Is it possible for news organizations to develop an AI chatbot that is considered reliable?

Is it possible for news organizations to develop an AI chatbot

News organizations have enthusiastically embraced artificial intelligence, employing generative AI tools to create unremarkable travel guides, inaccurate film blogs, and clickbait-style explainers for the purpose of increasing ad revenue without expanding their writing staff.

Currently, a cluster of tech-focused media outlets is striving to integrate generative AI into their websites, although readers will not come across articles attributed to machines anytime soon. On August 1st, an AI chatbot feature was introduced to Macworld, PCWorld, Tech Advisor, and TechHive. This new addition promises readers the ability to “receive answers to their technology-related inquiries from AI, based exclusively on content authored by our experts.”

The AI chatbot, named Smart Answers, is visible throughout nearly all articles and on the homepages of these websites, all of which are under the ownership of media/marketing company Foundry. Smart Answers has been trained solely on the collection of English-language articles from these four sites, omitting sponsored content and promotional posts. User interaction mirrors that of consumer tools such as ChatGPT: users input a question, and Smart Answers generates a response. Alternatively, users can choose from a list of frequently asked questions (FAQs), which are AI-generated but grounded in actual user queries and interactions. The responses from Smart Answers include links to the original articles from which the information was drawn.

The underlying objective, as stated by the leadership, is to aid readers who seek specific information within the vast expanse of content.

Neil Bennett, the Global Director of Product and Data at Foundry, commented, “We possess an extensive repository of content. However, there are instances where you don’t want to read a lengthy 2,000-word piece of top-notch journalism – sometimes you just need a specific piece of information.”

The development of Smart Answers spanned approximately five months before its launch, involving rounds of testing with editorial staff for two to three months. An initial group of editorial personnel evaluated early iterations of the tool and assessed responses until they were content with the accuracy.

Marie Black, the Editorial Director at Foundry, explained, “Our testing primarily revolved around ensuring that the provided answers are genuinely reliable, equivalent to what an editor would provide.”

The intention behind Smart Answers is to mitigate a common issue observed with other AI tools: the tendency to make mistakes, reference non-existent sources, or fabricate plausible yet false information. Smart Answers is exclusively trained on content that meets the standards for publication by reputable sources. Nevertheless, claiming that the chatbot could deliver responses at an editor-level is a substantial assertion.

The effectiveness of Smart Answers falters when confronted with seemingly simple queries. For example, when asked, “When was the most recent release of the iPod Nano?” the chatbot inaccurately stated that it was both released and discontinued in 2022. Some days later, the response had changed to, “I lack sufficient information to address this question.”

This type of response is frequent from the chatbot; for instance, it doesn’t know who the CEO of TikTok is. This limitation could reflect the scope of coverage by the Foundry outlets. While this limitation might be viewed as a drawback, it also serves as an advantage—users of the tool are likely to have remaining questions that they need to explore elsewhere. However, in these scenarios, the chatbot refrains from generating fabricated responses.

Regarding the incorrect iPod Nano response, Black speculated that the chatbot might have referenced outdated stories about iPods that hadn’t been updated in years. (The iPod Nano was discontinued in 2017.) To address such errors, Foundry relies on user feedback regarding the usefulness of the responses. If issues are identified by readers or staff, they are relayed to Miso.AI, the AI company developing Foundry’s chatbot.

Beyond its function of condensing information from complete articles, Smart Answers also serves the purpose of generating affiliate revenue. Beneath the generated responses and above the relevant story links, the tool includes purchasing options for items it deems pertinent to the user’s search. For instance, a question regarding iPhone battery life might yield shopping links to several retailers offering phones.

Similar to Foundry, many outlets monetize their content through affiliate marketing ads, earning revenue each time a reader makes a purchase using their provided link. In this manner, Smart Answers becomes a new avenue for accommodating these ads, tailored to align with the questions posed by users.

Black affirmed, “There are links included for purchasing, which we consider to be part of the reader service. If a user inquires about a product, they are likely interested in purchasing it.”

Sustaining user engagement on the site until a purchase is made, whether through AI-generated summaries or comprehensive reviews, has gained added importance as generative AI takes center stage in Google Search. Publishers, particularly those who create product reviews or recommendation lists, stand to lose considerable traffic (and consequently ad revenue) as Google’s AI-driven Search Generative Experience places excerpts from articles above the sources of information.

Black remarked, “I believe it’s unrealistic to anticipate that Smart Answers will directly replace the traffic we are losing from Google Search.” The aspiration is that it will deepen engagement with readers who are already navigating the site.

Amid criticism concerning the implementation of AI initiatives, other media companies like G/O Media have redoubled their commitment to AI content generation, leading to discontent among staff and prompting unionization efforts. Black asserts that Macworld, PCWorld, Tech Advisor, and TechHive will not be publishing AI-generated articles and that the chatbot could prove valuable for the reporting process. User queries or unsatisfactory responses could alert reporters and editors to areas where further coverage is required.

Black clarified, “This is in no way a threat to our journalists. It simply cannot exist without our journalists crafting in-depth content that addresses every possible reader question.”