AI tools like ChatGPT, Perplexity, Claude, and Gemini can now search the web and pull together sources for you in seconds — whether through NotebookLM's Deep Research, ChatGPT's web browsing, or Claude's web search.
That's incredibly useful, but it also shifts a key responsibility: instead of choosing your own sources, you're evaluating ones the AI chose for you.
This guide covers three parts to help you get better results and trust what comes back:
Part 1: Prompting for Better Sources — How to write prompts that steer AI toward higher-quality, more relevant material from the start.
Part 2: Verifying AI-Found Sources — A seven-point checklist for checking that what the AI returns is real, accurate, and trustworthy.
Part 3: Choosing the Right Approach — How to decide when a quick check is enough versus when you need to dig deeper, and when to let AI lead versus keeping full control.
Can you actually trust AI research tools like ChatGPT and Perplexity?
How do you write prompts that get better sources from the start?
What's the quickest way to verify what AI finds?
When should you let AI lead versus keeping full control?
Which AI tool is best for research — and when?
SUMBIT the form below to
We’ll only use your details for the purposes you select.
See our Privacy Policy for more information.
When you ask an AI to research something, the quality of your prompt influences the quality of sources it finds. You can't force perfect results, but you can guide the AI toward more reliable material.
"Research the benefits of intermittent fasting"
...will return sources, but you have no influence over:
Whether they come from credible publishers
Whether they represent balanced perspectives
Whether they're current
Whether they're primary or secondary sources
The AI will find something. Your job is to steer it towards something useful.
1. Specify the source types you want
Tell the AI what kinds of sources to prioritise and what to avoid.
Instead of:
"Research electric vehicle adoption in Europe
Try:
"Research electric vehicle adoption in Europe. Prioritise peer-reviewed studies, official government statistics, and reports from established research organisations. Avoid opinion pieces and industry marketing materials."
2. Request recency where it matters
For fast-moving topics, specify a time frame.
Instead of:
"Research AI regulation in the UK"
Try:
"Research AI regulation in the UK, focusing on developments from 2024 onwards. Prioritise official government sources, legislation, and analysis from legal or policy experts."
3. Ask for multiple perspectives on contested topics
Prevent one-sided source selection by making balance explicit.
Instead of:
"Research the effectiveness of remote work"
Try:
"Research the effectiveness of remote work. Include sources presenting different perspectives — both research supporting remote work benefits and research highlighting challenges or limitations. Prioritise academic studies and surveys with clear methodology over opinion pieces."
4. Request primary sources explicitly
Push toward original material rather than summaries and commentary
Instead of:
"Research the recent Supreme Court ruling on [topic]"
Try:
"Research the recent Supreme Court ruling on [topic]. Prioritise the original court decision, official court documents, and legal analysis from qualified legal scholars. Include the primary source document if possible."
5. Exclude source types you don't want
Be explicit about what to filter out.
Example:
"Research health benefits of meditation. Prioritise peer-reviewed medical and psychological research. Exclude wellness blogs, product marketing sites, and sources without named authors or citations."
Adapt these templates to your specific research needs.
Academic or Professional Research
Research [TOPIC].
Source requirements:
- Prioritise peer-reviewed academic sources, official reports, and recognised expert analysis
- Include sources published within the last [X] years unless historical context is needed
- Exclude opinion pieces, blogs without citations, and marketing content
- Where possible, find primary sources rather than secondary reporting
If the topic is contested, include credible sources representing different positions.
Current Events or Policy
Research [TOPIC], focusing on developments from [DATE] onwards.
Source requirements:
- Prioritise official government sources, established news organisations with editorial standards, and recognised policy experts
- Include primary documents (legislation, official statements, court filings) where available
- For news reporting, prefer outlets with clear editorial policies and named journalists
- Exclude anonymous sources, unverified social media, and overtly partisan commentary
Note any significant disagreements between sources.
Business or Market Research
Research [TOPIC].
Source requirements:
- Prioritise industry reports from recognised analysts, official company filings, government statistics, and trade publications with editorial oversight
- Be cautious with company-produced content — include it for stated positions but note it may be self-serving
- Include multiple perspectives if there are competing views on market trends
- Note publication dates — market conditions change quickly
Flag any sources that may have conflicts of interest.
Health or Medical Information
Research [TOPIC].
Source requirements:
- Strongly prioritise peer-reviewed medical journals, official health body guidance (NHS, WHO, CDC), and systematic reviews
- Include the level of evidence where possible (randomised controlled trial, observational study, expert opinion)
- Exclude wellness blogs, alternative health sites without medical credentials, and anecdotal sources
- If scientific disagreement exists, represent the consensus view and note where credible dissent exists
This is for informational purposes — note that it does not constitute medical advice.
Investigative or Fact-Checking Tasks
Research [TOPIC/CLAIM].
Source requirements:
- Prioritise primary sources: original documents, official records, direct statements from people involved
- For secondary sources, prefer established investigative journalism with clear sourcing
- Cross-reference key claims — find multiple independent sources where possible
- Note the provenance of each source: who published it, when, and what access they had
- Flag any sources that rely on anonymous claims or cannot be independently verified
I need to verify this information, so source quality is critical.
