Meta AI is probably already sitting inside the app you use to text your family. That’s the strange part. It’s built into WhatsApp, Instagram, Messenger and Facebook, it has its own standalone app, and it lives inside a pair of sunglasses. Which means millions of people have access to a genuinely capable assistant and have opened it exactly once, typed “hi”, watched it say hi back, and closed it forever.
This guide is about the second visit. Below are concrete tasks Meta AI handles well, the prompts that make them work, and the places it quietly falls apart. No product tour, no theory.
Know Which Meta AI You’re Actually Opening
There is no single Meta AI. What you can do depends entirely on where you’re standing.
- The standalone app — launched in 2025, separate from Facebook and Instagram. Best for longer sessions and voice conversation.
- WhatsApp, Instagram, Messenger and Facebook — the blue-purple ring in your chat list, or an @Meta AI mention inside a group thread.
- meta.ai in a browser — fine for drafting on a laptop keyboard.
- Ray-Ban Meta glasses — a camera and microphone on your face, which changes what’s worth asking.
The engine underneath is Meta’s Llama model family. That’s why the tone feels different from ChatGPT: chattier, more willing to riff, occasionally too eager to please. Availability still varies by country, and features roll out in waves, so if something described below isn’t there yet, it may simply not have reached you.
If you want the fuller picture of how the assistant spread across almost every Meta product, that’s worth ten minutes. This piece is about what to do once you’ve found it.
The One Habit That Separates Good Output From Garbage
Most people type a question. Good results come from typing a brief — and a brief has four parts: the task, the audience, the format, and the constraint.
Weak version: “Write a caption for my holiday photo.”
Better version: “Write three Instagram captions for a photo of my dog on a Cornish beach at sunset. Dry humour, no emoji, under 12 words each.”
That’s maybe fifteen extra seconds of typing. The constraint — under 12 words, no emoji — is the part people skip, and it’s the part doing the heavy lifting. Without it you get the same bland caption the model produces for everyone.
Five Tasks Meta AI Handles Well
1. Collapse a runaway group chat into decisions
If Meta AI has been added to a WhatsApp group, you can @ mention it. Otherwise, forward the thread or paste the text and ask:
“Summarise this conversation. List every decision we made, who agreed to do what, and anything still unresolved.”
On a 300-message thread about a stag weekend, this takes about four seconds and reliably surfaces the one thing everyone forgot — usually the deposit.
2. Rewrite a message before you send it
This is the single most useful habit I’ve built with Meta AI, and it takes under a minute. Type your angry version first, then hand it over:
“I’m annoyed about a delivery that’s nine days late. Rewrite this so it’s firm but not rude, under 80 words, and ask for a specific resolution with a date.”
You still send it. You just send the version you won’t regret at 2am. It’s especially handy when you’re writing in a second language and can’t tell whether your phrasing sounds blunt or obsequious.
3. Plan something with real constraints
Vague requests get vague itineraries. Load it up instead:
“Plan a Saturday for four adults and one dog near Bristol. Budget £60 per person, one indoor option if it rains, no more than 20 minutes’ drive between stops, and somewhere to eat that takes walk-ins.”
You’ll get something usable. You’ll also get one suggestion that doesn’t exist or closed in 2023, so spot-check addresses before you commit.
4. Pressure-test a purchase
Screenshot a used bike listing from Marketplace and ask: “List five questions I should ask the seller and three red flags to look for in these photos.” The list is rarely clever. It’s just that having it in front of you stops you buying a cracked frame because the paint is nice.
5. Fix a photo before you post it
Meta AI’s image tools can edit rather than just generate. Point it at a decent shot with a wheelie bin in the corner and ask for the bin to go. Ask it to extend the sky in a photo where you cropped too tight. Ask for a specific restyle — “same photo, warmer light, film grain” — rather than “make it better”, which gets you a plastic filter and a slightly different face.
Where Meta AI Falls Over
Being honest about the limits saves you a lot of wasted time.
- Anything requiring current, verifiable facts. Opening hours, prices, sports results, who currently holds a job title — treat every answer as a rumour until you check.
- Long documents. In-app, it’s built for chat-sized input. Forty pages of a contract will get skimmed, not read.
- Precise arithmetic and figures. It will produce a confident number that is simply wrong.
- Disagreement. It leans agreeable. If you say “I think the second option is better”, expect it to find reasons you’re right. Say “argue against the second option” instead.
The gap here isn’t cosmetic. The most interesting work in AI right now is on systems that reason through a problem rather than pattern-match a plausible answer, and chat assistants of Meta AI’s type are mostly on the other side of that line.
Same Prompt, Different Assistant
Worth a quick reality check, because nobody uses only one of these.
Meta AI wins on proximity. It’s already open, it’s free, it handles voice well, and it can see your group chat — which no competitor can. ChatGPT remains stronger for long documents, code and structured output, and if you want to know where OpenAI’s model lineup stands in 2025, that’s a separate read. Gemini has the Google data behind it.
The practical rule: use the one that’s already on screen. Copying a prompt into a second app costs more than the quality difference is usually worth.
Making It Repeatable Instead of Novel
The novelty wears off in a week. What sticks is a small set of prompts you reuse, which is basically the same process as setting up an assistant for real work, just shrunk down.
Three things that help:
- Save your five best prompts in WhatsApp’s “message yourself” chat so they’re one tap away.
- Use voice mode for anything you’d otherwise type one-handed — drafting replies, brainstorming a gift, thinking out loud on a walk.
- Check the memory settings. If you want it to remember your job, your kids’ ages or your writing style, you have to let it; if you’d rather it forgot, turn that off.
Prompts Worth Stealing Tonight
Copy these, change the details, see what happens.
“Here’s a message I need to send to my landlord. Make it polite, short, and make sure it includes a deadline.”
“I have 40 minutes and £15. Give me three dinner options using what’s probably already in a normal fridge.”
“Read this paragraph back to me and tell me what I actually said, not what I meant to say.”
“Ask me five questions about my week, one at a time, then write a three-line update I can send my manager.”
That last one is the real unlock. Meta AI works best when you treat it as a colleague who needs briefing rather than a search box that needs filling — and the briefing is the skill worth practising, whichever assistant you end up using next year.