Skip to content

Prompt Engineering Foundations · Modern assistant features: modes, memory, research and media · lesson 7 of 16 · 12 min

Quick answers, thinking modes and deep research

Three ways an assistant can answer

Modern assistants give you a choice of how much effort they put into an answer. The names differ between products and change often, but there are three broad modes:

| Mode | What it does | Typical wait | Best for | |---|---|---|---| | Quick answer | Replies straight away from what it already knows and what you gave it | Seconds | Drafting, rewriting, summarising, simple questions | | Thinking (reasoning) | Works through the problem privately before answering: breaks it down, tries approaches, checks its work | Seconds to minutes | Maths, multi-step plans, tricky decisions, logic, code | | Research (web search or deep research) | Searches the web, reads many pages and writes a sourced report | Seconds for a search; several minutes for deep research | Current facts, comparisons, market scans, anything that needs sources |

Some assistants pick the mode automatically depending on your question; others give you a switch, a menu or a model picker labelled with words like "Thinking", "Reasoning", "Extended thinking", "Pro" or "Deep research". Check your assistant's interface and help pages, because these labels change frequently.

Why thinking modes help on hard problems

A quick answer is like asking a colleague to reply on the spot. A thinking mode is like saying "take a few minutes and work it out properly". The model spends extra computation reasoning before it writes the answer. For problems with several steps, this usually reduces mistakes, because the model can catch and fix errors along the way.

It is not magic. Thinking modes still make mistakes, cost more on paid plans or use up usage limits faster, and are slower. On simple tasks they rarely add much.

A decision guide

Ask yourself three questions:

  1. Does the answer depend on current information? Prices, opening hours, laws, news, product specs. If yes, use web search or deep research, and check the sources.
  2. Does the task have several steps or constraints that must all be satisfied? A schedule, a budget, a comparison with a recommendation, a calculation. If yes, use a thinking mode.
  3. Is it mainly about wording? An email, a caption, a rewrite, a summary of text you pasted. Use the quick mode.

Worked example: one business owner, three modes

Hina runs an online modest-fashion shop from Lahore that ships to the UAE and UK.

Quick mode: "Rewrite this product description to sound warmer, under 60 words." Done in seconds.

Thinking mode:

I sell 3 abaya styles. Costs and prices are below. Shipping to the UAE
costs 18 USD per parcel and to the UK 22 USD. A parcel holds up to 3
items. Work out my profit per order for 1, 2 and 3 items to each
country, then tell me the minimum order value for free shipping that
keeps at least 25% margin. Show the calculation as a table and list any
assumptions.
[costs and prices]

The thinking mode is more likely to keep all the constraints straight. Hina still copies the table into a spreadsheet and checks the formulas.

Deep research:

Research the current rules and typical costs for a small Pakistani
online seller shipping clothing to customers in the UAE and UK: customs
duty thresholds, VAT, and documentation needed. Use official government
and courier sources where possible, cite each claim, and flag anything
that changed in the last 12 months.

She then opens the official sources cited, because customs rules change and a report can misread them.

How to prompt a thinking mode

Thinking modes need the same good briefing as any prompt, with three extra tips:

  • Give all the information up front. The model can reason, but it cannot invent facts you did not supply.
  • Say what "done" looks like. "A table with these columns and a one-line recommendation."
  • Don't micromanage the steps. You do not need to write "think step by step"; the mode already does that. Describe the goal and constraints clearly instead.

How to prompt deep research

Deep research tools plan a search, read many sources and produce a long report. Get more from them by stating:

  • The decision the research supports ("so I can choose a courier").
  • Scope: countries, time period, types of source to prefer or avoid.
  • Output: length, sections, a comparison table, a list of sources.
  • What not to do: "Don't include general advice; focus on rules and costs."

Many deep research tools ask clarifying questions before they start. Answer them properly; it is the best moment to shape the report.

Hands-on: the same question in three modes

Pick one real question from your work that involves a few steps (a budget, a comparison or a plan). Ask it three ways:

Quick mode:     [your question]
Thinking mode:  [same question] + "Check your answer against every
                constraint and list any assumptions."
Research mode:  [same question] + "Use current web sources and cite them."

Compare the three answers on accuracy, usefulness, time taken and how easy each was to check.

Pitfalls

  • Using thinking mode for everything and hitting usage limits by mid-week.
  • Treating a long, confident research report as verified truth. Open the sources.
  • Forgetting that a quick mode without web access may not know recent events.
  • Assuming the "best" mode is always the newest or most expensive one.

How to know you are choosing well

You are choosing well when simple tasks come back in seconds, hard tasks come back correct more often, and every fact you rely on has a source you have opened.

Video lecture: Quick answers, thinking modes and deep research

Quick answers, thinking modes and deep research — Three ways an assistant can answer Modern assistants give you a choice of how much effort they put into an answer. The names differ between products and…

11 chapters · about 7 minutes · captions and full transcript below.

  1. Quick answers, thinking modes and deep research
  2. Three modes
  3. Labels change, principles don't
  4. The decision guide
  5. Worked example: one shop, three modes
  6. Prompting the modes
  7. Example 1: a dinner for eight
  8. Example 2: a free-delivery threshold (illustrative)
  9. Four traps
  10. Practice: pick the mode
  11. Recap

Lecture transcript

Quick answers, thinking modes and deep research

Your AI assistant probably has more than one way to answer you, and choosing the right one can be the difference between a wrong budget and a right one. In this lecture you will learn the three broad modes, quick answers, thinking, and research, when each one pays off, how to prompt each one, and the traps to avoid. By the end, you will know exactly which switch to flip for any task.

