Phase 4 · Gemini · Level 2 · Practitioner
Deep Research: when to send Gemini digging
By the end, you'll be able to…
- Recognise the plan-approval step and edit the research plan before Gemini runs it
- Judge when a Deep Research report is worth the wait and when a normal prompt is faster
- Check the sources in a finished report instead of trusting the polished write-up
Why it matters
Deep Research is Gemini working on a slower, bigger job: browsing many web pages, then writing you a structured, cited report. It's useful for the 'I need to get across a topic I don't know' task, but it's slow, and a long report with footnotes can feel more trustworthy than it is. This lesson shows the approval step that lets you steer it, when it earns its wait, and how to check what it hands back.
What Deep Research does differently
Most Gemini answers come back in seconds from a single prompt. Deep Research is a different mode: you give it a topic, it browses many web pages over several minutes, and then it writes a structured, multi-section report with citations. It's the difference between asking a colleague a quick question and asking them to spend an afternoon and come back with a briefing.
The important honest point on availability: Deep Research is available to try on the free tier, with monthly limits, and those limits are more generous on the paid Google AI plans. During busy periods, people without a paid plan may hit the limit sooner. So you can absolutely learn it for free. Just don't be surprised if a heavy week runs you out; the current allowances are on Google's Gemini Apps limits and upgrades page. It's one of the more capable features you can use without paying, which is why it's worth learning properly.
Because Deep Research reaches out to the live web, treat everything it reads as grounding material of unknown quality. Some sources will be excellent, some will be marketing or plain wrong. That's the theme of this lesson: the tool does the digging, but you stay the editor.
The plan-approval step: your chance to steer
Here's the feature that sets Deep Research apart and that most people skip past. When you give it a topic, Gemini first shows you a research plan (a numbered list of the questions and angles it intends to investigate) before it does any browsing. You can read that plan, edit it, add or remove points, and only then approve it.
This step is the single biggest lever you have on the quality of the final report. If the plan is aimed at the wrong questions, no amount of polish in the write-up will save it. Say Tom wants to understand the market Fernway is selling into. His first topic is broad.
Research the current UK market for small-business subscription software: who the main players are, typical pricing models, and what small firms say they value most. I want a briefing I can use to sharpen how Fernway positions its Starter and Pro plans.
Why this works: A specific topic with a stated audience and purpose gives Deep Research a clear brief, so the plan it proposes is aimed at what you actually need rather than a generic overview you'll have to redirect.
Gemini comes back with a plan, perhaps "1. Identify major providers, 2. Compare pricing tiers, 3. Summarise buyer priorities, 4. Note recent trends". Now Tom edits the plan instead of accepting it blindly.
Good plan, but change two things: focus only on the UK, and add a section specifically comparing how providers justify a mid-tier "Pro"-style plan over a cheaper starter tier. Drop the general trends section. I only need what's useful for positioning.
Why this works: Steering the plan is far cheaper than fixing a finished report. Adding your real angle (UK focus, a specific comparison) up front means the whole research effort points where you need it, not where the default aimed.
Approve the edited plan and Gemini goes to work. A few minutes later you get a report with headings, a synthesis, and a list of the sources it drew on. Editing the plan is what turned a generic market overview into a briefing aimed at Tom's actual decision.
When it's worth it, and when it isn't
Deep Research is a heavy tool. Use it when the job justifies the wait:
- You're getting across an unfamiliar topic (a market, a regulation, a set of competitors) and want a structured starting point rather than ten open browser tabs.
- The task really spans many sources and a single quick answer would be too shallow or too likely to miss things.
- You'll act on it. A report you'll actually read and use, not idle curiosity that burns one of your monthly runs.
Reach for a normal prompt instead when you want a quick fact, a definition, or a draft; when you already know the answer and just need it written up; or when speed matters more than breadth. Firing Deep Research at "what's a good subject line?" wastes minutes and a monthly allowance on a job a one-line prompt does better. The skill is matching the tool to the size of the question.
Verifying the sources: the part that matters most
A Deep Research report is the most convincing-looking thing Gemini produces: sections, citations, a confident synthesis. That polish is exactly why it's dangerous to take on trust. The report is a summary of web pages of unknown quality, stitched together by a model that can still misread or overstate, and a well-formatted hallucination is still a hallucination. So before you rely on it:
- Open the actual sources. The citations are links, so follow the ones behind any claim you'll act on. Check the figure or statement is really there and really says what the report claims.
- Judge each source. A vendor's own marketing page and an independent review are not equal evidence. Deep Research doesn't grade credibility for you; you do.
- Watch for confident synthesis over thin evidence. If a strong conclusion rests on one weak source, treat it as a lead to check, not a fact.
- Note what's missing. Ask the report what it couldn't find or where sources disagreed. The gaps are often where the real risk sits.
For this report, list the three claims that rest on the weakest or fewest sources, and any points where your sources disagreed with each other. For each, give me the link so I can check it myself.
Why this works: Asking the report to expose its weakest evidence and its disagreements flips it from a polished conclusion into a checkable draft, so you find the soft spots deliberately instead of hoping there aren't any.
That single follow-up turns a report you'd have skimmed and trusted into one you can actually stand behind.
Try it now
Common mistakes
- Skipping the plan-approval step. Rubber-stamping the default plan is the most common way to get a report aimed at the wrong questions. Read it and edit it; it's your cheapest chance to steer.
- Using it for quick jobs. Deep Research is slow and rationed. For a fact, a definition or a draft, a normal prompt is faster and doesn't spend your monthly allowance.
- Trusting the format. Headings and citations make a report look authoritative. Over-trusting that polish is the classic Deep Research error, because a fluent, well-cited report can still rest on weak or misread sources.
- Not opening the citations. A link you never click is not a check. The claims you'll act on are exactly the ones to trace back to the source.
- Treating all sources as equal. Deep Research doesn't rank credibility. A marketing page and an independent study can sit side by side in the report, and telling them apart is your job.
Keeping current
Deep Research's availability, limits and the exact look of the plan step change as Google develops it. Trust your screen over this lesson and check Google's official pages: Use Deep Research in Gemini Apps and Gemini Apps limits and upgrades. Accurate as of 13 July 2026.