Polling & Analytics
September 2026
Principles for Leveraging AI
Artificial intelligence tools are transforming the modern world at incredible speeds. At Searchlight, we’re eager to leverage this newfound potential and innovate where others have not. Integrating AI into our policy, research, and editorial workflows offers tremendous advantages, even as it poses certain challenges. Searchlight is committed to demonstrating what thoughtful AI adoption looks like – embracing tools that make our work sharper, faster, and more impactful, while maintaining the transparency, rigor, and integrity that our mission depends on.
Permissible use of AI at Searchlight is guided by three main principles:
- Preserve human creativity in ideation and avoid using AI as a source of ideas.
- Maintain quality of information, identify and remedy hallucinations, and fact-check everything.
- Be transparent and contribute to public understanding.
Creativity in Ideation
A major concern when using AI at an organization committed to novel ideas is that AI may constrain the scope of these ideas or bias them toward variations on existing concepts. There’s also a risk that AI tools may not provide the same feedback that humans naturally would, and it may draw policies into structurally unworkable paths. Searchlight mitigates these concerns in a couple of ways.
Practice 1: Human Ideation
New ideas come from people first. Searchlight’s policy professionals approach their work with a wealth of expertise and historical knowledge. They neither lack creative new proposals nor rely on AI for novel ideation. Instead, Searchlight uses AI to research, stress-test, and verify the impact of policies and concepts. Researchers are encouraged to use AI as a constructive thought partner and to leverage the tools to iterate on policy models or legislative impacts in real-time.
Practice 2: Human Check-Ins
As a practice, researchers regularly engage with colleagues, stakeholders, and other experts about ongoing work. Stepping back from AI sessions and engaging with humans about specific policy direction or draft ideas is critical to remaining grounded. What an AI thinks is feasible and what the real world knows to be achievable are often misaligned, and Searchlight researchers know never to lose sight of that reality.
Information Quality
The risks of AI inaccuracy and hallucination are well-documented. Plausible information is not always correct information, and we must prevent the crafting of policy based on false or misleading material. At the same time, AI tools are enormously useful for background research, data aggregation, and other time-intensive tasks. Searchlight employs multiple practices to mitigate the related risks.
Practice 1: Show the Work
Attributing AI-generated work to human authors is not acceptable. Wherever possible, Searchlight notes the origins of AI-generated or AI-reliant information. When AI is leveraged in a material or substantive way, Searchlight’s final products include a description of how the AI was used in the product’s preparation.
Practice 2: Verify Claims
Searchlight uses a robust process to fact-check our work and ensure accuracy in human- and AI-generated outputs. This includes substantiating polling and policy conclusions and verifying that supporting materials are accurate. No work is published without a final, thorough, human review, and humans bear responsibility for product veracity.
Transparency
At any organization that values shared facts and positive impact, showing the work is critical to trust and reputation. In polling and analytics work, Searchlight already practices robust disclosure of sources and methods. So too will we in our deployment of AI.
Practice 1: Attribution
Artificial intelligence is everywhere, and not all tools are created equal. The reputation and reliability of the agent and model are as relevant as the manner in which they’re used. Searchlight discloses exactly what AI platforms and models it leverages in each of its published works.
Practice 2: Reproducibility
In all good policy and analytics work, strong output should endeavor to provide enough context and information to replicate its conclusions. How this looks depends heavily on the work product, and AI makes reproduction particularly challenging, given the non-deterministic nature and evolution of models and agents. For major policy modeling that relies on AI, Searchlight captures and publishes the AI parameters or query traffic, so others can see exactly what questions are asked and assumptions are made.