Why the setting matters as much as the tool
Most people pick an AI tool once and never touch the settings again. That works until the answer comes back shallow on a hard problem, or a simple question takes a minute to answer.
Every major AI assistant now gives you two choices before you type: which model to use, and how hard it should think. ChatGPT, Microsoft Copilot, Google Gemini and Claude all offer some version of both. Getting those two choices right does more for the quality of your answers than switching between vendors.
This article is a plain-English guide to making that choice. It isn't about which vendor is best. If you're still deciding which tool to start with, Cameron covered that in ChatGPT, Claude or Copilot: which AI should your business start with?
Two dials: the model and the effort
The simplest way to think about it is staffing a job. The model is who you give the job to. The effort is how long you let them spend on it.
The model sets the ceiling on what the AI can do. Every vendor now sells a family of models in roughly three sizes:
- Fast and light. Quick, cheap and good at routine work. Think of a capable junior who turns things around in seconds.
- Balanced. The everyday workhorse for most business tasks. This is usually the default.
- Most capable. Slower and more expensive, built for complex analysis, long documents and hard problems. Your senior specialist.
The effort (also called thinking or reasoning) controls how much work the model does before it answers. At low effort it replies straight away. At high effort it plans, checks its own working and considers alternatives first, which can take from several seconds to a few minutes.
The two dials are separate. A big model at low effort can still give a quick answer. A smaller model at high effort can work carefully through a problem that's within its reach. You get the best result when both match the job.
A simple guide: match the dials to the job
Start on the default or Auto setting, and only turn a dial up when the answer tells you to. Most everyday work never needs the top settings.
| The job | Model | Effort |
|---|---|---|
| Rewording an email, fixing tone, a quick factual question | Fast or default | Lowest (Instant, Quick response, Low) |
| Summarising a meeting, email thread or short document | Default | Low to medium, or Auto |
| Drafting a proposal, policy or report from your notes | Default | Medium |
| Building spreadsheet formulas or working through data | Default | Medium to high |
| Comparing quotes or contracts, or finding gaps in a long document | Default or most capable | High (Think deeper) |
| Troubleshooting something unusual, planning a strategy, multi-step analysis | Most capable | High |
| Deep research where waiting several minutes is fine | Most capable | Highest (Extra High, Pro, Deep Think) |
A few rules of thumb I use myself:
- Turn effort up before you change tools. If an answer skips steps or misses detail, re-run the same prompt at a higher effort. That fixes more than jumping to a different product.
- Turn it down when you're waiting on easy work. If a simple question takes a long time, you're paying in time (and often usage limits) for thinking you don't need.
- Effort doesn't replace context. A clear prompt with the right document attached beats the highest setting on a vague question.
- More thinking isn't a fact check. A higher setting reasons more carefully, but it can still be wrong. Check anything that matters before it goes to a client.
What it means for cost, speed and your data
Bigger models and higher effort cost more, either in usage limits or in dollars. On subscription plans, heavy use of the top settings burns through your allowance faster. Keeping simple tasks on lower effort stretches it further, whichever vendor you use.
If you're building AI into a process through an API, the dials become a line item. Larger models cost more per word processed, and higher effort produces more words behind the scenes. The sensible approach with any vendor is to prove the workflow on a capable model, then find the smallest model and lowest effort that still does the job.
The dials don't change where your data goes. Your plan and your tenant settings decide that, not the model menu. A business or enterprise plan with the right data protections matters far more than which model a staff member picks. We covered that side in Private, secure AI for Australian businesses.
Admin control also varies. Microsoft doesn't give administrators controls over Copilot's mode selector, so staff choose for themselves. Some other business plans let an administrator limit which models and effort levels a role can use. Either way, a one-page guide for staff on when to use each setting will do more than a technical lock.
The short version
Use the default for everyday work. Turn effort up when the task is hard or the answer falls short. Reach for the biggest model only when the problem genuinely needs it. That habit works the same in ChatGPT, Copilot, Gemini or Claude, and it'll keep working as the menus change.
If you'd like help putting together a simple staff guide for the AI tools your business already uses, or working out whether those tools are set up safely, our AI Readiness assessment is a good place to start. Or book a discovery call and we'll talk it through.