Add these phrases to any research prompt:
Goal
Add This
Improve Authority
"Prioritise sources from recognised experts, institutions, or peer-reviewed publications"
Improve Recency
"Focus on sources published after [DATE]"
Get Primary Sources
"Include original documents and primary sources where available"
Ensure Balance
"Include credible sources representing different perspectives"
Exclude Low-Quality Sources
"Exclude blogs without citations, marketing content, and anonymous sources"
Request Transparency
"Note any limitations or potential biases in the sources found"
They can:
Guide the AI toward better source types
Reduce (but not eliminate) low-quality sources
Encourage balance on contested topics
Make your requirements explicit
They cannot:
Guarantee the AI follows instructions perfectly
Prevent all hallucinated or misrepresented sources
Replace your own verification
Force the AI to find sources that don't exist
Better prompts improve your starting point. You still need to check what comes back.
AI research tools can hallucinate sources, misrepresent what sources say, or select sources that aren't appropriate for your needs. Before relying on AI-found sources, run through these checks.
1. Does the source actually exist?
AI can hallucinate URLs or mangle web addresses
Click through to the actual page
If you get a 404 error, search the article title directly
If you still can't find it, don't use it
Red Flag: A URL that looks plausible but leads to nowhere
2. Does it say what the AI claims?
The source exists, but the AI may have misrepresented it.
Read the relevant section yourself and not just the AI's snippet
Look for context that changes the meaning
Check whether opinions are presented as facts (or vice versa)
Red Flag: The AI's summary sounds more definitive than the original text
2. Does it say what the AI claims?
The source exists, but the AI may have misrepresented it.
Read the relevant section yourself and not just the AI's snippet
Look for context that changes the meaning
Check whether opinions are presented as facts (or vice versa)
Red Flag: The AI's summary sounds more definitive than the original text
3. Who published this and why?
Not all sources deserve equal trust.
Identify the publisher (news outlet, academic journal, company, individual)
Check if the author is named and has relevant expertise
Look for an "About" page; who runs this site?
Consider whether the publisher has a financial or ideological stake
Generic trust hierarchy:
Higher trust: Peer-reviewed journals, official government sources, established news organisations with editorial standards, recognised professional bodies
Medium trust: Reputable trade publications, named expert commentary, well-sourced journalism
Lower trust: Anonymous articles, company blogs on topics where they're selling something, content farms, unverified user-generated content
Red Flag: No author, no about page, no way to identify who's behind the content
4. When was it published?
Information ages at different rates.
Check the publication date
Consider whether the situation has changed since then
For fast-moving areas (technology, health, policy), even 6-12 months can matter
Red Flag: No visible date, or the date suggests the information may be outdated
5. Is this a primary or secondary source?
Primary sources are closer to the truth.
Primary: Original research, official documents, court filings, direct interviews, first-hand accounts
Secondary: News-reporting, summaries, analysis, commentary on primary sources
Where possible:
Trace secondary sources back to the primary source they cite
Prefer primary sources for important claims
be cautious if a claim only appears in secondary sources
Red Flag: Multiple articles all citing the same single source (circular sourcing)
6. What's Missing?
AI research finds sources, but not all sources
Consider what perspectives might be underrepresented
On contested topics, did it find sources from different viewpoints?
Are there obvious types of sources missing (academic research, official data etc.)?
Think about what search terms might find sources the AI missed
Red Flag: All sources share the same perspectives on a genuinely contested issue
7. Do multiple independent sources agree?
Corroboration increases confidence.
Cross-reference key claims across different sources
Check that sources are genuinely independent (not all citing eachother)
If only one sources makes a claim, treat it with more scepticism
Red Flag: A striking claim that appears in only one source
Check
Key Question
Exists
Can I actually access this source?
Accurate
Does it say what the AI claims?
Authoritative
Who published this and why?
Current
Is this information still relevant?
Primary
How close is this to the original?
Complete
What might be might be missing?
You don't need specialist software to assess source credibility. These quick checks work for most situations:
Check
What It Tells You
Can you find an "About" page?
Legitimate organisations identify themselves
Is the author named with credentials?
Anonymous content is harder to verify and trust
Does the site have ads everywhere?
Ad-heavy sites often prioritise traffic over accuracy
Does the site have ads everywhere?
Commercial motive may bias the information
Can you find this organisation mentioned elsewhere?
Real institutions have external footprints
Does the URL match a known organisation?
Watch for typosquatting and impersonation sites
Full check (all seven points):
High-stakes decisions (health, legal, financial, professional)
Content you'll publish or share publicly
Claims that seem surprising or too good to be true
Contested or politically-sensitive topics
Quick check (existence, authority, currency):
Background research and general understanding
Low-stakes personal decisions
Well-established factual matters
Quick check (existence, authority, currency):
Casual exploration
Topics where you already have expertise and can spot errors yourself
AI research tools sit on a spectrum between convenience and control.
Approach
Speed
Control
Best For
AI finds and synthesises sources
(Deep Research, Perplexity, ChatGPT browsing)
Fast
Low
Initial exploration, time-pressured research, broad overviews
AI works with sources you provide
(NotebookLM with uploaded docs, ChatGPT/Claude with files)
Medium
High
Initial exploration, time-pressured research, broad overviews
Traditional research
(you find and read sources yourself)
Slow
Full
Critical decisions, original research, maximum reliability
For important work, consider a hybrid approach:
Use AI research for initial exploration and source discovery
Verify and filter the sources it finds
Upload the vetted sources to a tool that works only from what you provide
Generate outputs from the verified material
This captures the speed benefit of AI research while maintaining quality control
AI research tools make finding information faster. They don't make verification unnecessary — they make it more important. The citations exist so you can check them. Use them.
Have questions about what you’ve just seen?
LEMA Logic is also a trading name of Gallagher Innovations, Inc. a company incorporated in Maryland, USA and registered in the Isle of Man - company 006459F.