It's been a couple of weeks since you heard from me. I'd love to tell you the AI world slowed down while I was away. It did not.
So this issue is a little different. First, a quick catch-up on what changed, keeping only what you can actually use here in Canada today. Then the question I get asked most: what's a model, and which one should you use? And finally, four prompts so you can try it yourself.
There's also a thread running through all of it: keeping your voice. More on that in a minute.
What changed while I was away
ChatGPT got a big update, mostly for paid plans. OpenAI started rolling out GPT-6 on October 7 to Plus and Pro subscribers. The interesting part isn't the name. It's that answers can now come back as a chart, a form or a small calculator instead of a wall of text. Free accounts didn't get GPT-6. They got a newer everyday model instead, which is still plenty for most things.
Free Gemini got a little smaller. Since October 9, the free version of the Gemini app is set to Auto, which picks the model for you, and most answers now come from Google's lightest model. If Gemini has felt a bit less sharp lately, that's probably why. For a bigger job, it's worth running the same prompt in free ChatGPT or Claude and comparing.
Claude's document, slide and design tools are now free. On October 8, Anthropic opened up Claude Docs, Slides and Design to every plan, including the free one. Anyone can now make a flyer, a short slide deck or a clean document right inside Claude. Prompt 3 walks you through a flyer. Anthropic also launched Dashboards and Motion, but those are paid and aimed mostly at workplaces, so I'll come back to them if that changes.
What's a model, anyway?
You've probably seen headlines like “OpenAI releases GPT-6” or “Google launches a new Gemini.” Those aren't new apps. They're new models.
Here's the simplest way I can put it. ChatGPT, Gemini and Claude are the apps you open. The model is the engine inside the app that reads your question and writes the answer. Each company builds several, from small fast ones to big careful ones, and most apps let you choose which one you're using.
Why does it matter? Bigger models handle complicated problems better. But they're slower, and they use up your allowance faster. To understand that part, you need one more word.
What's a token?
A token is a small piece of a word. It's how AI companies measure how much work you're asking for. OpenAI's rule of thumb is that 100 tokens is about 75 words. Their own example even comes from the Great One: Wayne Gretzky's line “You miss 100% of the shots you don't take” is 11 tokens.
Every plan, free or paid, has a limit on how many tokens you can use. Bigger models and more thinking use more tokens for the same question. That's the trade-off.
Free doesn't mean unlimited. You may have seen OpenAI say free ChatGPT comes with unlimited everyday chats. Here's what that actually means: unlimited plain back-and-forth typing on its fastest setting. The moment you upload a photo or a document, ask for an image, or use voice, you're on a different allowance with its own limits. Free accounts, for example, get three file uploads a day.
Uploads add up fast because the AI has to read every word of the file. A 10-page lease can run several thousand tokens before you've even asked your question. Google and Anthropic both say longer chats and the features you use count against your limit too. Hit the limit and you wait for it to reset.
The models in each app
GPT-5.6 Luna: the free plan's model. OpenAI's fastest and lowest-cost. Fine for everyday questions.
GPT-6: the new main model, for paid plans (Plus and up). This is the one that can answer with charts and small tools.
GPT-5.6 Sol: paid plans. Built for heavier work like research and analysis.
GPT-6 Pro: the top plans only (Pro, Business and Enterprise). For the hardest, longest jobs, with its own usage cap.
How to switch: in the box where you type, click Thinking effort, just left of the microphone. On a paid plan, click the level name (like Medium) to choose GPT-6 or GPT-5.6 Sol, then drag the slider. Further left is quicker, further right thinks longer. On a free account you can't change the model, but the same button lets you ask for a more careful answer.
Haiku: the fastest, and the lightest on your limit. Free and paid plans.
Sonnet: quick and capable. Good for everyday writing and planning. Free and paid plans. On the free plan, Haiku and Sonnet are what you get.
Opus: the default on paid plans. Anthropic describes it as built for complex work and everyday tasks.
Fable: Anthropic's biggest model, for the hardest and longest jobs. Included with the Max plan. On Pro, you pay extra for it with usage credits. It's the slowest, and it uses the most.
