Agents · Lesson 02 Pro ~11 min read Citations + MCP

Deep Research: depth over breadth.

Deep Research is ChatGPT's long-running research mode. Where ChatGPT Agent acts (clicks, fills forms), Deep Research thinks — pulling from dozens of sources, synthesizing, and producing a cited report that reads like an analyst's work. As of February 2026, it can also connect to MCP servers and restrict searches to trusted sites.

Workflow 01 Industry analysis with sources

1

Get a real research report on a topic

The classic Deep Research use case. Give it a research question, get back a fully-cited report in 10-30 minutes.

The prompt that works

Industry researchProduce a deep research report on: 'The state of AI training data licensing as of mid-2026.' Specifically address: • Who's licensing data to AI companies (the big publisher deals) • Pricing patterns (per-token, per-content-item, flat licenses) • Lawsuits in progress and their current status • How AI companies are responding (synthetic data, opt-outs, content provenance) • What's likely to change in the next 12 months Length: 2,500-3,500 words. Citations inline with hyperlinks. End with a 5-bullet 'what to watch' section.

Best use cases

  • Market research for strategy work
  • Competitive landscape analysis
  • Regulatory or legal landscape reports
  • Pre-meeting prep for executive conversations
Deep Research can take 30+ minutes. Don't stare at it — kick it off, do something else, come back to the result.
Time savings: A real research report: 1-2 days analyst time → 30 min Deep Research + your review.

Workflow 02 Trusted-site mode for high-stakes research

2

Restrict Deep Research to sources you trust

For research where source quality matters (medical, legal, financial), restrict Deep Research to a curated set of trusted sites.

The prompt that works

Trusted-source researchResearch question: 'Current best-practice treatment for [condition], evidence level [strong vs emerging].' Restrict sources to: • PubMed.gov • NIH.gov • CDC.gov • Mayo Clinic, Cleveland Clinic, Johns Hopkins • Cochrane reviews • UpToDate (where available) Do NOT include: • Health blogs or WebMD-style sites • Vendor or treatment-clinic sites • Reddit, Quora, forums Produce a report with: standard of care, current evidence quality, recent guideline changes, and what to ask a specialist. Cite every claim. Include disclaimer that this is research synthesis, not medical advice.

Best use cases

  • Medical or healthcare research
  • Legal precedent research
  • Investment research from primary sources
  • Scientific literature review
Trusted-site mode doesn't make the output authoritative — it just constrains the inputs. Deep Research can still misread a source. High-stakes decisions need expert review, not just an agent.
Time savings: Source-quality control: built-in instead of manual filtering.

Workflow 03 MCP-enabled internal research

3

Connect Deep Research to your own tools and data

As of February 2026, Deep Research supports MCP connectors — so it can pull from your Salesforce, your internal docs, your databases, alongside public sources.

The prompt that works

Internal + external researchRun a competitive analysis on our top 3 competitors: Internal sources (via MCP): • Salesforce — lost-deal notes mentioning these competitors over the last 6 months • Notion — our internal competitive intel docs • Slack #competitive-intel channel External sources: • Competitor websites, pricing pages, blog posts • G2 / Capterra reviews from the last 6 months • Recent funding announcements Produce a report: • For each competitor: positioning, pricing, recent moves, observed strengths from our lost deals, weaknesses you can identify • Conclude with: 3 actions our sales team should take based on these findings

Best use cases

  • Internal data + market context combined
  • Customer success deep-dives (CRM notes + public sentiment)
  • Product positioning research
  • Sales enablement content built from real win/loss data
MCP connectors need proper auth and scopes. Use read-only credentials for any sensitive data source.
Time savings: Cross-source research: combine 4-5 tools' worth of analysis in one task.

Run one Deep Research task this week

Pick a topic you'd normally Google for an hour and synthesize in your head. Frame it as a Deep Research prompt with specific subquestions. Set Deep Research running and walk away. The report you come back to is the kind of work an analyst would charge for.

Back to ChatGPT Mastery → Try a prompt in the playground →

What you can do now

  • Frame queries with specific subquestions, not vague topics
  • Use trusted-site mode for high-stakes research
  • Connect MCP servers for hybrid internal + external research
  • Expect 10-30 minute runtimes — don't watch it work
  • Treat output as a strong draft requiring fact-check and review
Pro
Up next in ChatGPT Mastery

Lesson 03 · Custom GPTs — build agents that match your work

Custom GPTs are the user-extension layer of ChatGPT. Build agents that know your context, your tone, your data. Where the GPT Store gets useful. See the track →