Synapse Fusion AI Value Briefing

AI Field Guide · Part 1: Foundations

Deep research: when AI researches for you

Guide 08 of 32For: Everyone5 min read
In one sentenceDeep research modes search many sources and write a cited report; give a clear brief, check the citations, and keep the decision yourself.

On this page:What it isWhy it mattersHow to do itExampleCommon mistakesTry it this weekGoing furtherKey words

What it is

Several AI assistants, such as Claude, ChatGPT, Microsoft Copilot and Gemini, offer a deep research mode. Instead of answering at once, the assistant plans a search, reads many web pages or documents over several minutes or longer, and then writes a report with citations: links or references showing where each point came from.

It is different from a quick question. A quick search answers "what is the opening time?" in seconds. Deep research tackles "what are the main options for replacing our booking system, and what do users say about each?" and takes longer.

Think of asking a junior researcher to spend an afternoon in the library. They come back with a neat summary and sources, but you still check the books and make the decision yourself.

Why it matters

  • It can save hours of opening tabs and reading pages for a first overview.
  • Citations let you trace each claim back to its source.
  • Sources can be weak, out of date, biased or simply misquoted by the assistant.
  • The report reads confidently, which can hide gaps in the evidence.
  • The decision, and the responsibility for it, stays with a person.

How to do it

  1. Decide whether you need it. For a single fact, use a quick search (see guide 02). Use deep research for questions that need many sources compared.
  2. Write a brief with four parts: the question, the scope (time period, region, topic limits), the sources to prefer or avoid, and the output format.
  3. Check the plan if the assistant shows one before it starts, and correct it if it misunderstood. Some ask clarifying questions instead; answer them carefully.
  4. Read the report for the overall picture, then mark the claims that matter most to your decision.
  5. Open the citations for those key claims and confirm the source actually says what the report says. Where possible, check the primary source, not someone's summary of it.
  6. Look at the date and type of each key source: official, independent, a supplier's own marketing, or an anonymous post.
  7. Decide yourself, and record which sources you relied on.

Example

Before:

Research project management software.

Result: a long, general report about well-known tools, mixing old reviews with marketing pages.

After:

Question: which project management tools suit a team of
[number] people who mainly need task lists, shared calendars
and simple reporting?

Scope: tools available in [country/region], information from
the last two years only.

Sources: prefer official product pages, independent reviews
and user forums. Avoid sponsored comparison sites.

Output: a table of up to five options with key features, pricing
model (as stated on the official site) and main complaints from
users. Then a short summary of trade-offs. Cite every claim.
Say clearly where information was missing or conflicting.

You then open the citations for prices and key features, find that one price was from an old page, and correct it. The report becomes input to a team discussion, not the decision.

Common mistakes

  • Using deep research for a simple fact. Fix: use a quick search; it is faster and easier to check.
  • A vague brief. Fix: state the question, scope, sources and format.
  • Trusting a citation because it is there. Fix: open it and confirm it supports the claim; links can be broken or say something different.
  • Ignoring dates. Fix: ask for publication dates, and treat older sources with care.
  • Including confidential details in the brief. Fix: describe your situation in general terms and use only approved tools.

Try it this week

  1. Pick a question you genuinely need answered that involves comparing options, using no confidential details (2 minutes).
  2. Write a brief with the four parts: question, scope, sources, format (5 minutes).
  3. Run it in an approved assistant with a deep research mode, and do something else while it works.
  4. Choose the three claims that matter most and check each citation (8 minutes). Note how many held up fully.

Going further

  • Ask for a short section called "evidence quality" that rates how strong the sources are for each main point.
  • Ask for counter-arguments or sources that disagree with the conclusion.
  • Deep research often has its own, smaller usage allowance, and some sources sit behind paywalls the assistant cannot read.
  • Some assistants can include your own approved documents alongside web sources; say which should take priority.
  • Save good briefs as templates for repeat research, such as supplier checks or market updates (see guide 11).
  • Use the checking routine (see guide 09) before any research finding goes into a report or decision.

Key words

  • Deep research: an assistant mode that searches many sources over several minutes or longer and writes a cited report.
  • Citation: a link or reference showing where a claim came from.
  • Brief: the instructions you give before research starts: question, scope, sources and format.
  • Scope: the limits of the research, such as time period, region or topic.
  • Primary source: the original origin of information, such as an official page or the original study.

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (244 KB), or the complete AI Field Guide (1.7 MB).

Want help putting this to work?

These guides are free to use and share. If you would like your teams to learn this with real work from your organisation, and see the hours it gives back, start with a conversation.

Book my AI Value BriefingA conversation, not a pitch. We will not ask for confidential data. How we handle your details.

Your move

You’re already paying for AI. Let’s make it pay you back.

Start with a conversation. Tell us what concerns you about AI in your organisation. We will listen, answer with practical solutions and propose a path forward that fits. Or write to jacques@synapse-photonics.com.

Supported by INSEAD LaunchPadINSEAD LaunchPad
Accelerator at Station F, Paris · selected 2026
AktantisAktantis
Member
Synapse Photonics SAS 535 Route des Lucioles · 06560 Valbonne SIREN 106 579 493 R.C.S. Grasse Prices exclude VAT