Core Build: Full Research Project

Part 1: Research Question Formulation (10 minutes)

Choose Your Research Topic

Select a topic you genuinely need to understand, not a hypothetical exercise. Consider what decision this research will inform. Are you evaluating market opportunities? Assessing technical solutions? Understanding competitive landscapes? Real stakes produce better questions, which produce better research.

Examples of decision-driven topics: Market analysis for product positioning, technical evaluation of infrastructure options, trend forecasting for strategic planning, competitive intelligence for partnership decisions.

Craft Effective Question

Use question patterns that work well with Gemini's autonomous research capabilities. These patterns signal clear intent and scope to the research engine.

Pattern 1: "What are the latest [topic] as of [year]?" Example: "What are the latest developments in retrieval-augmented generation (RAG) as of 2025?"

Pattern 2: "How are [actors] using [technology] for [purpose]?" Example: "How are financial institutions using large language models for risk assessment?"

Pattern 3: "What are the key differences between [options]?" Example: "What are the key differences between vector databases for production LLM applications?"

Refine for Specificity

Add constraints that narrow scope without limiting insight: timeframe, geography, industry, use case, scale. Specificity guides the autonomous agent toward relevant sources and prevents generic overviews.

Bad question: "Tell me about AI in healthcare" Good question: "What are the current applications of generative AI in clinical documentation as of 2025?"

Bad question: "How does machine learning work?" Good question: "What are the latest techniques for fine-tuning large language models on domain-specific data?"

The difference between vague and specific questions is the difference between Wikipedia summaries and actionable intelligence.

Part 2: Run Deep Research and Monitor (10 minutes)

Launch Deep Research

Enter your refined question in Gemini's interface. Activate Deep Research mode explicitly if not automatically detected. Gemini will present a suggested research plan showing the topics it intends to explore. Review this plan before starting. If it misses critical angles or includes irrelevant tangents, refine your question now rather than after the research completes.

The research plan preview is your chance to course-correct before investing 10+ minutes of autonomous browsing.

Monitor Autonomous Research

Watch sources appear in real-time as Gemini browses the web. The interface shows what sites it's visiting, which topics it's exploring, and how many sources it's collecting. Notice the diversity of source types: academic papers, industry reports, technical documentation, news articles, case studies, product comparisons.

Typical runtime for comprehensive topics: 8-12 minutes. Complex or emerging topics may run longer. Very short runtimes (under 5 minutes) often indicate overly narrow questions or limited available information.

During monitoring, you can observe Gemini's research strategy. Does it start broad and narrow down? Does it explore multiple perspectives? Does it follow citations from authoritative sources? This transparency builds trust in the autonomous process.

Initial Review

When research completes, skim the generated report structure. Check section organization: Introduction, Key Findings, Analysis, Implications, Conclusion. Well-structured reports indicate good synthesis. Flat lists of facts indicate shallow research or poor question formulation.

Count sources discovered. Target: 40+ sources for most comprehensive topics. Fewer sources may indicate niche topics or need for question refinement. More sources (60-80+) suggest rich literature and strong synthesis potential.

Part 3: Source Quality Evaluation (10 minutes)

Review Source List

Navigate to the Sources section at the report bottom. Gemini lists all sources with titles, URLs, and publication dates. Check source diversity across dimensions: authoritative institutions versus niche publications, global versus regional perspectives, primary research versus secondary analysis.

Source diversity indicates comprehensive coverage. Homogeneous sources (all news articles, all from one region, all from one timeframe) suggest limited perspective or search bias.

Evaluate Source Authority

Look for authoritative signals: research institutions, peer-reviewed journals, established industry analysts, official documentation, recognized publications. These sources have editorial standards and citation accountability.

Acceptable secondary sources: Technical blogs with clear citations, product documentation, detailed case studies, conference presentations, reputable news outlets covering specialized beats.

Red flags: Promotional content disguised as research, domains with low authority, very old sources for rapidly evolving topics, sources that contradict consensus without strong evidence, broken links or paywalled content Gemini cannot access.

Authority is contextual. A startup engineering blog may be the best source for emerging technical patterns. A 2019 paper may be foundational for theory. Use judgment based on claim type.

Cross-Check Key Claims

Pick three to five key findings from the report that would influence your decisions. Trace back to source citations by clicking inline citation numbers. Verify: Does the source actually support the claim? Is the claim accurately represented or overstated? Is critical context preserved or lost?

This verification builds confidence in report accuracy and reveals synthesis quality. Strong synthesis preserves nuance. Weak synthesis cherry-picks facts or misrepresents findings.

Part 4: Report Refinement and Synthesis (10 minutes)

Assess Synthesis Quality

Check whether the report synthesizes insights across sources rather than just summarizing each source individually. Strong synthesis identifies patterns, reconciles contradictions, highlights consensus, and notes outlier perspectives. It connects dots across sources to generate insights no single source provides.

Look for evidence of cross-source analysis: comparison of findings, integration of complementary research, acknowledgment of conflicting evidence, temporal progression of understanding. These indicate deep research, not shallow aggregation.

Evaluate whether the report provides insights beyond facts. Does it explain why trends emerged? Does it identify implications for different stakeholders? Does it contextualize findings within broader developments? Insight transforms information into intelligence.

Identify Gaps or Weaknesses

Review the report critically for missing perspectives. Are there obvious stakeholders not represented? Recent developments not covered? Alternative viewpoints not explored? Geographic or demographic blind spots?

Consider whether additional targeted queries could fill gaps. Sometimes a follow-up Deep Research session with a narrower question yields complementary insights. Other times, manual source additions are appropriate for very recent developments or niche perspectives.

Evaluate report balance. Does it present multiple viewpoints fairly? Does it acknowledge uncertainty where evidence is limited? Does it distinguish established facts from emerging hypotheses? Balanced reports earn reader trust.

Export and Format

Copy the report to your preferred format: Google Docs, Notion, Obsidian, Markdown files. Preserve source citations and hyperlinks. These are essential for verification, sharing, and future reference.

Add your own annotations: executive summary highlighting key takeaways, margin notes on particularly relevant sections, questions for further research, action items based on findings. Personal annotations transform research into action.

Consider sharing the report with collaborators or decision-makers. Deep Research reports are self-contained: they explain context, cite sources, and present synthesis. Recipients don't need to attend the research process to benefit from the outcome.