AI-Assisted Literature Discovery

AI literature discovery tools work differently from traditional database searching. Instead of matching exact keywords, they use semantic understanding to find papers related to your question conceptually. Tools to know: Semantic Scholar: A free academic search engine that uses AI to analyze paper content, identify key findings, and surface related work. It covers over 200 million papers. Its 'TLDR' feature provides one-sentence summaries. Elicit: Designed specifically for researchers. Enter a research question in natural language, and it finds relevant papers, extracts key findings, and organizes them in a table. Useful for rapid evidence mapping. Consensus: Searches the scientific literature and synthesizes findings using AI. Presents a 'consensus meter' showing the balance of evidence for or against a claim. Useful for quick evidence checks. Research Rabbit: A citation-based discovery tool. Add seed papers and it recommends related papers based on citation networks and content similarity. Useful for exploring a topic you are new to. What these tools do well: Find conceptually related papers regardless of keyword overlap. Provide quick summaries of findings. Help you discover papers in adjacent fields. Reduce the initial overwhelm of a new topic. Critical limitations: 1. Incomplete coverage: No AI tool indexes all journals. Many index only open-access papers or papers from specific publishers. You cannot know what they missed. 2. Non-reproducible: AI search results may differ between users and over time. You cannot document a reproducible search strategy. 3. AI hallucination: Large language models (like ChatGPT, Claude, and others) can fabricate citations — inventing realistic-looking paper titles, authors, and DOIs that do not exist. NEVER cite a paper based solely on an AI-generated reference without verifying it exists in PubMed or another database. 4. Not accepted for systematic reviews: PRISMA requires transparent, reproducible search strategies. AI-assisted tools currently do not meet this standard. Best practice: Use AI tools for initial exploration, brainstorming search terms, and discovering papers you might have missed. Then verify every reference in a real database. Always build your formal search in PubMed, Embase, or Cochrane using the methods from previous lessons.