ChatGPT gives you a few dials. The good news: the defaults are sensible, so you can ignore all of this and still do good work. But knowing what the dials do turns a slow, so-so answer into a fast, sharp one. There are three worth learning: how hard ChatGPT thinks, whether it searches the web, and when to hand it a full research job.
Models and thinking levels
Behind ChatGPT sits a model, the engine that writes the answers. OpenAI renames and replaces models often (at the time of writing, the current one is called GPT-6), so do not memorise names. Learn the dial instead.
That dial is the thinking level, set from the model menu in the message box. It runs from fast to thorough:
Instant answers straight away. It is the default for good reason: drafting, summarising, rewriting, and everyday questions. If you are not sure, stay here.
Medium and High make ChatGPT reason step by step before it answers. Slower, but much better on problems with real teeth: a knotty calculation, a plan with many moving parts, weighing a hard decision. Think of a colleague who says "give me a minute" and then gets it right.
Extra High and Pro are for the hardest jobs, the kind where you would happily wait a few minutes for a careful answer. On Business you have all of these; on Plus, Instant, Medium, and High.
Higher levels take longer and use more of your allowance, so cranking everything to the top is the most common beginner mistake. It just makes you wait. You can switch level in the middle of a chat; it applies to the next reply.
One Business detail: a setting called Higher intelligence (in Settings, General) lets ChatGPT add extra thinking on its own when you are on Instant and a question looks hard. Leave it on for convenience, or switch it off if you want Instant to stay instant.
Web search: for anything current
A model learns from data up to a cut-off date. It does not know this morning's news, this week's prices, or a policy that changed yesterday. For that, ChatGPT uses web search.
Often it searches by itself when a question needs current facts. You can also ask for it: pick Search from the tools behind the + button, type / and choose Search, or simply say "search the web for...". The answer comes back with citations, small links you can open, and a Sources list.
One rule keeps you safe: trust, but verify. OpenAI's own help pages warn that results can be incomplete, outdated, or wrong. On anything that matters, open the source and check the date. Let ChatGPT do the finding; you own the fact.
What does checking look like? Say ChatGPT tells you a supplier raised its prices by 8% this month. Open the citation. Is it the supplier's own announcement, or a blog repeating a rumour? Is it from this month, or from last year? Does the page actually say 8%? Thirty seconds, and you know whether the sentence can go into your report. If the source is weak, ask ChatGPT to search again and name the source you trust.
Deep research: when you need a report
For a bigger question, the kind where you would normally spend an afternoon with twenty browser tabs, there is deep research. You describe what you want to know, ChatGPT proposes a research plan you can edit, and then it works through many sources and writes a structured report with citations, which you can download as a Word, PDF, or Markdown file.
It takes minutes, not seconds, and each plan gets a set number of deep research tasks a month, so save it for questions that deserve a report: a market scan, a comparison of suppliers, a summary of new regulation. On Business it can also read from connected apps such as SharePoint or Google Drive, read-only, within your own access rights.
Matching the setting to the task
Put it together and a simple habit falls out. Stay on Instant for most things. When a problem is genuinely hard and self-contained, raise the thinking level. When you need something current, search, and check the sources. When you need a proper written answer from many sources, run deep research. They combine, too: a thinking level and web search work fine together.
Try it
Take one real question with a clear right answer, a bit of logic or a calculation from your work. Ask it on Instant, then ask it again on High. Compare: the thinking answer usually shows its working and catches a detail the quick one missed.
Now ask something about this week, a price or a recent change in your sector, and make sure Search is used. Open two of the citations and check them. Two short tests, and the dials stop being a mystery.
(General exercise. Role and industry versions come once accounts are in.)