How to switch: click the model name at the bottom right of the message box, next to the send button. Pick one, or click More models to see the rest. Effort, in the same menu, sets how long Claude thinks.
Flash-Lite: quick answers to everyday questions. The free plan is set to Auto, and most answers come from this one.
Flash: a balance of speed and ability. Paid Google AI plans.
Pro: for complex questions and in-depth problem solving. Paid Google AI plans.
Argon: Google's top model for the most complex work. Google AI Pro and Ultra only.
How to switch: on gemini.google.com or in the Gemini app, tap the model name at the bottom of the chat box. Free accounts can't change the model, but can choose a thinking level there: Low, Medium or High. It starts on Low.
So which one should you use?
For everyday stuff, the free models are plenty. That's GPT-5.6 Luna in ChatGPT, Haiku or Sonnet in Claude, and Gemini on Auto. Rewording a text, a recipe substitution, a thank-you note, a quick question. They'll handle it without making a dent in your limit.
For planning and comparing, work on your prompt before you change the model. Building a week of meals around a food allergy and a budget, or writing a firm email to your landlord about a repair. On a free plan, stay right where you are: GPT-5.6 Luna in ChatGPT, Sonnet in Claude, or Gemini on Auto. If you pay for ChatGPT, GPT-6 on Medium is where I do most of my own work. Either way, the bigger difference usually comes from your follow-up, not the model. If the first answer misses something, tell it what's missing and ask again.
Pulling details out of a document is a Medium job too. Asking AI to find the key dates, costs and conditions in a lease, an insurance renewal or a mortgage letter works well without the biggest model. Just remember the upload uses up tokens. And for anything that really matters, AI gives you a head start on your questions, not the final answer. Talk to a professional before you sign or decide.
Save the biggest models for deeper research and analysis. This is where they earn their keep: pulling together information from many sources and working through it carefully. Think researching whether a heat pump makes sense for a prairie winter, or which used minivans hold up best in Canadian cold. On a paid plan, that's GPT-6 on High in ChatGPT, Opus in Claude, or Pro in Gemini. ChatGPT also has Deep research in the left sidebar, which digs through many sources and takes several minutes. Gemini and Claude have their own versions. Google notes these use more of your allowance than a regular chat, so save them for questions that are worth it.
Start with the default. If the answer feels thin, improve your prompt first. If it's still thin, try a bigger model, then switch back. And bigger never means always right. You still read it, check it and make it sound like you.
Want to check it yourself? Here's what each company says: OpenAI on its models and tokens, Google on its models and limits, and Anthropic on its plans and switching models.
Why your voice matters more now
Every one of these updates makes AI writing smoother. Mostly, that's good news. But smooth isn't the same as you. When a note to your kid's teacher, a message to your landlord and a post in your neighbourhood Facebook group all come out sounding like the same polished stranger, people notice.
The fix isn't to stop using AI. It's to give it more of you to work with. Prompts 2 and 4 are built around exactly that.
Four prompts to try this week
1. Ask for an answer you can play with
If you pay for ChatGPT, you can ask for something interactive and it will often build it right there in the chat. On a free account you'll get a regular written answer, which still works for this one.
Danielle in Moncton grabs a coffee and a muffin on the way to work most mornings. She asked ChatGPT what that adds up to, with a slider to test cutting back to three days a week. Watching the yearly number drop did more than any budget lecture ever could.
“Show me what my [HABIT] costs over a year. It's about $[AMOUNT] each time, [NUMBER] times a week. Make it interactive, with a slider so I can change how many times a week, and show the yearly total as a chart.”
“Show me what my morning coffee and muffin run costs over a year. It's about $6 each time, 5 times a week. Make it interactive, with a slider so I can change how many times a week, and show the yearly total as a chart.”
If you get plain text back instead, just reply “Can you make that interactive?” The update has been rolling out gradually, so not everyone sees it right away.
2. Talk it out, then tidy it up
This is my favourite way to keep my voice in anything AI helps me write. Instead of typing a careful prompt, I just talk. Tap the microphone in the ChatGPT, Gemini or Claude app and ramble about what you need, the way you'd explain it to a friend. Then ask AI to organize it without sanding off your words. I do this almost every day with Wispr Flow, a voice-to-text app that works in any app on my computer, not just the AI ones. It's the biggest reason my drafts still sound like me.
