Optimizing Desk Research with AI | Forum Desain Publik
Optimizing Desk Research with AI
This project utilized artificial intelligence to accelerate two main stages of desk research: literature collection and insight synthesis. With AI support, a small team of two researchers and two writers was able to explore a broader range of literature, discover new concepts, and generate sharper, more actionable insights in less time.
Role
ResearcherContent Designer
Process
Problem Discovery & DefinitionDesign & Solutioning
Output
Research & Insight Synthesis
Context & Background
In the early research stage, the team needed to build a thorough understanding of the subject, from social context to supporting theories. The challenges included:
The volume of literature was vast, covering academic articles, whitepapers, and international reports (e.g., from UNICEF).
Limited resources: only 4 people (2 researchers and 2 writers).
A need for speed while maintaining quality.
To address this, the team brought in AI to:
Speed up the search and filtering of literature.
Expand the exploration and uncover overlooked ideas and concepts.
Generate insights that could immediately support internal discussions and product development.
Approach & Process
⇢ Tools
Sourcing
ChatGPT → Used for initial reference gathering, drafting exploratory questions, and suggesting paper titles/topics. Ideal for quick brainstorming and wide exploration.
Perplexity → A real-time Q&A engine with citation features. Copilot and related questions allow deeper exploration from a single starting point. Useful for sourcing academic and general texts.
ReadWonders → Curates articles clustered by theme. The exploration feed and topic journey features help the team find new perspectives from different angles, especially in academic texts.
Writing & Communication Systems
Processing
Anara → Used to compare multiple literatures covering similar topics. Anara highlights key points and perspectives from each source, helping the team spot contrasting views or thematic threads. Most effective for academic papers and theory-heavy research.
Notebook LM → Used to synthesize insights. It summarizes and constructs narrative scripts from a bundle of texts (including screenshots), producing presentation-ready scripts.
⇢ Process
Literature Sourcing
The team crafted specific prompts in each platform (ChatGPT, Perplexity, ReadWonders) to obtain:
A list of relevant papers or articles
Summaries of main points
Uncommon or novel perspectives
Results: Each tool delivered outputs that accelerated the initial orientation. The team didn't have to manually read through each abstract or introduction.
New Topic Exploration
Exploration features in ReadWonders and Perplexity were used to:
Discover alternative viewpoints
Reveal derivative questions rarely found in conventional searches
Uncover connections between topics (e.g., between literacy, habit formation, and social learning)
Results: Several new ideas and frameworks emerged, such as "separate related learning," which became the thematic direction (North Star) for the next phase of the research.
Synthesis & Collaboration
The exploration outputs—paper summaries, article highlights, and analytical notes—were fed into Notebook LM along with the research context or problem statement.
Results:
Notebook LM produced structured summaries and synthetic scripts (formatted like a narrative podcast).
The resulting insights were mapped in Miro as team discussion material.
The synthetic script format helped present ideas more engagingly to non-research stakeholders.
Final outputs included:
Paper insights and summaries mapped in Miro for internal team discussions.
A "podcast-style" narrative to aid stakeholder presentations.
Impact & Reflections
⇢ Impact
Early exploration time was significantly reduced—the team could jump straight into analysis and discussion.
New ideas emerged that would have been difficult to find using traditional methods.
The synthesis process became more collaborative and accessible, especially for cross-functional teams.
