In the past few weeks, News Corp, the Financial Times and Dotdash Meredith became the latest news organizations to strike licensing deals with OpenAI, while Axel Springer and Informa teamed up with Microsoft.

Under the terms of the FT deal, financial details of which are unknown, OpenAI will get access to content from both sides of the FT’s paywall to help train the tech firm’s generative AI technology. Once trained, OpenAI’s flagship chatbot, ChatGPT, will be able to answer questions with summaries from FT journalism and will provide links back to the original source.

A day after the FT and Axel Springer deals were announced, news broke that eight newspapers owned by subsidiaries of Alden Global Capital, including the Chicago Tribune, New York Daily News, and Denver Post, were following the New York Times in suing OpenAI and Microsoft for infringing their copyright when training their respective generative AI products, ChatGPT and Copilot. The lawsuit also accuses the tech firms of making unauthorized use of the newspapers’ trademarks, removing copyright management information, and causing reputational damage by attributing “hallucinated” answers to the newspapers’ reporting.

These developments typify the latest chapter in a fraught relationship between platforms and publishers, where the big question for news organizations – or some might argue, the chosen few that are given a choice – is: Deal or no deal? (For those not invited to the table, it’s more a case of: Deal with it.)

Comparing previous chapters is not a perfect case of comparing oranges and oranges;  however, aspects of the current moment contain echoes of the recent past.

Around a decade ago, having faced up to the reality that social platforms had become an unavoidable part of the information ecosystem, news organizations were weighing the extent to which they would shift their business models and practices to align with tech platforms’ promises of jam tomorrow. A few years later, the chosen few were deciding whether to accept offers to be paid launch partners on new products like Facebook Live and Instant Articles. More recently, the question has become whether to take money, credits, or training sessions offered as part of big-money journalism initiatives like Google’s Digital News Initiative and the Facebook Journalism Project.

OpenAI’s licensing deal with the FT was the fifth such arrangement it has struck since ChatGPT burst onto the scene in November 2022. The first came last July when it reached an agreement with the Associated Press. Since then similar deals have been done with Axel Springer, Le Monde and Prisa Media.

But licensing deals are not the only way tech firms have forged financial relationships with the journalism world. 

Hot on the heels of its licensing deal with the AP, OpenAI announced a partnership with the American Journalism Project for which it would pay $5 million in cash and up to $5 million in OpenAI API credits. Five months later, just before the turn of the year, AJP distributed 13 grants of $25,000-200,000 via a Product & AI Studio created as part of the OpenAI partnership.

In February, Microsoft announced five partnerships that covered news organizations (Semafor, and the GroundTruth Project, which places journalists in local newsrooms through its Report for America program), education (Craig Newmark School of Journalism at the City University of New York), industry organizations (Online News Association), and software development (Nota).

Like Microsoft, OpenAI has also delved into the education sector, awarding $395,000 to NYU’s Ethics and Journalism Initiative.

According to news reports, the big players – including Apple, Amazon, and Adobe, as well as OpenAI, Microsoft, and Google –  have engaged numerous news organizations in discussions about potential AI partnerships. Shortly after the FT announced its deal with OpenAI, The Information reported that Google is set to pay News Corp $5-6 million per year to develop AI content as part of an existing deal through the Google News Initiative. (A News Corp spokesperson told Reuters, “We absolutely do not have an AI content licensing deal with Google, though we do have a number of partnerships with Google across our businesses.”)

We’ve collated reports of confirmed deals and reported discussions in a database that we will continue to update as partnerships are pursued and formed.

The current flurry of activity raises a number of questions – many of which feel like callbacks to previous episodes, and were not satisfactorily answered the first time around.

To what extent are powerful technology companies picking winners (again)? On what basis are news partners chosen? How are “high quality”/”reliable”/”trustworthy” sources determined?

The pitch deck for OpenAI’s Preferred Publisher Program – a version of which was leaked to Adweek – describes the company’s pursuit of “select, high-quality editorial partners.” The basis upon which it is deciding which news outlets are “high-quality” is unknown.

