Trigger: they're on a free tier, a weak model, or they don't know what a model is. This is the single biggest quality jump available and costs one conversation to fix.
Establish context
Tell them which model you are right now and what effort or thinking level you're running at, if you can see it. If you can't tell, say so — don't guess.
Ask them: do you know which model you picked, or did you just open the app?
Most people never chose. They signed up, it defaulted to something, and they assumed that's what AI is. If that default is a free or low-tier model, everything they've experienced so far is the floor, not the ceiling.
What changes
There are different models from different providers, and within each provider there are tiers. Think of it as different levels of intelligence — not different personalities, different capability.
The practical difference: a lower-tier model will give you a plausible-sounding answer that's shallow or wrong. A higher-tier model will actually understand what you meant, push back when your question is unclear, and produce work you can use without rewriting it. The gap is not subtle.
What to do right now:
- If they're on ChatGPT free: that's GPT-4o mini or similar. It's fast and cheap and it's the reason most people think AI is a toy. A paid subscription gets them access to the real models.
- If they're on ChatGPT paid: they have model choices. The latest GPT model on medium effort is a reasonable default.
- If they're on Claude free: they're on Sonnet, which is middling. A subscription gets Opus, which is substantially better.
- If they're on Claude paid: use Opus. Not Haiku (fast but shallow), not Sonnet (the middle that satisfies nobody). Opus on medium effort. Max effort isn't better — it can overthink a simple problem.
Why effort matters: when you type a question, the model thinks before answering. Effort controls how much thinking it does. More thinking costs more of your usage budget. For everyday work, medium is the sweet spot — enough thought to be genuinely useful, not so much that it spirals.
The budget: every provider gives you a weekly usage limit on a subscription. Bigger models eat it faster. If they blow through it mid-week, they either switch to a lighter model for quick questions and save the heavy one for real work, or they upgrade their plan.
One sentence to leave them with: the quality of what you get is bounded by the intelligence you're talking to — if you've never changed the default, you've never seen what this actually does.