Three modes

Mode one is the quick answer. The assistant replies straight away from what it knows and what you gave it. Perfect for drafting, rewriting, summarising pasted text and simple questions. Mode two is thinking, sometimes called reasoning or extended thinking. The assistant works through the problem privately before answering, breaking it down, trying approaches and checking its work. Slower, but better for maths, multi-step plans, logic and tricky decisions. Mode three is research. The assistant searches the web and cites sources, and with deep research it reads many pages over several minutes and writes a structured report.

Labels change, principles don't

The labels differ between products and they change often. You might see a model picker, a switch that says thinking, a menu item called deep research, or an assistant that chooses automatically. So do not memorise buttons. Memorise the question each mode answers. Quick: can you write this for me now? Thinking: can you work this out carefully? Research: what do current sources say?

The decision guide

Here is a simple decision guide with three questions. One: does the answer depend on current information, like prices, laws, opening hours or news? Then use web search or deep research, and check the sources. Two: does the task have several steps or constraints that must all be satisfied, like a schedule, a budget or a comparison with a recommendation? Then use a thinking mode. Three: is it mainly about wording? Then the quick mode is fastest and usually just as good.

Worked example: one shop, three modes

Meet Hina, who runs an online modest-fashion shop from Lahore, shipping to the UAE and the UK. To make a product description warmer, she uses quick mode, done in seconds. To work out profit per order for one, two and three items to each country, including shipping costs and a minimum margin, she uses thinking mode, and asks for a table with assumptions listed. She still pastes the table into a spreadsheet to check the formulas. To learn current customs and VAT rules for parcels, she uses deep research, asks for official sources, and opens them, because rules change and reports can misread them.

Prompting the modes

How do you prompt a thinking mode? Give it all the information up front, because it can reason but it cannot invent facts you did not supply. Say what done looks like, such as a table with these columns and a one-line recommendation. And do not micromanage the steps. You do not need to write think step by step; the mode already does that. For deep research, state the decision the research supports, the scope, the type of sources to prefer, and the output you want. Many research tools ask clarifying questions first. Answer them properly; that is your best moment to steer.

Example 1: a dinner for eight

Let us make the choice concrete with a simple example. You are planning a dinner for eight friends, two vegetarians and one with a nut allergy, on a budget of eighty pounds. Quick mode can suggest dishes instantly. But if you ask for a full menu, a shopping list with quantities, and a total cost that stays under budget, you now have constraints that must all hold at once. That is when a thinking mode earns its extra seconds. It is more likely to keep the allergy, the vegetarians and the budget straight. And if you ask, which supermarket has the best price on salmon this week, that needs current information, so switch on web search.

Example 2: a free-delivery threshold (illustrative)

Now a business scenario, with illustrative numbers. Zara runs an online accessories store from Karachi and sells about three hundred orders a month. She wants to decide whether to offer free delivery above a threshold. Quick mode can draft the announcement later. The decision needs thinking mode. She pastes her average order value, delivery cost per order, and margin, and asks: model three thresholds, two thousand, three thousand and four thousand rupees, estimate the margin impact for each, and list your assumptions. The thinking mode produces a clear table. She checks it in a spreadsheet and finds one rounding difference. Then she uses deep research to see what competitors' published delivery policies look like, with sources she opens. Three modes, one decision, each used for what it does best.

Four traps

Watch for four traps. Using thinking mode for everything and hitting your usage limits by Wednesday. Treating a long, confident research report as verified truth; open the sources. Forgetting that a quick mode without web access may not know recent events. And assuming the newest or most expensive mode is always best. Often it is not; the right mode is the one that matches the job.

Practice: pick the mode

Here is a quick way to practise choosing. For each of these, decide the mode. One: shorten this email to eighty words. Quick mode. Two: which of three delivery options is cheapest for a hundred and twenty parcels a month, given different weight bands and a fuel surcharge? Thinking mode, with all the rates pasted in. Three: what are the current requirements for registering a small online business in Saudi Arabia? Research mode, asking for official sources, and then you open them. After a week of making this choice deliberately, it becomes automatic.

Recap

To recap. Quick mode for wording. Thinking mode for multi-step problems and calculations. Research for current facts with sources you open. Give thinking modes complete information and a clear definition of done, and give research tools a goal, scope and source preferences. Try this now: take one real multi-step question from your work and ask it in all three modes. Compare accuracy, time, and how easy each answer was to check. Next, we will stop you repeating yourself, with projects, custom instructions and memory.

Key takeaways

  • Assistants offer quick, thinking (reasoning) and research modes; names differ by product and change often.
  • Use thinking modes for multi-step problems, calculations and plans; quick mode for wording tasks.
  • Use web search or deep research for current facts, and open the sources.
  • Give thinking modes all the facts and a clear definition of done instead of micromanaging steps.

Try it

Choose one real multi-step question from your work. Ask it in quick mode, thinking mode and research mode, and note which answer was most accurate, fastest and easiest to check.