I spent a few fascinating days at the INMA Media Subscriptions Summit in New York recently, chatting with attendees about generative AI. Maybe it’s just because their backgrounds are in consumer revenue, but here’s something that popped up a few times: the question of money.

The keynote speaker at the summit was Professor Rajkumar Venkatesan, who specialises in marketing with technology at the University of Virginia. He mentioned AI-powered search in his presentation, which almost every publisher I have spoken to over the past few months has mentioned as a topic of great concern. 

The reason is simple: As Google rolls out search generative experience (SGE), it will no longer send much traffic to publishers’ sites, giving consumers a bullet-point summary of what they are searching for instead.

This means that publishers may need to rethink their business models — whether they are predicated on ad impressions, affiliate revenue, or subscriptions. This has the potential to take out one-third of news publishers’ business, according to my brilliant colleague Greg Piechota, who organised the summit as is the INMA Readers First Initiative lead.

In our conversation after his presentation, Venkatesan had a lot to say about SGE:

1. Google will need to monetise AI-powered search. “There’s no way they can keep providing it for free. It costs money to run search queries with GenAI at scale,” Venkatesan said. It costs Google 10 times more to generate a list of summarised results than it does to return a typical, old-fashioned page of results with links, according to Piechota. (He also points out that Google brought in US$31 billion in revenue in 2023 from its ad network on publishers’ Web sites.)

This means Google could charge the consumer for SGE results, since it is providing them with a time-saving service by summarising results for them. Will customers be willing to pay for that? Or perhaps they are simply used to treating the Internet as “free and god-given,” as the chief product officer at a large European publisher told me.

The publisher points to a different trade-off: “Do you want good results or do you want cheap results?”

Alternatively, Google could charge publishers to be referred to and linked to in SGE results. Charging publishers could undermine trust in results — or, if Google links only to reputable publishers who can afford to pay, maybe it will not. 

So, instead of generating revenue for publishers, search could become another expense.

2. Remove the middleman. Stop relying on Google and other platforms to send audiences your way. GenAI makes it easier to generate content and build that direct relationship with the reader, whether it is a low-tech solution like newsletters or something relatively new like a chatbot that answers user queries. Publishers need to remember that focusing on their brand is worthwhile.

Venkatesan cited the travel industry as an example. Consumers realised they could get better deals and perks by going directly to a hotel Web site than by going through an intermediary travel-booking Web site. 

It is now up to news publishers to show consumers the value of their content by repeatedly surfacing what consumers actually find valuable. This means investing in data to really understand what consumers are looking for and find worth paying for.

3. Join forces and negotiate. News publishers should block GenAI platforms from simply taking their content and using it as training material for their models, Venkatesan said. (Only about half of the world’s publications do this at present). They should band together and force tech companies to pay for it, he said, as Reddit and Axel Springer did, and as News Corp is doing.

OpenAI itself has said that it is impossible to train models without copyrighted material. The news industry should cash in on that. 

QUICK ASK: Could you please take 60 seconds to fill in a two-question survey on GenAI tools? Your answers will help us serve INMA members better.

GenAI for commerce: Extracting value from chat products

The question of ROI is increasingly popping up in connection with another popular GenAI product: chat. 

Many news organisations are busy building chat products to see if they can open the door to deeper user engagement. 

For example, The San Francisco Chronicle recently launched a chatbot that provides restaurant recommendations after combing through nearly 1,000 different restaurants in the newspaper’s database. The bot has been built off hundreds of reviews in the Chronicle’s Top Restaurant lists, all written and vetted by its food and wine writers and critics, according to Sarah Feldberg, The Chronicle’s deputy director of product and strategy.

Forbes is testing a chat product called Adelaide (named after the wife of founder B.C. Forbes), where the user can input a query and get a summary response as well as a list of related articles.

Skift, a niche travel and hospitality publishing site, does something similar. Ask Skift was trained on 11 years of content.

Germany’s Ippen Media is building a chatbot that appears on the article page and lets the reader ask further questions about the topics mentioned in the story. 

In India, both Times Internet and The Hindustan Times are also experimenting with chat, with Times Internet clearly building a product that would allow it to generate more revenue.

