Deep research
# Work Users
# Advanced Features
Learn how to use deep research for thorough external research tasks
July 23, 2025 · Last updated on February 12, 2026What is deep research?
Deep research is a research agent in ChatGPT that can scan many sources, synthesize what it finds, and produce a structured report with citations—useful when you need more than a quick answer.
Deep research helps when you’re trying to get oriented in an unfamiliar space, compare options, or build an evidence-backed brief. It’s especially helpful for topics where the relevant information is scattered across many sources.
Deep research supports high-stakes, source-heavy work across consulting/strategy, finance, and legal—where teams need to pull from many inputs, keep analysis aligned as questions change, and avoid rework. It combines web and internal sources (and, where available, authenticated data providers) into a traceable, structured output with clear citations that’s easier to review, validate, and reuse.
What’s new in the deep research app
The updated experience is designed to make it easier to start, monitor, and refine research end-to-end:
- Connect to any app with read access so research can use authenticated or domain-specific sources when available. Only connect sources you’re authorized to access, and follow your organization’s data-handling policies. Your workspace admin also has control over access to deep research.
- You can now connect higher-quality data sources – like approved URLs, custom integrations and trusted third-party apps – including connections to private market data sources.
- The new experience also gives you more control, letting you shape the plan before it starts, follow progress in real time, and adjust direction or sources as the work runs.
- Review + sharing/export options to make outputs easier to reuse. If you share externally, re-check citations, remove sensitive info, and confirm rights/permissions for any included material.
- Improved discovery: a landing experience with starter prompts and quick access to recent reports.
Getting started
- Open a chat and choose Deep research from the tool menu or type /deep research. Write a specific research request: your goal, scope, timeframe, and the output format you want.
- Choose your sources (as needed):
- Add Apps for authenticated or specialized sources (read-only). Reminder that admins of ChatGPT Enterprise control access to apps via role-based access controls.
- Set Sites / allowed domains to restrict web research to specific domains or to prioritize specific domains, while allowing ChatGPT to search the open web
- After you start the chat, watch the live progress and the running plan; interrupt to clarify, narrow scope, or add sources.
- Once generated, review the report: validate key claims, inspect citations, and request alternate versions (e.g., different scope, audience, or depth).
Examples for your role
Role | Use Cases | Prompt |
Marketing | Competitor positioning studies, audience sentiment monitoring, campaign performance benchmarks, trend scanning | Produce a competitor messaging analysis for the top three cloud security platforms launched globally in the last year. Include citations and call out where evidence is weak or conflicting. |
Product | Feature adoption analytics, user interviews and usability testing, feedback clustering, competitive teardown | Find the three most discussed usability issues for our mobile onboarding based on public reviews and forums. Cite sources and propose fixes, clearly separating evidence from hypotheses. |
Finance | Financial statement review, regulatory monitoring, economic indicator tracking, scenario modeling | Create a report on how upcoming global banking regulations may affect capital ratios at major international banks. Cite sources and note any assumptions or areas requiring expert review. |
Sales | Account research, buyer committee mapping, solution comparison, deal retrospective interviews | Research the buying criteria large enterprises use when selecting customer data platforms. Cite sources and summarize the most common patterns and outliers. |
Engineering | Repository analytics, incident post mortems, knowledge base mining, performance benchmarking | Locate examples of memory leak fixes in open-source Rust projects similar to ours. Cite sources and recommend approaches, with clear caveats where details are missing. |
Consulting/strategy | Market scans, competitive analysis, and strategy research using trusted sources | Summarize [company]’s diversification strategy beyond GPUs, including moves into cloud, networking, and software ecosystems. |
HR | Labor market intelligence, policy benchmarking, engagement surveys, diversity metrics research | Compile an overview of new pay transparency rules introduced in major regions for 2025. Cite sources and flag where legal interpretation is required. |
IT | Threat intelligence feeds, vendor evaluation, usage analytics, root cause reference search | Generate a comparison of the top three enterprise password managers by certifications, adoption signals, and cost. Provide sources and note which claims should be verified with vendors. |
How to incorporate deep research outputs
Deep research is designed to produce more than a one-off report. Its real value is how the output can be reused across workflows—supporting understanding, analysis, decision-making, and execution without requiring you to re-research the same topic.
