How NewsGuard Uses and Doesn’t Use AI

NewsGuard aims to be transparent in all aspects of our work. Users can see NewsGuard’s ratings process and criteria, corrections policy, staff backgrounds including any potential conflicts of interest, ownership details, and leadership on the NewsGuard website. (All of that is written by us, with no assistance from AI.)

This policy explains the limited ways we use AI in producing our reliability ratings, public reports, Reality Check newsletter, or False Claim Fingerprints—including how we do so with adequate human oversight to ensure accuracy and accountability.

NewsGuard’s Work is Produced by Humans

All of NewsGuard’s journalism is produced by human analysts with journalism training, with multiple layers of editing, fact-checking, and review by our editorial staff. We do not use AI to draft or write our journalistic and analytical work—though, as described below, we do use it for some limited purposes such as to aid in copy editing and translation.

Our articles, our website and source ratings, our “Nutrition Label” reports, our Misinformation Fingerprints, and every other substantive analysis we publish are written by human staff.

Artificial intelligence plays no role in any judgments of how each site performs on NewsGuard’s criteria. Those judgments are made entirely by our human analysts and editors, and the reasoning behind our assessment of each criterion is explained in detail in the Nutrition Label report accompanying each rating.

How We Use AI with Significant Human Oversight

Copy Editing: Our analysts and editors sometimes use AI for basic copy editing of work that has already been drafted by a human analyst. Analysts or editors receive copy edits in the form of bulleted suggestions for changes that the human authoring or editing the piece can either make or not make, depending on their editorial judgment.

Translations: All of NewsGuard’s work is drafted in English and reviewed by our English-speaking senior editors. Once a drafted report, rating, or other piece of journalism is fully drafted and edited, our global editorial team sometimes uses AI to translate that fully edited, human-written work into the relevant languages for each country in which NewsGuard operates. In every case, a native speaker of the relevant language then reviews the translation line by line to confirm that it accurately conveys the original meaning.

Internet Search: NewsGuard analysts sometimes use AI to assist in internet searches for basic research. However, analysts review, verify and are required to cite direct source material in any journalism they produce for NewsGuard, rather than relying on AI summaries of facts.

Human Editing and Review: We never publish work that has been assisted by AI in any way, including content copy edited or translated by AI, without close review by multiple human editors. In addition, in all cases, if NewsGuard determines that a site might fail any of its criteria, our human analysts attempt to contact the site for comment and to ensure that there is not any information they might have missed that could change their assessment. You can learn more about this on our rating process and criteria page.

Potential False Claim Detection: Outside of our writing, we use some automated systems, including ones that incorporate AI, to detect content that may repeat false claims that our team has cataloged in our False Claim Fingerprints database. When the AI-based check determines that content reviewed may convey a false claim, the content is then flagged for a human analyst to review manually to determine whether that is indeed the case.

AI-Generated Site Detection: We use an AI text classifier as one factor in identifying undisclosed AI-Generated content and websites that primarily publish such content. However, our analysts never rely entirely on AI detection tools in assessing content to be AI-generated—pairing any such automated analysis with a human review of the content or website being assessed.

NewsGuard AI: NewsGuard offers its own consumer AI tool, NewsGuard AI, that answers questions on any topic in the news based on articles from 12,000 reliable sources vetted by NewsGuard’s team. That tool uses AI, which is clearly disclosed, to deliver answers to consumers by summarizing reporting from those reliable outlets. The tool also can answer fact-checking questions with access to NewsGuard’s database of 64,000+ examples of false claims spreading online. The tool is built with strict editorial instructions and safeguards aimed to ensure answers are as accurate, evenhanded, transparently sourced, and as factual as is possible in an AI product.