The BBC is also exploring “the potential for chatbots to provide interactive and tailored learning on BBC Bitesize,” its educational service for children (similar to Khan Academy’s Khanmigo effort, perhaps), an initiative that any publisher with a rich trove of evergreen content could emulate. 

But is it worth the effort? This is the question many publishers are asking. It costs money to run GenAI search queries. How does one monetise chatbots?

One option is to place advertisements inside chatbots to generate revenue. It is an option that both Google and Microsoft are exploring. 

Another option would be to initiate the subscription flow from chat itself. 

How about building a chat product that is so powerful, valuable, and useful that it becomes a premium product in itself worth charging for? It could serve your most valuable niche of customers and provide specialised information that you have already published and which your data already shows users are willing to pay for. 

We already have a news-adjacent use case in the legal services bot that Thomson Reuters has launched. Bloomberg also offers chatbots as a service to its clients. 

Could we draw inspiration from these B2B cases? Could we not create a product that draws on publishers’ archives and expertise to create, and perhaps even help us book, an entire vacation, where we specify our budget and the travelers’ interests?

Or one where you can share a photograph of what is inside your fridge and ask the bot for customised recipes that would feed a family of five, where one diner is a senior citizen and two are teenage boys with dietary restrictions?

Or one that helps a young professional draw up a comprehensive list of everything she needs to do ahead of a tax-filing deadline, based on hundreds of personal finance columns and her own financial goals? 

GenAI makes all of this possible, especially as techniques to minimise hallucinations make rapid strides. Most news sites have mediocre search experiences. This could solve that problem — and add in a layer of personalisation, which was a recurring theme at the INMA Media Subscriptions Summit. 

The other gold mine in chat is data. Users willingly engage with the bot and share information about themselves. It’s often worth spending money to gather data (I’ll be writing more about this soon).

Other industries, such as retail or banking, have already discovered that chat fits nicely with a broader commercial strategy, where chat can serve as a tool for lead generation, payments, and retention. Publishers like Rede Gazeta in Brazil are already using chat to supplement the effectiveness of their customer service offering, allowing humans to focus on tougher, more specialised questions that the bot cannot answer. 

All of this translates into money down the line.

Dates for the calendar 

• Webinar on March 22: Future-proof your business in the AI-powered search era. Please join speakers from The Wall Street Journal and Hearst Newspapers to gain a nuanced understanding of the role search plays in building a successful subscription business and ensure your news content differentiates itself while maintaining its reputation for quality and accuracy. 
• Webinar on April 3: A glimpse behind the FT’s first GenAI product. Join Lindsey Jayne, chief product officer at the Financial Times, to learn how the FT is planning on meeting its audience where they are headed. Register here.

Worthwhile links

• Voicebots, AI anchors, automated coverage … your peers are doing cool stuff. Check out the finalists in the INMA Global Media Awards’ AI categories. 
• Strange but true: Your mum was right: You do get better results when you ask nicely — even with LLMs.
• Strange but true II: And you also get better results if you pretend you are in Star Trek.
• Never mind: Just let the machine handle the prompts.
• How do LLMs actually learn? We’re not quite sure. But it looks like forgetting helps LLMs learn.
• Beware the hype: Don’t believe everything you read about productivity gains from GenAI.
• First Reddit, now Automattic: WordPress and Tumblr are selling their user-generated content to OpenAI and Midjourney as training material.
• Dependence on platforms: Google is paying smaller newsrooms to use its tools. Does that simply solidify its grip over them? 
• Music from text? Reporters, this is not what the editor means when he says, “Make it sing.” 

A non-AI diversion

Who among us has not dreamt of having a house with a secret room, accessible only through a cleverly concealed door?

About this newsletter

Today’s newsletter is written by Sonali Verma, based in Toronto, and lead for the INMA Generative AI Initiative. Sonali will share research, case studies, and thought leadership on the topic of generative AI and how it relates to all areas of news media.

This newsletter is a public face of the Generative AI Initiative by INMA, outlined here. E-mail Sonali at sonali.verma@inma.org or connect with her on INMA’s Slack channel with thoughts, suggestions, and questions.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .