Imagine asking Joanna Stern your exact iPhone question and getting an instant answer in her voice. Or navigating complex tax codes through conversation with The Wall Street Journal’s expert tax reporters. Or uncovering secrets buried in 80,000 government documents within hours. The Journal isn’t just adopting AI; we’re using it to change the way readers digest the news and to deepen our investigative power, all while upholding rigorous journalistic standards. Our North Star: generative AI should enhance, rather than replace, human-powered journalism.
We created Joannabot and Lars, two specialized AI assistants that transformed static reporting into interactive experiences for readers. Joannabot, launched alongside our iPhone 16 coverage, allowed readers to engage directly with Joanna’s technology expertise (and quirky personality) through a conversational interface. Rather than a one-size-fits-all iPhone review, readers could ask specific questions about features, upgrades and comparisons, and receive responses grounded in Joanna’s professional assessments and verified iPhone specifications.
Lars, our tax assistant, similarly transformed our traditional tax guide into an interactive resource, helping readers navigate complex tax questions by drawing from more than 1,200 WSJ tax articles and IRS publications. Named after WSJ tax writers Laura Saunders, Ashlea Ebeling and Richard Rubin, Lars gave personalized guidance during the stressful tax season. Alongside its helpful guidance about tax code changes, deduction eligibility and investment taxes, Lars also included a robust citation system that showed readers exactly which WSJ articles or IRS documents informed each answer.
Behind the scenes, our newsroom has implemented powerful AI tools that accelerate and enhance investigative journalism. When the government released thousands of JFK assassination documents in March, our data team rapidly deployed an internal tool that processed more than 80,000 pages of poorly-scanned documents. This tool generated initial summaries to guide reporting focus, enabled semantic search across the full document set and tracked connections between key figures and events. The bot allowed a small team to discover significant findings within hours, including revelations about CIA monitoring of Lee Harvey Oswald before the assassination and previously undisclosed diplomatic communications about Cuba.
Our approach to AI innovation is based on three core principles:
The technical infrastructure supporting our AI efforts combines large language models with custom knowledge bases, sophisticated prompt engineering and robust monitoring systems. For Lars, we used Google’s Discovery Engine and Gemini 2.0 models, which allowed for inline citations, ensuring transparency for readers while also improving accuracy of the responses. Our public deployments include clear guardrails through carefully crafted system instructions and conversation wrappers to ensure AI responses remain within appropriate journalistic boundaries while maintaining a consistent voice.
Most importantly, our AI innovation strategy has delivered meaningful results—enhancing reader engagement, accelerating investigative stories and creating new formats for delivering WSJ journalism while maintaining the human expertise that defines the Journal brand.
