Skip to content

Citations & Traceability

Every claim in ResearchCrew output is directly traceable to a source.

Why Citations Matter

The Problem

Generic AI research produces un-sourced claims:

"AI adoption in healthcare is accelerating. 
Companies report significant improvements in diagnostic accuracy and efficiency."

Issues:
- Where is this information from?
- Can I verify it?
- How do I cite this in my own work?
- Is it recent or outdated?

The ResearchCrew Approach

Every claim includes a direct link to the source:

"[AI adoption in healthcare is accelerating](https://healthcare-journal.com/trends-2025), 
with [companies reporting 30-40% improvements in diagnostic efficiency](https://tech-report.com/healthcare-ai-2025)."

Benefits:

  • You can click and verify each claim
  • Readers trust the information (transparent sources)
  • You can cite this report in your own work
  • Source recency is clear (what year? what publication?)

Citation Format

ResearchCrew uses markdown inline citations:

[claim text](URL)

Example:

[GPT-4 can solve 92% of LeetCode problems](https://openai.com/research/gpt4)

This creates a clickable link in markdown viewers and web browsers.

Citation Strategy

Stage 1: Extraction

During content extraction, claims are linked to their source page:

{
  "url": "https://healthcare-journal.com/study-2025",
  "title": "AI in Medical Diagnostics 2025",
  "claims": [
    {
      "claim": "AI diagnostic accuracy reached 95% in 2024",
      "quote": "...our AI system achieved 95% accuracy in identifying tumors...",
      "confidence": "HIGH"
    }
  ]
}

Stage 2: Synthesis

When synthesizing across sources, each source's contribution is tracked:

## Diagnostic Accuracy

Multiple studies demonstrate high accuracy in AI diagnostics:
- https://study-a.com reports 95% accuracy
- https://study-b.com reports 92% accuracy
- https://study-c.com reports 97% accuracy

Stage 3: Reporting

Final report includes direct inline citations:

[Multiple studies demonstrate AI diagnostic accuracy of 92-97%](https://study-a.com), 
with [cardiac imaging seeing the best results](https://study-c.com).

Citation Accuracy

Matched Claims

The crew ensures citations match claims:

Correct:

[AI adoption in US hospitals reached 60% in 2024](https://valid-source.com/stat)

The URL actually contains this statistic.

Incorrect (hallucination):

[AI adoption in US hospitals reached 95% in 2024](https://valid-source.com)

The URL says 60%, not 95% — citation doesn't support claim.

ResearchCrew avoids this through:

  1. Extracting verbatim quotes alongside claims
  2. Scoring confidence based on source reliability
  3. Human review (you can catch mismatches)

Citation Freshness

Citations should be recent and relevant:

Good:

[Recent studies from 2024-2025 show...](https://2024-study.com)

Questionable:

[Recent studies show...](https://2015-study.com)

5-10 year old source may not be "recent"

ResearchCrew mitigates by:

  1. Prioritizing recent sources in web search
  2. Using terms like "2024-2025 studies" in search queries
  3. You can provide feedback if sources are outdated

Multi-Source Citations

Single Source

When one source is sufficient:

[AI is transforming healthcare delivery](https://authoritative-source.com/report)

Multiple Sources

When multiple sources support a claim:

[AI is transforming healthcare delivery](https://source-a.com/report), 
with [adoption rates accelerating globally](https://source-b.com/trends).

Conflicting Sources

When sources disagree:

Adoption rates vary significantly by region:
- [US adoption reached 80% in 2024](https://us-report.com)
- [EU adoption is at 45%](https://eu-report.com)
- [Asia-Pacific adoption is 60%](https://apac-report.com)

This allows readers to see the full picture, not just consensus.

Citation Chains

Some claims require multiple citations to fully support:

[AI diagnostics can be 95% accurate](https://source-a.com/accuracy), 
but [implementation in rural hospitals faces infrastructure barriers](https://source-b.com/barriers), 
with [estimated deployment costs of $500K-$1M per facility](https://source-c.com/costs).

This chain:

  1. Establishes the capability (accuracy)
  2. Identifies constraints (infrastructure)
  3. Provides implementation context (costs)

Each claim is independently verifiable.

Citation in Report Sections

Example Report Structure

# Research Report: AI in Healthcare

## Executive Summary

[AI is transforming healthcare with diagnostic improvements](https://source.com),
with [adoption rates increasing 40% annually](https://source2.com).

## Diagnostic Accuracy

[Multiple AI systems now achieve 90%+ accuracy](https://source3.com),
compared to [human radiologist accuracy of 88%](https://source4.com).

### Limitations

However, [AI systems require substantial training data](https://source5.com),
and [regulatory approval timelines can exceed 2 years](https://source6.com).

## Implementation Challenges

[Large hospitals have infrastructure for AI deployment](https://source7.com),
while [rural healthcare providers often lack adequate IT infrastructure](https://source8.com).

Every claim is linked. Readers can verify any statement.

How to Use Citations

Sharing Research

When you share report.md:

  1. Citations are embedded as clickable links
  2. Readers can verify claims
  3. They can cite your work, which cites original sources

Citing ResearchCrew Reports

If using ResearchCrew output in your own work:

According to my research on AI in Healthcare 
(conducted with ResearchCrew on 2025-05-16):
- [AI diagnostic accuracy reaches 95% in some applications](https://...)
- [Implementation challenges persist in rural settings](https://...)

Sources are directly cited and verifiable.

Academic/Professional Use

ResearchCrew reports are suitable for:

  • Research papers (cite the original sources)
  • Business reports (data is backed by sources)
  • Team documentation (shared context with full sources)
  • Presentations (every slide can be substantiated)

Citation Best Practices

When Reading ResearchCrew Output

Do:

  • Click citations to verify claims
  • Check if source supports claim
  • Validate citation freshness (recent enough?)
  • Consider source bias or perspective

Don't:

  • Assume all citations are correct (verify!)
  • Use claims without reading source
  • Share without understanding source quality
  • Assume sources can't be wrong

When Providing Feedback

If a citation seems wrong:

## User Feedback

The report cites [Source A showing AI accuracy of 95%](https://source-a.com).
I read the source and it actually says 92%, not 95%.

Please verify this citation and correct the report.

The crew will re-validate the citation in the next iteration.

Citation Formats

ResearchCrew uses markdown inline citations. This format:

  • Works in markdown files
  • Renders as clickable links in browsers
  • Preserves sources in text files
  • Is human-readable

Other Formats (Future)

Could be extended to:

  • [ ] Markdown inline citations (current)
  • [ ] Footnotes with bibliography
  • [ ] BibTeX entries
  • [ ] APA/MLA/Chicago style
  • [ ] Structured JSON with full metadata

Limitations

ResearchCrew citations have limitations:

  • URL Links May Break

  • If source is deleted or moved, link becomes invalid

  • Save PDFs of important sources for archival

  • Sources Can Be Biased

  • ResearchCrew filters for authority, not objectivity

  • A credible source can still present a biased perspective

  • Sources Can Be Wrong

  • Even reputable sources can contain errors

  • Always apply domain expertise to validate

  • No Permanent Archives

  • Citation links depend on URLs staying live

  • Consider archiving important sources

Next Steps