For every news publisher out there who has been worried about the existential risk posed by AI-powered search, the past couple of weeks have brought positive news.

It is no longer a foregone conclusion that Google’s search generative experience (SGE), which summarises results instead of sending readers to news sites, will knock out the top of publishers’ sales funnels. The reason: Google is considering putting in place a paywall for AI-powered results. Users will need to pay to receive this summarisation service and will still have the option of clicking on links if they do not want to pay.

Ah, look at that: Big Tech joins the ranks of the many readers of this newsletter who have already put their faith in paywalls. And for much the same reason: It costs money to provide information that adds value for users. 

In fact, 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 my colleague INMA Researcher-in-Residence Grzegorz Piechota.

Google is not alone in coming up with ways to cover its expenses. Perplexity, another generative AI platform, is planning to help fund the costs of its AI-powered search results by placing ads against them. (Of course, if you read my March 14 newsletter, you already saw this coming.) 

A few publishers have already hedged their bets on search by signing licensing agreements with GenAI platforms to provide training material for their language models. They will receive not only millions of dollars in revenue but also preferential treatment in AI-powered search results, with links back to their sites.

But for every other news publisher out there: Surely greater friction for users on search engines for intelligent results has to be positive. How many readers at the top of your funnel are willing to pay for information or love the user experience with ads? Would they rather pay Google up front for saving them some time — or would they rather click through to your site in the hope of finding what they are looking for without spending a penny? 

Before you heave a sigh of relief, consider this: This decision by search engines to monetise GenAI search is hardly unalloyed good news.

It is a reminder that GenAI is an immature technology. Even the providers of it are still evaluating the costs of compute and of running queries at scale. You will recall ChatGPT had entirely free access when it was publicly launched, and the folks at OpenAI were shocked when the bot gained a million eager users within five days, with each query costing OpenAI a few cents to run. Two months after its launch, OpenAI started charging for a premium tier and limited the number of users accessing its free product during peak hours.

Across the world, every news publisher that I have spoken to over the past three months is using GenAI tools. This means the industry now faces the risk of escalating costs — because whichever GenAI platform is in use, it is almost certainly going to be more expensive. 

Are we ready for this? It is likely that costs will eventually decline as the technology matures, and competition among GenAI providers could also keep a lid on prices — but we may need to ride this out.

Getting search to work for you, even with AI-powered search

By the way, if you’re still worrying about the impact of search generative experience (SGE), you’ll want to hear from Edward Hyatt, director of SEO at The Wall Street Journal. 

Hyatt, who appeared at an INMA Webinar recently, pointed out that AI-powered search is focused on commodity news.

“The question is how to provide something that nobody else can provide. More scoops, more in-depth analysis. Google will provide answers for generic stuff,” Hyatt said.

In other words, understand deeply what it is that makes your publication and its content distinctive.

“While commodity news is easy to summarise for AI — and easy for other publishers to mirror — exclusive and high-quality journalism is not. Focusing on the never-ending parade of trending topics is not the answer to Generative AI and the new landscape of search. It’s the production of meaningful journalism that is audience-first.

“Our strategy is to really focus on core coverage. We will provide something that Google can’t do itself. Trying to shoot for the moon on every single topic and story is not going to play in this new world.”

Google actually wants high-quality content, he said, pointing to criteria used in search engine optimisation (SEO) and adding that Google favours results that provide high-quality, authoritative coverage. 

In other words, our instincts as journalists to write well about topics we know well are perfectly aligned with the future state of the Internet. We simply have to ensure those topics are also aligned with our audience’s interests.

“Focus on your brand and the type of content that your audience — your existing subscription base — and your future audience find valuable,” Hyatt said. “If you have valuable stuff, Google will show that. Our exclusives perform really well.”

Hyatt suggested checking to see which content is most at risk from SGE and then optimising it — but also ensuring the content is not being crawled by GenAI platforms as training data. 

“You want your content to be indexed by Google and other major tech firms, but do you really want to help them build a product for free and get little traffic back in return? Making news costs money. Don’t give it away.”

Building trust with readers is vital, as are using images in search and possibly changing the format of an article, say, from text to video, and creating a subscription journey for the reader that works well with search.

He also spoke of search intent: “You have to understand what users really want when they go into search and meet them where they are.” 

Most SGE features include the bar at the bottom that encourages readers to ask a follow up question, which can be valuable in understanding reader intent. Readers could interact with these questions and provide “great fodder for understanding what audiences care about.” 

Hyatt was speaking 10 days before Yahoo and Perplexity both said they will look at search intent, in real time, when evaluating whether to present a user with AI-powered search result summaries or an old-fashioned list of blue links.

Dates for the calendar 

Friday, April 26: The GenAI workshop at the INMA World Congress of News Media promises to be exciting, and there is a strong vein of AI running through the entire conference. I hope to see you in London! 

Thursday, May 16: Presenting our inaugural Generative AI Master Class. We have a dynamite line-up of speakers with practical, concrete use cases and best practices. 

Worthwhile links

• Transparency and trust: How transparent should we be with our audience when using AI? A group of respected Swedish publishers got together and agreed on a set of guidelines.

• GenAI licensing: The platforms really, really need your content as training material.  

• GenAI for local news: Five tips.

• GenAI for audio: The New York Times will soon offer most articles via automated voice.

• GenAI and search: What do people actually search for on Google? 

• GenAI for fact-checking: Looks like it works quite well.

• GenAI deals: A handy guide to who in the news business is making deals with GenAI companies — and who is suing.

• GenAI for marketing: The BBC said it would not use GenAI in its Doctor Who promotions again after receiving complaints.

• GenAI for marketing II: About three-quarters of CMOs believe AI can generate deep detail into customer behaviour and can facilitate real-time decision making as well as predict the effectiveness of creative imagery and messaging. Yet, as many are completely unprepared for this future.

• Beware of chatbots: GenAI can be remarkably persuasive — especially when tailoring content by crafting messages that resonate with users’ specific background and demographic.

• Beware of chatbots II: This one randomly gives out incorrect information, despite having an impressive pedigree. 

• GenAI for advertising I: Reddit uses GenAI-driven headline generation, asset cropping, and bidding and budget-management tools.

• GenAI for advertising II: Looks like media planning and buying roles at ad agencies will be decimated, according toMartin Sorrell.

• GenAI and the end of learning languages: People may no longer need to learn a second language.

A non-AI diversion

I stumbled upon this fascinating, timely course and thought that you — the logical, analytical readers of my newsletter — would find the case studies interesting as well. Enjoy it with some machine learning-driven beer, of course.

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.

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