When the latest deals have been announced, it’s often been left to OpenAI’s news partners to self-describe their journalistic credentials, such as the “fact-based, nonpartisan news content” supplied by the AP, “recent and authoritative content” from Axel Springer, “accurate, verified, balanced news stories” and “reliable information” from Le Monde, “in-depth, quality journalism” from Prisa Media, or “reliable sources” of the FT. 

For its part, OpenAI has tended to shape its public narrative around the great benefits it claims these partnerships will bring to the public and the news industry as a whole. The deal with the FT, for example, is said to be “about finding creative and productive ways for AI to empower news organizations and journalists, and enrich the ChatGPT experience with real-time, world-class journalism for millions of people around the world.”

The power imbalance between platforms and publishers has long led to accusations the former is able to “pick winners” – both through product choices and partnerships.

In the Tow Center’s initial research, published eight years ago, a newsroom manager from a local outlet said, “The [social media] algorithms, because they favor scale and reach, they’re naturally going to favor national and international stories, and so local journalism gets deprioritized. I think we do run the risk of selecting winners in this game.” The following year, a digital media executive with experience at major metro newspapers, said, “I feel as if we are collateral damage in the war between these platforms. They’ll give some publishers a chance to play, but not others. They’ll give favorable rates and treatments to some and not to others. They are already picking winners.”

Perks for so-called winners have fluctuated, but have included: receiving a flood of traffic or engagement on account of producing output that chimed with the whims of a platform’s algorithm, being invited to the table to discuss product development or partake in testing, receiving financial incentives to partner on new or expanding publishing products, and enjoying priority access to platform personnel. More recently, they have involved direct payments or been selected to participate in programs with perks such as platform credits, training programs, and coaching.

Indeed, an FT report about the US launch of the Facebook News tab in October 2019 was headlined, “Facebook picks winners and losers ahead of news page launch: Social media group offers millions to Bloomberg and Dow Jones but nothing to Reuters or AP.” (The FT was among the winners once Facebook News made its way to the UK.)

One constant has been a lack of transparency around how and why certain publishers have seemingly had the ear of platform companies while others have felt shut out.

There’s no reason to think we won’t be asking the same questions again.

How much are these deals worth to news organizations? Alternatively, when it comes to licensing deals for journalism used to train LLMs, what value are technology companies placing on news archives? (Or, from the other side, what do news outlets consider an acceptable valuation of their archives? And to what extent is the lure of short-term income so strong that they are willing to accept a sum below their valuation?) How do they differ from one organization to the next?

Thomson Reuters CEO Steve Hasker has said that “there appears to be a market price evolving” for news content used to train LLMs. What this is remains unknown, although the $250 million OpenAI is reportedly paying News Corp in cash and credits over the next five years suggests deals with the most influential organizations come at a particular premium It also raises interesting questions about the implications for OpenAI’s previous deals, given earlier reporting that “Associated Press pact with OpenAI gives it the right to reset terms if another publisher gets a better deal”

Platforms have long been reluctant to disclose details about the checks they’re signing for chosen news organizations. The recipients, for their part, are usually subject to non-disclosure agreements.

Details about the AI partnerships struck to date are similarly scarce. Away from deals with news organizations, we know from AP reporting that Google’s deal with Reddit is valued at “roughly $60 million.” NYU disclosed that a grant from OpenAI for its Ethics and Journalism Initiative was worth $395,000.

According to The Information, some news organizations have been underwhelmed by offers from OpenAI starting at $1 million.

Bigger numbers are occasionally touted, such as the “$5 million+ partnership” between OpenAI and the American Journalism Project. But once that money started to be funneled to news organizations, the figures got smaller and the specifics became more vague. A December 2023 announcement states that 13 named “portfolio organizations” were awarded “one- to two-year grants, ranging from $25,000 to $200,000.”

Is any level of exclusivity built into such deals, or will news organizations have carte blanche to sell their archives to whichever technology firms are willing to write them a check?

Between partnerships on tools/technology and/or licensing deals, are the next few years set to become an era of Microsoft/OpenAI newsrooms and Google newsrooms? If so, what are the implications of this?

What impact, if anything, will this have on, say, the FT’s relationship with Google?

