Protecting Research Integrity: Best Practices for Authors Using AI-Assisted Research

Generative AI tools are fundamentally altering how scholarly work gets done and are rapidly becoming embedded in scholarly workflows. From literature discovery and summarization to reference list generation, AI helps researchers navigate information overload, accelerate discovery, and uncover new connections. Due to these time-saving benefits and efficiencies, studies show that a third of researchers today now begin their information seeking process with AI (Rapple, 2026), and more than 80% of researchers use AI in some phase of their work (ExplanAItions, 2025). But many researchers are unaware of the extent to which their view of the scientific literature is now AI-mediated. Alarmingly, 40% mistakenly believe that AI overviews generated on the fly in search engines have been created or curated by a human (Rapple, 2026).

However, while the research process has changed dramatically with the use of AI, it also brings a high degree of risk. The improper use of AI can introduce hallucinations, bias, loss of provenance, and false citations into research. If your research team uses AI-assisted tools during the research process, the following best practices can help validate findings and support the most accurate outcomes possible.

Best Practice Tips for Authors for AI-Assisted Research

• Apply Critical Thinking: AI can be persuasive as it produces confident and convincing answers. But researchers need to scrutinize AI results and practice critical thinking before accepting answers from any LLM tool to avoid being misled. Common warning signs could be results that appear overly simplified, very definitive without clear justification, or lack references.
  • Questions to ask yourself: Can I verify this? Does the model point to real, verifiable sources, or is it guessing based on patterns? Is it hallucinating? Does the logic appear coherent and consistent?
• Introduce Deliberate Friction into the Workflow: Instead of the common approach of making every process as frictionless as possible, it is important to slow down and create friction along the research journey to verify findings before proceeding to the next step. Removing all obstacles when using GenAI tools can lead to anchoring bias and lower analytical quality. Integrating deliberate friction can act as a cognitive speed bump that prompts users to pause and test assumptions before finalizing insights, and improves the accuracy of the final results in AI-enabled research workflows.
• Cross-Reference Facts: Double-check LLM-provided facts, statistics, and claims against multiple trustworthy, primary sources.
• Validate All Citations: Verify all citations and references provided by the LLM. If you can’t find the provided citation, the source likely doesn’t exist, and this could be a hallucination. You can usually find that information on a publisher’s website or a trusted index.
• Check for Currency and the “Version of Record”: AI systems don’t always make a distinction between current and dated information. Ensure that the information provided is current and still relevant. Be sure you are referencing the most current version of the material, or the “version of record,” and there have been no new amendments, retractions, etc. Again, you can usually obtain that information from a publisher’s website or a trusted index.
• Watch for AI-Writing Patterns: Are you unsure if the information you are citing was in fact an article that was created by AI? Always seek to verify by viewing the original, primary source to ensure the article and the publication it was published in are in fact authentic. Some common cues of AI-generated prose to look for:
  • “Pretentious diction” or dressing up simple statements
  • Use of polysyllables and rare words (“interdependence”)
  • Less punctuation and longer sentences
  • Fewer commas and semicolons
  • Lack of parentheses
  • Increased use of em dashes
• Check AI Summaries Against Original Texts: AI summaries can include errors as it may misconstrue information, lose nuance and context, omit critical information, overgeneralize findings, ignore scientific uncertainties and study limitations, and hallucinate inaccurate details. View the original source to confirm that the AI tool summarized or paraphrased the conclusions of the original source correctly, and that the meaning is accurate and intact.
• Identify Potential Biases: Look for potential biases introduced by LLM usage, checking against original sources. Some questions to consider:
  • Is the AI relying on old data trends, historical stereotypes, or unverified patterns?
  • Does the answer favor a specific cultural or political viewpoint?
  • Is there a perspective that is left out of this answer?
• Ensure Prompts are Correct: Compare LLM-assisted results with traditional research methods to ensure your prompts are delivering the desired results that are relevant, on-topic and aligned with your research objectives.
• Avoid Using Generic Prompts: General terms trigger answers derived from generic, SEO-optimized blog posts rather than more authentic primary sources. The more specific the instructions, the more likely that results will be returned from trusted sources. To unlock trusted sources, structure your searches using specific constraint formulas to seek primary sources.

Responsible AI Starts with Responsible Research
AI has the potential to be one of the most powerful research tools ever developed. It can accelerate discovery, reduce administrative burdens, and help researchers uncover connections that may otherwise be overlooked.

But with those benefits comes responsibility. Research integrity depends on maintaining rigorous standards of verification, transparency, and critical evaluation. AI should be viewed as an assistant, not an authority. By verifying sources, checking citations, questioning outputs, and prioritizing primary evidence, researchers can leverage AI’s advantages while protecting the accuracy and credibility of the scholarly record. As AI becomes more deeply embedded in the research process, researchers must balance efficiency with vigilance. Whenever possible, work with independent researchers who can verify key findings from the original sources.

AI can assist in the scholarly workflow, but authors bear full responsibility for the accuracy and integrity of their work. Authors should follow the best practices and tips above to help verify their research results and ensure the most accurate outcomes. Also, all authors should review and follow IEEE’s Author Guidelines for Artificial Intelligence (AI)-Generated Text when preparing their manuscript for submission to IEEE publications.

Reference List:
ExplanAItions 2025: The evolution of AI in research. (2025). Wiley.
Rapple, C. (2026, July 15). AI has captured a third of research discovery.