

Yes, it’s more efficient to read a summary than to read a chat. It’d also be more efficient if an AI wrote your responses for you as well. Is that where we’re heading? Seriously, if you’re going to summarise a chat, you might as well opt out of it altogether.

WhatsApp’s last new features drop was in April. Find out if you’re using them all.

And if you’re in the market for a new phone, check out Tech Advisor’s top Android phone and iPhone recommendations.