Raj in Saskatoon is hosting his family for the holidays for the first time. He talked for three minutes about who's coming, who doesn't eat what, and his oven that only fits one roasting pan. AI turned it into a to-do list and a group text to his siblings that still sounded exactly like Raj.
“I'm going to talk through [WHAT YOU'RE PLANNING OR WRITING]. Turn what I say into [WHAT YOU WANT BACK]. Keep my words and phrasing wherever you can, and only clean up what's needed. Here it is: [TAP THE MIC AND TALK]”
“I'm going to talk through my plans for hosting the family over the holidays. Turn what I say into a simple to-do list and a short group text to my siblings. Keep my words and phrasing wherever you can, and only clean up what's needed. Here it is: [tap the mic and talk it through]”
If you pay for ChatGPT, you can now also upload a recording you already have, like a voice memo, and it will transcribe and summarize it. Either way, if anyone else's voice is on a recording, make sure they know and are okay with it.
3. Make a flyer in Claude Design
Open Claude on your computer and look for Design in the sidebar, or go to claude.ai/design. Describe what you need and you get a flyer you can tweak, download and print. The trick for keeping it yours: write the words yourself and let Claude handle the layout, not the message.
Colleen in Winnipeg volunteers at her community club and got roped into making the flyer for the family skate. She wrote three sentences the way she'd say them to the neighbours and asked Claude Design for something bright and wintery. Ten minutes later it was on the club bulletin board.
“Make a [SIZE] flyer for [EVENT] on [DATE AND TIME] at [PLACE]. Here's the wording I want, in my own words. Please keep it as written: [YOUR WORDS]. Style: [THE LOOK YOU WANT].”
“Make a letter-size flyer for our community club's family skating party on Saturday, December 12, from 1 to 4 p.m. at the outdoor rink. Here's the wording I want, in my own words. Please keep it as written: ‘Dust off the skates and bring the whole crew! Hot chocolate is on us, and the warm-up shack is open if your toes give out. Everyone welcome, no skills required.’ Style: bright, friendly and wintery, easy to read from across the room.”
A quick privacy note. On anything you'll post publicly, use a general contact like the club's email, not your personal phone number or home address.
4. Teach AI how you actually sound
Paste in two or three things you've actually written and ask AI to describe your style. Save what it gives you in a note on your phone, then paste it into any prompt where you want the result to sound like you.
Linda in Kelowna kept getting AI drafts that sounded like a bank's customer service email. She pasted in three texts to her sister and an email to her book club. AI described her as warm, a little dry, fond of exclamation marks, and someone who would never in her life write “utilize.” Now she pastes that list in whenever she wants help writing something.
“Here are [NUMBER] things I've written: [PASTE YOUR WRITING]. Describe my writing style in five short points: my tone, my sentence length, words and phrases I tend to use, and words I would never use. Keep it short enough that I can paste it into future prompts.”
“Here are three things I've written: [paste a text to a friend, an email to a coworker and a social media post]. Describe my writing style in five short points: my tone, my sentence length, words and phrases I tend to use, and words I would never use. Keep it short enough that I can paste it into future prompts.”
Before you paste anything in, take out names, addresses and anything personal about other people. AI only needs to know how you write, not who you wrote to.
Your challenge this week
Start with Prompt 4. It takes five minutes. Then use your style description with one of the other three prompts, or with something real you need to write this week. Read what comes back out loud. If it doesn't sound like you, tell AI what's off and try again.
Coming up next issue
Next issue we're talking about AI and your kids. What they're already using, what's worth paying attention to, and what honestly isn't worth losing sleep over. Whether you're a parent, a grandparent or the favourite aunt, this one's for you.
Whatever AI produces for you, always read it, verify it, and make it sound like you. AI is an incredible first draft machine. Your judgment, your voice, and your values are always the final layer.
Has AI ever written something for you that just didn't sound like you? Read it out loud, tell it what's off, and try again.