Explore Prompts

Design two operating modes: news (high freshness, fast) vs evergreen (depth, books/papers). Include different source scoring, caching, and output formats.

Design cross-language research: translate queries, search in multiple languages, normalize entities, and merge evidence. Include how to detect mistranslations and preserve proper nouns.

Design an entity extraction + graph pipeline: NER, entity resolution, relationship extraction, and graph storage. Include queries to answer “who is connected to what.”

Create a method to cluster collected sources into themes: embeddings, hierarchical clustering, and label generation. Include how to keep clusters stable across updates.

Design an evidence table format (CSV/JSON): each claim mapped to sources, snippets, dates, and reliability scores. Include how to generate it automatically from the pipeline.

Create a standard research brief template: executive summary, key findings, evidence table, open questions, and citations. Include options for analyst vs layperson formats.

Design a bias-control policy: require diverse sources (geography, institution, viewpoint), detect monocultures, and add “missing perspective” prompts. Include reporting in outputs.

Create a workflow for conflicting sources: identify disagreements, categorize (definition vs data vs interpretation), and present balanced conclusions with confidence levels.

Design a claim extraction step: identify factual assertions, attach supporting evidence, and flag unsupported claims. Include scoring for evidence strength and contradictions.

Design a citation system: store supporting snippets, anchor to URL + section/page, and enforce quote-length limits. Include how to generate citations in final responses automatically.

Draft a policy for paywalled/restricted content: detect paywalls, avoid bypass, rely on accessible sources, and suggest user-provided excerpts when needed. Include user messaging guidelines.

Create a plan for handling PDFs: downloading, parsing text, capturing page numbers, and generating citations. Include limitations and a quality checklist for extraction accuracy.