The Financial Times, and its consultancy arm, FT Strategies, have a long history of partnering with Google.

The FT was a launch partner for the Google-developed Accelerated Mobile Pages (AMP), and Subscribe with Google. The pair also joined forces for the Hidden Cities project in 2015, which led to Hidden Cities: Rio, a VR collaboration that marked the FT’s “first major virtual reality project.”

FT Strategies has been a prominent part of the tech giant’s European journalism program, the Digital News Initiative, and its US equivalent, the Google News Initiative.

Indeed, Google highlighted the partnership on the first page of written evidence to the UK Parliament’s Digital, Culture, Media and Sport Select Committee inquiry into the sustainability of local journalism: “From increased revenue through our search engine and advertising tools to financial support for bespoke consultancy services (such as our Subs Lab with FT Strategies, below) Google aims to provide local news publishers with the security and skills to adapt their business models and grow their revenues.”

More recently, FT Strategies has partnered with GNI for AI Foundation and AI Launchpad programs in Europe, the Middle East, and Africa. A deck for the former states, “Delivered in partnership with the Google News Initiative, the AI Foundation programme helps publishers who wish to explore and start acting on the opportunity that AI brings, built upon the FT’s deep understanding of publishing and Google’s expertise in AI.”

In late April, Axel Springer provided an early model for cosying up to multiple partners, announcing a new partnership with Microsoft to supplement the one it formed with OpenAI last year.

When, where, and how will news organizations disclose relationships with the technology companies from which they have received payments?

Local outlet The City used OpenAI’s ChatGPT to map the New York neighborhoods covered in 2,750 of its articles.

Nowhere in the article is it disclosed that the City received one of 13 grants of $25,000-200,000 from the American Journalism Project’s Product & AI Studio, which was created as part of AJP’s partnership with OpenAI, for which AJP reportedly received $5 million in cash and up to $5 million in OpenAI API credits.

AP reporting on OpenAI ends with a disclaimer stating, “The Associated Press and OpenAI have a licensing and technology agreement that allows OpenAI access to part of AP’s text archives.”

Should The City have disclosed that it has received money originating from OpenAI? Or does the nature of the award and/or its size and/or the fact it was distributed via a third-party reduce any such obligation? By extension, when, where and how will the FT disclose its relationship when reporting on OpenAI ? And how, if at all, might the relationship impact the paper’s coverage of its latest partner?

To what extent will these deals shape tech stacks and workflows? To what extent do news organizations risk short-term satisfaction translating into some form of vendor lock-in? To what extent can – or should – news organizations prepare for the knock-on effects of sudden shifts in providers’ AI products, or the technology companies’ priorities more broadly?

Prior examples of the latter are too plentiful to list. Some may have bitter memories of being sold on the infamous – and, in some cases, terminal – “pivot to video”. Others will recall how traffic plummeted when friends and family suddenly took priority over news on Facebook. Others still will have revised their distribution in step with Google’s pivot to and from AMP, or Facebook’s push for and abandonment of Instant Articles.

In some ways, the breathless push for generative AI has hallmarks of Google’s aggressive attempts to strongarm publishers into adopting a publishing format they insisted would be beneficial in an ecosystem over which they wielded huge control. As The Verge’s Dave Pierce recalls, “In 2015, Google hatched a plan to save the mobile web by effectively taking it over. And for a while, the media industry had practically no choice but to play along.” (For anyone in need of a recap of how the story ends, the subtitle of Pierce’s piece states, “Google promised to create a better, faster web for media companies with a new standard called AMP. In the end, it ruined the trust publishers had in the internet giant.”)

The old guard have also threatened or implemented blocks on news in locations drawing up legislation they deem undesirable.

In June 2021, Facebook announced the UK launch of Facebook News, a product through which selected news organizations would receive a share of $105 million over three years. Under the headline, ‘Facebook News Will Help Sustain Quality Journalism,’ Nick Clegg, the company’s VP of global affairs, explained that the relative paucity of news links surfaced by the algorithm Facebook controls was not illustrative of journalism’s lack of value to Facebook.