Use case (report topic) | Example prompt | How to incorporate the output |
Market scan | “Run deep research on the current landscape for [category] in [region/industry]. Map the main segments and common capabilities, typical pricing/packaging patterns, key adoption drivers, and unresolved customer pain points. Provide a sourced summary and a short list of hypotheses to validate with customers.” | Turn the report into a 1-page landscape brief (market map, table-stakes vs differentiators, top hypotheses). Create an interview guide + validation backlog (who to talk to, questions, what evidence would confirm/deny). Use it to draft the PRD context and a v1 scope proposal. |
Competitive teardown | “Deep research [Competitor A] vs [Competitor B] vs us/our product] in [domain]. Compare key workflows and features, security/compliance claims (as stated publicly), integrations, and pricing signals. Separate ‘marketing claims’ from ‘documented evidence’ and cite sources for each.” | Convert into a battlecard (when they win/when we win, objections → responses) + feature comparison table with citations. Feed into sales enablement and produce a backlog shortlist for meaningful gaps. |
Policy analysis | “Deep research common policy approaches for [policy area] (e.g., travel, expenses, data handling) in [industry/geo]. Summarize typical definitions, required elements, common exceptions, and rollout practices. Include example language excerpts with citations and note where norms vary by jurisdiction.” | Extract into a redline-ready draft structure (definitions, scope, exceptions) + decision log (choices needed). Generate FAQ and manager talking points. Build an approver checklist for exceptions. |
Regulatory landscape summary | “Deep research the regulatory and standards landscape relevant to [topic] in [jurisdictions]. Identify authoritative sources (regulators/standards bodies), summarize what they cover at a high level, recent changes, and practical implications for a company like [company type]. Provide a ‘questions for counsel’ section.” | Produce an internal guidance note (what likely applies / unsure / links to primary sources). Create counsel review questions and a release gating checklist (disclosures, approvals, recordkeeping). |
Vendor evaluation | “Deep research [Vendor 1] vs [Vendor 2] for [use case]. Compare integration requirements, product capabilities, documented security/compliance posture (public artifacts), SLA/support model, and known implementation risks. Output a scorecard table and a list of validation steps to run in a trial.” | Turn into a vendor scorecard + due diligence checklist (security questions, SLA must-haves, integration requirements). Create a trial validation plan (“what we must test to confirm claims”) and draft an RFP email with targeted questions. |
Customer/user research synthesis | “Deep research the most common user workflows and pain points for [job-to-be-done] in [industry/persona] using public sources (forums, case studies, help docs, reviews). Summarize recurring workflows, failure points, and ‘must-have’ moments. Output a prioritized list of assumptions to validate with 8–10 interviews.” | Convert into a workflow library (jobs, steps, friction points) + prioritized interview questions. Use it to draft journey maps, UX requirements, and a support readiness list (likely confusions → docs needed). |
Implementation playbook | “Deep research best practices for rolling out [tool/process] across a mid-sized/enterprise org. Cover change management steps, training approaches, governance models, common failure modes, and metrics that indicate adoption/value. Produce a 30/60/90-day plan with roles and deliverables.” | Repurpose into a rollout plan (milestones, RACI, comms templates, training plan, success metrics) + risk register. Create a reusable 90-day checklist and a weekly status update template. |
Important considerations
- Validate before relying. Citations help you trace claims back to sources, but you should still review the underlying sources—especially for decisions, external publishing, or high-stakes topics.
- Not professional advice. Outputs may touch on legal, medical, financial, security, or regulatory topics; treat them as informational and consult qualified experts as needed.
- Respect IP. Use outputs as summaries/synthesis. Avoid copying long passages verbatim into external materials, and keep citations when you reuse content.
Table Of Contents