Deep Research That's Actually Right: ChatGPT vs Gemini vs Claude vs Grok vs Perplexity (Step-by-Step)

Turn a big question (which car, which school, which supplier, is this medical bill right?) into a one-page brief you can trust, with a 4-step method that works in any chatbot.

Checked October 17, 2026. Researched and drafted with help from Grok Bot.

Quick answer

• "Deep research" is the AI doing homework for you. It searches dozens of web pages over several minutes and writes a long report with links.
• It's not automatically right. Long reports look trustworthy. The method below makes them actually trustworthy.
• The 4 steps: 1) scope the question, 2) edit the research plan before it runs, 3) run it in two chatbots, 4) cross-check the links and squeeze it into one page.
• Best picks: Gemini Deep Research if you want to include your Gmail and Drive. ChatGPT deep research for control over which sites it uses. Claude Research if you're on a paid Claude plan and want it to use your connected apps. Perplexity for quick, heavily linked answers. Grok Expert (or Heavy on Ultra) for hard reasoning with live web search.
• Free plans: Gemini and ChatGPT both offer some deep research with limits. Claude Research is paid only.

Comparison table

Chatbot	What it's called	Where to start it	Edit the plan first?	Your own files and email	Plan
ChatGPT	Deep research	Type /Deepresearch, or + → Deep research, or the sidebar	✓	Connected apps (read-only)	Limits vary by plan; resets 30 days after first use
Gemini	Deep Research	Add Files (+) → Deep Research	✓ Edit plan	✓ Sources: Gmail, Drive, files	18+; free with limits; visuals on Ultra
Claude	Research	+ → Research (web search must be on)	Not documented	✓ connectors on web and desktop (not mobile)	Paid plans only
Grok	Expert or Heavy mode with web search	Mode picker	✗	Connectors (Gmail, Drive and more)	Expert on plans we checked; Heavy needs Ultra
Perplexity	Research mode	Mode picker	✗	Files in Projects	Free with limits
Microsoft Copilot	Researcher (Microsoft 365)	Copilot in Microsoft 365	Not checked	Your Microsoft 365 files	Microsoft 365 Premium; consumer Deep Research retired Aug 18, 2026

Grok note: SpaceXSI calls Grok's research agent DeepSearch. We couldn't confirm on a current official page how it appears in today's app (not verified), so this guide uses Expert or Heavy with web search.

Perplexity row: details not verified this round (Perplexity's help center blocked our checks).

Which one should I use? The verdict

Quick questions

Is deep research better than a normal chat? For big questions with many sources, yes. For a quick fact, no. It takes minutes and uses up a limited number of runs.

Why did ChatGPT say I hit my deep research limit? Limits vary by plan, and the counter resets 30 days after your first use. Save runs for real decisions.

Can it read my email? Gemini can use Gmail and Drive as sources if you choose them. ChatGPT can use connected apps (read-only). Claude can use connectors on web and desktop, not on mobile.

Can I trust the links? Mostly, but check. Reports sometimes cite a page that doesn't say exactly what the report claims. Step 4 shows you how to spot it fast.

Does Grok have deep research? Yes: Grok's research agent is called DeepSearch. Where it sits in today's app isn't confirmed on an official page (not verified). Grok's Expert mode also thinks harder and searches the web, and Heavy (Ultra plan) uses a "team of experts."

Before you start

• Pick a question that deserves it. A decision with money, health or time at stake. "Which used SUV under $25,000 is most reliable for a family of five in Minnesota?" is a good one. "What's the capital of Peru?" is not.
• Know your limits. Each run counts against your plan. Do Step 1 in a normal chat first, so you don't waste a run on a vague question.
• Have your own facts ready. Budget, location, deadlines, must-haves.
• Health, legal and money: deep research is a starting point. Bring the brief to a professional.

Step 1: Scope the question (in a normal chat)

Paste this into any chatbot, in normal mode:

I'm going to run a deep research report. Before I do, interview me.
My rough question: [your question]
Ask me up to 6 short questions about my situation, budget, location, deadline and what "good" looks like.
Then write the final research brief:
- the exact question
- what to include and exclude
- the kinds of sources to prefer (official sites, recent reviews, government data) and to avoid
- the time range (e.g. last 2 years)
- the output: a comparison table + top 3 recommendations + "what I couldn't confirm"

Copy the final brief. That's your research prompt.

Step 2: Edit the plan before it runs

ChatGPT deep research

1. Type /Deepresearch, or choose + → Deep research (also in the sidebar).
2. Optional: Sites → Manage sites to restrict research to sites you trust, or prioritize them.
3. Optional: turn on connected apps you want it to read (they're read-only).
4. Paste your brief and send.
5. ChatGPT shows a research plan. Edit it: remove off-topic parts, add what's missing.
6. While it runs, you can interrupt to add a detail or change direction.
7. When it's done, download as Markdown, Word or PDF.

