Wisconsin Watch to test whether AI assistants can cite local reporting

Cartoon · AI-generated
A $35,000 Reynolds Journalism Institute grant will fund an open-source connector to give compatible AI tools access to Wisconsin Watch’s archive, with a public trial due in early 2027.
A $35,000 Reynolds Journalism Institute grant will support a trial of whether AI assistants can use local reporting more accurately and make its origins clear. The project brings together journalism technology network Hacks/Hackers and Wisconsin Watch, a nonprofit investigative newsroom serving communities across Wisconsin. Its central tool will be an open-source Model Context Protocol server, or connector, designed to give compatible AI systems a route into the newsroom’s archive.
The protocol provides a standard way for AI applications to connect with external information, rather than requiring a separate bespoke integration for every service. Anthropic, which introduced MCP, describes a system in which servers expose selected data or capabilities to AI applications. The protocol documentation sets out the same basic architecture, but leaves decisions about what a newsroom makes available, and how an assistant uses it, to the project’s implementation.
For readers, the intended benefit is that an assistant answering questions about Wisconsin issues could draw on relevant Wisconsin Watch reporting and point users to the journalism behind its response. RJI executive director Randy Picht said: “Shutting the door on AI is not the answer to declining search traffic,” adding that the trial would test whether publishers can make material available while setting conditions for its use. The project is an experiment, however, not evidence that a connector will guarantee accurate answers or reliable citations.
The urgency reflects wider uncertainty over how audiences will find news. The Reuters Institute’s 2026 report records publishers’ concern that referrals from search engines and social platforms will weaken, while noting that observed search traffic has fallen in aggregate and that the causes are not fully understood. Its findings do not show that AI alone is responsible. News organisations are nevertheless exploring new routes to audiences, alongside continued investment in original reporting and explanation.
For Wisconsin Watch, the issue extends beyond website readership. Managing editor Jim Malewitz said that people relying on AI summaries might miss the newsroom’s original work, and that lower direct traffic could also affect donations. The organisation says it is funded by individuals, foundations and businesses, as well as partnership work, and that funders do not influence its editorial decisions. Malewitz also sees a potential reader benefit in making the newsroom’s 17-year archive easier to search, including material that may help show how long-running issues have changed.
Hacks/Hackers co-founder Burt Herman said the pilot’s findings and code are intended to be shared with the wider industry. The public test is scheduled to begin in early 2027, with a report expected in the summer. Its results should help establish whether this approach can give local journalism a useful presence in AI-mediated discovery, while leaving open the practical questions of accuracy, sourcing and editorial control.
Inspired by headline at: [1]
Sources by paragraph: - Paragraph 1: [2], [5] - Paragraph 2: [3], [4] - Paragraph 3: [2] - Paragraph 4: [7] - Paragraph 5: [6], [5] - Paragraph 6: [2]
Source: Noah Wire Services · Original