Person of the Year is one of our most enduring franchises. Every year, we aim to name a Person of the Year—the person, group, or concept that had the most influence on the world during the previous 12 months. With this year’s release, we wanted to do something that reflected a recent change in how we interact with information and news.
Our goal was to create an AI chatbot specifically designed to answer questions about TIME’s Person of the Year—past or present—based on a curated body of verified TIME articles, other trusted sources, and the bot’s built-in general knowledge. This chatbot is tailored to provide focused, reliable, and accurate responses while maintaining a safe and professional tone. It is engineered to prevent users from steering the conversation toward unrelated, controversial, or potentially biased topics.
TIME’s AI chatbot represents a significant step forward in using AI to enhance journalism. With its guardrails and focused content retrieval, the chatbot aims to deliver a safe, accurate, and engaging experience for readers. This tool exemplifies how innovation can support and amplify trusted reporting.
Here’s a closer look at how it works.
What is TIME's AI chatbot?
It’s an interactive tool designed to help readers explore TIME’s Person of the Year package. Powered by verified articles, the chatbot provides accurate, in-depth information about past and present Persons of the Year. Unlike general AI tools, it’s purpose-built for meaningful, focused conversations about a single aspect of TIME’s journalism. It combines accuracy, depth, and interactivity in a way that aligns with TIME’s commitment to trustworthy reporting. By making Person of the Year coverage more interactive, the chatbot deepens reader engagement and accessibility.
How do we ensure the chatbot provides accurate information?
The system’s responses are rooted in vetted content from TIME articles and other trusted sources, all in the service of adhering to rigorous journalistic standards. The goal is to avoid speculation and keep conversations within the scope of the chatbot’s expertise.
How does the chatbot handle misinformation?
By limiting its source material to approved articles and sources, the chatbot minimizes the risk of spreading inaccurate or unverified information. If a question strays into controversial territory, it redirects the conversation to safer, fact-based topics. It politely shifts the conversation back to its core purpose—exploring TIME’s Person of the Year package.
What kind of questions can readers ask?
Readers can ask about a number of topics, including:
- Background on people previously named Person of the Year
- Historical context from TIME’s vast archives
- Details and insights from related TIMEcoverage
Where does the chatbot’s information come from?
Its knowledge is drawn from the following:
- TIME’s Person of the Year articles
- Related TIME reporting
- Historical archives
- General knowledge embedded within the large language model (LLM)
What safeguards are in place?
The chatbot employs a dual-layer protection system:
- Guardrails (safety score, rationale, alternative response) ensure inputs are safe and relevant. See more on guardrails below.
- Focused content limits responses to verified TIME material.
How does it work?
The system employs a combination of three techniques:
- Prompting establishes a framework for how the chatbot should operate throughout user interactions.
- Retrieval ensures the chatbot draws upon vetted TIME articles, either directly or through generated summaries.
- Guardrailing classifies user inputs as safe or unsafe, redirecting responses as necessary to maintain focus and integrity.
The chatbot system integrates several components, as illustrated below:
- User Input Processing: Each user message, along with the conversation history, is analyzed by two models: the primary chatbot and the guardrail model.
- Guardrail Evaluation: The guardrail model determines whether the message is safe to process. If deemed safe, the primary chatbot generates a response. If not, the guardrail model provides an alternative response tailored to redirect or address the issue appropriately.
Why use two models?
A single model handling all tasks would face challenges due to prompt complexity. Large language models (LLMs) struggle to follow intricate, multi-layered instructions consistently. By dividing responsibilities between the primary chatbot and the guardrail model, the system maintains higher precision and reliability.
IMAGE HERE
Primary Chatbot Setup
The primary chatbot, based on the GPT 4o Mini model, is optimized to answer questions about TIME’s Person of the Year based on curated content. The system prompt directs the chatbot to rely on the main article and supplementary articles when crafting its responses.
In addition to TIME articles, the chatbot uses summarization and tool calling to retrieve relevant information dynamically:
- Summarization: Key points from articles are condensed and included in the model’s context for quick reference.
- Tool Calling: Depending on the conversation, the chatbot can retrieve additional details by referencing specific articles through synthetically generated summaries. This enables precise and context-aware responses while minimizing information overload.
Guardrail Model Setup
The guardrail model, also based on GPT 4o Mini, acts as a safety filter. Its system prompt defines the boundaries for acceptable inputs, ensuring conversations remain relevant and appropriate. For every user message, it outputs:
- A safety score indicating the risk level of the input,
- A rationale explaining why the input is considered safe or unsafe, and
- An optional alternative response for unsafe messages.
Based on the safety score, the system either routes the input to the primary chatbot or delivers the guardrail’s alternative response.
More Must-Reads from TIME
- Caitlin Clark Is TIME's 2024 Athlete of the Year
- Where Trump 2.0 Will Differ From 1.0
- Is Intermittent Fasting Good or Bad for You?
- The 100 Must-Read Books of 2024
- Column: If Optimism Feels Ridiculous Now, Try Hope
- The Future of Climate Action Is Trade Policy
- FX’s Say Nothing Is the Must-Watch Political Thriller of 2024
- Merle Bombardieri Is Helping People Make the Baby Decision
Contact us at letters@time.com