Integrating AI Tools into Academic Research: A Step‑by‑Step Guide Using Perplexity, Elicit, and NotebookLM

Introduction

The session demonstrated a fully browser‑based workflow for conducting academic research with generative AI. Participants were asked to bring a research topic; the facilitator used impact of AI on small‑holder farming in East Africa as a live example.

1. Discovery Phase

• Perplexity.ai – an LLM‑powered search engine that returns concise, sourced answers while minimizing hallucinations. The query "impact of AI on small‑holder farming in East Africa" produced a structured summary covering:
• Precision‑farming tools
• Early‑warning climate systems
• Computer‑vision pest detection
• Economic benefits and challenges (infrastructure, cost, ethics)
• Source links from reputable organisations (World Bank, FAO, BMZ).
• Elicit.com – a research‑article finder. After signing in, the same query returned the top‑ranked papers with DOI, citation count, year, and a short abstract. The facilitator downloaded PDFs for five papers and noted that one required a subscription.

2. Organising Sources

• All PDFs and web links were collected in a local folder.
• The facilitator highlighted the importance of curating high‑quality sources before feeding them to an LLM to avoid “garbage‑in, garbage‑out”.

3. Interrogation & Analysis with NotebookLM

• NotebookLM (Google’s AI notebook) provides three panels:
• Sources – upload up to 50 URLs, PDFs, YouTube videos, or plain text.
• Chat – interact with the uploaded material; the model answers questions using only the supplied sources, eliminating hallucinations.
• Studio – generate enriched outputs such as audio overviews, video summaries, mind maps, flashcards, and quizzes.
• The facilitator added the PDFs from Elicit and the URLs from Perplexity to the Sources panel.
• Sample prompts:
• “Create a table summarising the core message, research focus, and main findings of each source.”
• “Generate a mind‑map of opportunities, challenges, and ethical concerns for AI in East African agriculture.”
• “Produce an audio overview of the collected material.”
• The model returned a structured table, a critical comparison of theoretical frameworks, and a visual mind‑map that could be expanded by clicking nodes.
• Note‑taking: Anything typed in the chat is transient unless saved as a note; the facilitator warned participants to click Save as note to preserve insights.

4. Multi‑Modal Learning Aids

• Audio Overview – converts the textual synthesis into a short podcast (≈12 min) for auditory learners.
• Video Overview – a brief visual recap (limited to ~3 min in the free tier).
• Mind‑Map & Gap Analysis – visual representations that help both visual and analytical learners grasp relationships between concepts.
• Flashcards & Quiz – optional tools for self‑testing knowledge.

5. Ethical & Practical Reflections

• The facilitator emphasized responsible AI use:
• Verify sources (FAO, World Bank, peer‑reviewed journals).
• Be aware of algorithmic bias, especially gender‑related design violence.
• Treat data as a community‑owned resource rather than a commodity.
• Participants asked about saving outputs; the answer was that notes, mind‑maps, and media can be downloaded or exported to Google Drive.

6. Next Steps & Future Sessions

• Participants are encouraged to repeat the workflow with their own topics.
• Upcoming breakout rooms will focus on turning the interrogated material into a research draft, followed by specialized tools for writing, citation management, and validation.
• The facilitator invited attendees to share the YouTube recording, Google Drive resources, and to spread the word for future AI‑training webinars.

Key Takeaways

• AI‑augmented research can be done entirely in the browser using free or freemium tools.
• Perplexity provides quick, sourced overviews; Elicit surfaces scholarly articles; NotebookLM lets you interrogate, summarise, and repurpose those sources without hallucination.
• Multi‑modal outputs (tables, mind‑maps, audio) cater to different learning styles and dramatically reduce the time spent manually reading and organising dozens of papers.
• Ethical vigilance—checking provenance, acknowledging bias, and respecting data sovereignty—remains essential throughout the workflow.

By chaining Perplexity, Elicit, and NotebookLM, researchers can streamline the entire research pipeline—from discovery to analysis to documentation—while maintaining source transparency and reducing the risk of AI hallucination. This browser‑only workflow empowers scholars to focus on critical thinking rather than tedious data gathering, making AI a reliable partner in ethical academic inquiry.