Each of the AI newsroom initiatives are continuously being monitored for editor efficiency (time saved at an individual level), quantifiable outcomes on editorial quality (SEO and content parameters, generated traffic and/or subscriptions sold), and impact created in terms of scale.
“Only when our initiatives impact all or some of these metrics in a positive way, do we think about scaling them,” said Aviral Mathur, Head of Data Science, at our recent AI Summit in Bengaluru.
The impact of the AI Newsroom, in partnership with the Google News Initiative, is evident in its significant contributions to traffic growth. Automated content is now instrumental in generating 3-4 million pageviews on the core websites, and doubling the traffic on the publisher’s niche websites.
Notably, in certain key geographies, for which components of this product have been leveraged, HT Media has seen market share gains of 42 percent, said Mathur.
HT’s three AI buckets
Mathur classified HT’s AI initiatives into three categories:
- Automate content: Complete end-to-end automation with minimal editorial intervention. This includes stock market liveblogs, gold and silver commodity blogs, and weather/air pollution blogs (HT AlphaInsight). This automation saved 20 percent of the editorial bandwidth.
- Assist editorial processes: Here, AI is merely a part of the overall process. This has been used for content quality monitoring, text summarisation, SEO parameters – keywords, backlinks (HT NewsNet), and has resulted in a 15 percent increase in page views.
- Augment editorial capabilities: This bucket includes HT Story Selection (HT Relevare), language translation, interview transcripts, large document summarisation. This has helped increase story traffic on the core sites by 15-20 percent.
HT Relevare
Inside the AI buckets
HT Relevare system – Story selection algorithm: The HT Relevare System is an advanced story selection algorithm designed to optimise editorial efficiency by prioritising high-yield stories. It is structured around three key components:
- Predictive content selection: This AI-driven system boosts content management efficiency by scanning, sorting, and scoring incoming stories from sources like Google Trends, third-party tools, HT keywords, and competitors. Using HT’s Relevare System, updated every 15 minutes, it prioritises relevant content for editorial and SEO teams. Higher-scoring stories attract more traffic.
- AI-powered content generation: This component is at the heart of the AI newsroom’s capability to produce innovative content, tailored to the interests of readers. The newsroom has been experimenting with GenAI-powered content for live blogs, web stories, company results, and affiliate content. It has also helped in producing story metadata (headlines, keywords, URLs, summaries, etc.). All generated content is actively vetted by the editorial team, with multiple checks in place to identify and address any issues related to fake news and misinformation from LLMs.
- CMS workflow integrations: The third pillar of the AI newsroom focuses on seamlessly integrating the AI-driven content with the existing CMS. “This is important in ensuring that the generated automated content is easily accessible and usable within the standard workflows of the editorial team,” he said. It also has checks within the CMS to flag content that does not match the brand content standards.
Building on the system’s success, they plan to adapt it for advertising by recommending content to attract targeted traffic, like mutual funds or automotives. They also aim to use the algorithm to identify content that can boost subscription acquisition, expanding its utility beyond just increasing traffic.
See also: HT Digital sees 20% boost in page views with AI
HT NewsNet System – Content quality monitoring system: Mathur explained the integration of various views to monitor content quality and provide recommendations to editors. The Chief Content Officer view monitors daily content quality trends across HT sites and desks. The Content Desk Head view tracks hourly content quality trends, allowing for immediate action. The Author Performance view evaluates the performance of authors within desks on a weekly and monthly basis, offering actionable feedback to improve performance. These quality parameters are embedded within HT’s CMS as GenAI-based recommendations for journalists, and are updated hourly.
The different views within the content quality monitoring system
HT AlphaInsight System – Monitor stock market trends and generate insights: This GenAI based algorithm collects stock news from both internal and competitor sources and extracts themes and sentiment. Additionally, it leverages the Refinitive Stock Market Database for trend identification, analysing changes every 15 minutes, daily, and monthly. Automated predefined thresholds help filter and prioritise significant trends. This data feeds into the Insight Engine, which processes the information and generates content using large language models (LLMs).
“Next steps include automating live trend analysis, identification of bullish and bearish stock trends, news theme identification by means of external and internal data, and stock sentiment analysis,” he said.
Tech maturity, constant testing: Lessons learned
Mathur also touched upon the importance of recognising when AI is not sufficiently developed for certain use cases and discontinuing those efforts to save time. In addition, he noted a few more learnings that he recommended to other publishers starting to work with AI:
Building trust and transparency: Constant communication and dialogue between journalists and tech talent is essential.
Think AI products, not algorithms: Focus on the end product to increase adoption in the newsroom.
Experimentation: Continuously experiment to adapt to the dynamic news industry. Prioritise successful experiments.
Think like a VC: Launch multiple experiments knowing that only some will work. Focus on those with the potential for high returns.