AI as a Research Tool: Possibilities and Hard Limits
1 / 5AI has transformed research workflows for many professionals. It has also led to some high-profile failures — lawyers citing non-existent cases, journalists reporting AI-fabricated statistics. Understanding what AI can and cannot do in research is essential before you rely on it.
What AI Does Well in Research
Orientation and background When approaching an unfamiliar topic, AI is excellent for getting oriented quickly. It can explain terminology, outline the key debates, identify the major thinkers or institutions, and summarise the state of knowledge.
Synthesis of provided materials If you paste in your own source material — papers, reports, transcripts — AI is highly effective at synthesising it. This is arguably its best research capability: finding themes, contradictions, and gaps across large amounts of text.
Identifying angles and questions AI is a useful thinking partner for research design. Ask it what questions you should be asking, what angles have been overlooked, or what the strongest counterargument to your current hypothesis is.
Summarisation Condensing long documents into usable summaries is one of AI's most consistent strengths. A 60-page report can become a 500-word summary in seconds.
What AI Does Poorly in Research
Real-time or recent information Unless the model has web access, its knowledge has a training cutoff. For anything requiring current data, use tools with web access (Perplexity, ChatGPT with browsing, Gemini) or verify against primary sources.
Specific statistics and numerical claims This is where AI most dangerously hallucinates. It will produce specific-sounding numbers with no real basis. Never use an AI-generated statistic without independently verifying it.
Citations and references AI will generate plausible-looking citations — including authors, journals, years, and titles — that do not exist. This has ended careers. Never use an AI citation without verifying it exists.
Niche or specialised knowledge AI knowledge quality is proportional to how much was written about a topic in its training data. For niche or cutting-edge topics, its knowledge may be shallow, outdated, or subtly wrong.
The Golden Rule
Use AI to find your research faster. Use primary sources to verify it.
AI is a navigation tool, not a primary source. It gets you to the right territory; you still need to read the actual documents.
The Verification Habit
- 1.Build verification into your workflow as a non-negotiable step:
- 2.AI provides a summary, claim, or statistic
- 3.You identify the primary source it should be traceable to
- 4.You find and read the primary source yourself
- 5.Only then do you rely on the claim
This sounds like it removes the efficiency gain. It does not — AI still saves you significant time finding and summarising; verification is a fraction of the total research time.