Gemini Deep Research

1. Go to gemini.google.com (18+).
2. Below the text box, choose Add Files (+) → Deep Research.
3. Choose Sources: web, plus Gmail, Drive or files if they help (for example, a quote or a school brochure).
4. Paste your brief and send.
5. Gemini shows a plan. Choose Edit plan to change it, then Start research.
6. It takes about 5–10 minutes. You can leave and come back.
7. The report opens in a canvas panel: Share & export → Export to Docs, or Create Audio Overview to listen.

On the Ultra plan, reports can include visuals, except when you use Workspace sources.

Claude Research (paid plans)

1. Make sure web search is on.
2. Choose + → Research.
3. On web or desktop, connected apps (connectors) can be used too. On mobile, connectors are skipped without a warning, so run research with your own files on a computer.
4. Paste your brief and send. Claude searches for several minutes and writes a cited report.

Whether you can edit Claude's research plan before it starts is not documented on Claude's help pages; not verified. Also note: Research uses your connectors automatically, without asking for approval each time, so only turn on the ones you want it to read.

Grok (Expert or Heavy)

1. Open the mode picker and choose Expert ("Thinks hard"). On the Ultra plan you can choose Heavy ("Team of Experts").
2. Paste your brief and add: Search the web. Link every claim. Plan your research first and show me the plan before you write the report.
3. Check the plan Grok writes, correct it, then say Go ahead.
4. Optional: use @ connectors (Gmail, Drive and more) if your own files matter.

Perplexity Research

1. Choose Research mode in the search box.
2. Paste your brief. Perplexity answers with numbered sources.

Step 3: Run it in two chatbots

Run the same brief in a second chatbot. Good pairs:

• Gemini + ChatGPT (both free with limits).
• Claude + Grok Expert (if you pay for one).
• Any one + Perplexity (quick second opinion).

Then paste both reports into one chat (any chatbot, normal mode):

Here are two research reports on the same question.
Report A: [paste]
Report B: [paste]
Make a table of every claim where they disagree or only one mentions it: Claim | Report A | Report B | Which source looks stronger and why.
Then list the 5 claims I should check myself before deciding.

Step 4: Cross-check and squeeze to one page

Check the 5 claims yourself. For each, open the link. Does the page actually say it? Is it current?

Then ask for the brief:

Using only claims I've confirmed below, write a one-page decision brief.
Confirmed: [paste the claims you checked, with links]
Format:
1. My question (1 line)
2. The answer in 3 sentences
3. Comparison table (max 5 rows)
4. Top pick and runner-up, with one reason each
5. What I couldn't confirm
6. What to ask a professional (if health, legal or money)

The citation check (paste after any report):

For the 10 most important claims in your report, quote the exact sentence from the source that supports it, with the link. If a source doesn't directly support a claim, say "NOT SUPPORTED" and remove the claim.

The others

Try it: 3 worked examples

1. A used car

Rough question: which used 3-row SUV should I buy?
Facts: budget $25,000 out the door, Minnesota winters, family of 5, 15,000 miles a year, keep it 6 years.

Prefer: owner-reliability surveys, government recall and safety data, recent long-term reviews. Avoid: dealer pages. Output: a table of 5 models with reliability, safety, cost per year and known issues.

2. A school

Rough question: which of these 3 middle schools fits my kid best? [names]
Facts: [your kid's interests, commute limit, special needs]

Add Drive as a source in Gemini if you have school brochures or emails.

3. A medical bill

Rough question: is this hospital bill correct, and what can I dispute? [describe the bill; remove your name, ID and policy numbers]

Prefer: your state's official consumer pages, your insurer's published rules and federal patient-rights pages. Bring the brief to your insurer's billing line.

Common mistakes and fixes

Mistake	Why it hurts	Fix
Vague question	You get a long, generic report	Step 1 interview first
Skipping the plan	It researches the wrong thing for 10 minutes	Always edit the plan (ChatGPT, Gemini)
Trusting one report	Every bot has blind spots	Run two, compare
Not opening links	Some citations don't say what the report claims	Citation check prompt
Using it for quick facts	Wastes your limited runs	Normal chat for quick facts
Claude Research on mobile with connectors	Connectors are silently skipped	Use web or desktop

Prompt pack

The scoping interview (Step 1, above).

The plan editor (paste when a plan appears):

Before you start: remove anything not about [my question]. Add [missing angle]. Limit sources to the last [2] years. Prefer [official sites / independent reviews]. Keep the final report under 1,500 words with a table.

The skeptic:

Argue against your top recommendation. What would make it the wrong choice for me? What evidence would change your mind?

The "what's changed" update:

Here is a report from [date]. Search only for what's new since then and tell me which conclusions still hold.

Next level

Take a Gemini Deep Research report further: drop it and its sources into a Gemini Notebook, ask questions with citations, and turn it into a podcast-style Audio Overview. See A7: The Google Research Stack.