Researchers, students and journalists increasingly use AI to explore topics, summarise papers and draft outlines. Used well, it can save hours of reading. Used carelessly, it can insert fake references, misrepresent studies and quietly bias your work. The key rule is simple: AI can help you find and understand sources, but you must verify them yourself.
What AI does well in research
- Explaining unfamiliar concepts at different levels
- Suggesting search terms, synonyms and related fields
- Outlining a literature review or structuring questions
- Summarising a paper you provide (with checks)
- Comparing methods or findings across documents you give it
- Extracting data into tables from text, with verification
- Drafting questions for interviews or surveys
- Improving clarity and grammar of your own writing
What it does poorly
- Providing accurate citations from memory. Language models can generate references that look real, with plausible authors, titles and journals, but do not exist.
- Knowing recent publications, unless connected to search.
- Reporting exact numbers, quotes and page numbers reliably.
- Judging study quality consistently.
- Representing minority or local research, which is underrepresented in training data.
A reliable workflow
Step 1: define your question
Write a clear research question and what kind of evidence you need. A vague topic leads to vague help.
Step 2: use AI to plan the search
Ask for key terms, subtopics and alternative phrases:
I am researching the effect of mobile phone use on sleep in teenagers. Suggest 15 search keywords and phrases, including synonyms and related concepts, and name the types of studies I should look for (for example, systematic reviews, cohort studies).
Step 3: search real databases
Use academic search engines, library databases and reputable repositories such as Google Scholar, PubMed (for health), Semantic Scholar, arXiv, institutional repositories and government statistics sites. Record the exact citations and links yourself.
Step 4: read, then use AI to help understand
Open the actual paper. If something is dense, paste the relevant passage and ask for an explanation or summary. Compare the summary against the original. Pay special attention to the abstract, methods, sample size, limitations and conclusions.
Step 5: verify every fact and citation
For any reference the AI suggests:
- Search for the exact title in a scholarly database.
- Confirm authors, journal, year and DOI.
- Open the paper and check that it says what the AI claims.
- Check quotes against the source.
- Look for retractions or corrections.
If you can't find it, treat it as nonexistent.
Step 6: evaluate quality
Ask critical questions about each source:
- Who produced it and why? Any conflicts of interest?
- Was it peer reviewed? Is it a preprint?
- How large and representative is the sample?
- Are methods clear and appropriate?
- Do other studies agree?
- Is the claim stronger than the evidence?
AI can help generate a checklist, but your judgement decides.
Step 7: synthesise in your own words
Take notes in your own words and track which source supports which statement. Use AI to check structure and clarity, but make the argument yours. Cite original sources, not the AI.
Prompts that reduce errors
- "Using only the text below, summarise the main findings. Do not add anything not in the text."
- "Quote the sentence from the text that supports each claim."
- "List the limitations the authors mention."
- "What would a sceptical reviewer criticise about this study?"
- "I will give you three abstracts. Create a table comparing sample size, method and main result. Write 'not stated' where missing."
Avoid: "Give me 10 peer-reviewed references on X", unless the tool has real search access and you verify results.
Checking references in code
You can sanity-check a DOI against a public registry such as Crossref:
import requests
def check_doi(doi: str):
r = requests.get(f"https://api.crossref.org/works/{doi}", timeout=15)
if r.status_code != 200:
return None
item = r.json()["message"]
return {
"title": item.get("title", [""])[0],
"journal": (item.get("container-title") or [""])[0],
"year": item.get("issued", {}).get("date-parts", [[None]])[0][0],
"authors": [f"{a.get('given', '')} {a.get('family', '')}".strip() for a in item.get("author", [])][:3],
}
print(check_doi("10.1038/nature14539"))A DOI that does not resolve, or resolves to a different paper than claimed, is a red flag. Note that existence is only the first check; you must still read the paper.
Ethics and integrity
- Follow institutional and journal policies on AI use and disclosure.
- Do not present AI text as your own where prohibited.
- Protect confidential or unpublished data; do not paste it into tools that may store it.
- Respect copyright when uploading papers to AI tools.
- Acknowledge AI assistance when required, for example in methods or acknowledgements.
- Be wary of bias: AI may favour well-known Western sources and miss local research. Seek out regional journals and data.
Research with local context
For topics related to Nepal, such as agriculture, health or education, look for national statistics, government reports, university theses and regional journals alongside international literature. AI may not know them, so you must search them directly.
Frequently Asked Questions
Can I trust AI search tools that provide links?
They are more reliable than models answering from memory, but they can still misread pages or choose weak sources. Always open the links.
Is it acceptable to use AI to summarise papers?
Usually yes, as long as you verify the summary against the paper and follow your institution's policies.
How do I cite AI use?
Follow the guidelines of your style guide or institution. Some require citing the tool and the prompt, others a disclosure statement.
What is the biggest risk?
Fabricated or distorted citations that you pass on without checking.
Conclusion
Treat AI as a research assistant that suggests search terms, explains and organises, and verify every source yourself. Real databases, careful reading and critical evaluation remain the foundation of trustworthy research.