“For many people, Facebook is one of the places they find news,” Clegg wrote. “While the amount of news on Facebook is small as a proportion of the content seen overall — less than 4% of what people see in their feed are posts with links to news articles — it undoubtedly plays a significant role in the distribution of journalism.”

On the contrary, this was “why Facebook has been working with news organizations… to figure out how it can best support the industry for the long term.” Facebook News, Clegg said, “is a result of those conversations.”

Two-and-a-half years later, the company announced it would be depreciating Facebook News in the US and Australia, just as it had in the UK, France and Germany four months earlier. This time, however, the relative scarcity of news links surfaced by the algorithm Facebook controls did indicate journalism’s lack of value to Facebook, who stated, “news makes up less than 3% of what people around the world see in their Facebook feed, and is a small part of the Facebook experience for the vast majority of people.”

LLMs are all the rage at the moment. It’s safe to assume they won’t be forever.

Could the short-term perk of a welcome check come at the cost of longer-term problems? 

OpenAI is reportedly getting access to the FT’s paywalled content as part of the deal.

Theoretically, this means that a reader who currently ends up on an FT article via search, opening the door to a relationship that may lead to a newsletter sign-up or subscription, may soon have their needs satisfied by OpenAI without needing to go anywhere near the FT’s website. Similarly, an existing subscriber may conclude their needs can be satisfied by FT content that is freely available to them from OpenAI.

What impact, if any, could this have on its subscriber churn or future conversions? If such deals could harm subscriptions, what sweeteners are necessary to mitigate those losses? Given that potential losses extend beyond the revenue of subscription fees – by limiting or removing capacity to collect user data, cultivate relationships, etc. – how is the value of any such sweeteners determined?

Is it too cynical to view these partnerships as a(nother?) lobbying exercise?

Some of the more critical assessments of Google and Facebook’s journalism programs have described these ‘philanthropic’ initiatives as PR and/or lobbying exercises strategically introduced as the prospect of legislation loomed.

In the current climate, with eyes on the New York Times’ copyright case against OpenAI and the Justice Department’s antitrust lawsuit against Google, similar questions arise.

Can – and should – the current raft of AI partnerships be interpreted as a contemporary equivalent, designed to deter powerful news organizations from (i) following the New York Times in filing lawsuits for copyright infringement, (ii) pushing governments to intervene, or (iii) participating in any future collective bargaining?

The Financial Times was among the publishers that made a submission to the UK House of Lords Communications and Digital Committee’s report on Large Language Models, which concluded the UK Government “cannot sit on its hands” while waiting for copyright case law to be established.

Per Reuters, News Corp CEO Robert Thomson told the Goldman Sachs Communacopia + Technology Conference in September:”What you will see over time is a lot of litigation; some media companies have already begun those discussions… Personally, we’re not interested in that at this stage. We’re much more interested in negotiation.”

While this didn’t entirely rule out the possibility of litigation (“we’re not interested… at this stage“), it did seem to imply News Corp’s interest in litigation was partly contingent on the outcome of negotiations, i.e. News Corp wouldn’t be interested in litigation if the price is right.

It seems, from Wall Street Journal reporting, the price was around $250 million.

Publishers in Scandinavia have shown a willingness to band together for litigation, collective bargaining, and developing ethical approaches to using AI in journalism. In 2021, the Danish Publishers Collective Management Organisation formed to take on Facebook and Google. Earlier this year, the Danish Rights Alliance, on behalf of the Danish Journalists’ Association and Danish Media Association, took legal action against Apple for alleged violation of the Danish Copyright Act.

How will the outcome of the legal wrangle over copyright between OpenAI and the New York Times impact these kinds of deals? Could the prospect of a ruling in OpenAI’s favor convince some publishers that something is better than nothing?

Will non-partners effectively be punished for not striking a deal?

Per the FT’s announcement, the deal with OpenAI “allows ChatGPT to respond to questions with short summaries from FT articles, with links back to FT.com. This means that the chatbot’s 100mn users worldwide can access FT reporting through ChatGPT, while providing a route back to the original source material.”

OpenAI is willing to provide “rich links” to FT stories it cites because of this partnership. It will presumably be less willing to identify – or maybe even consider – non-partners as a source, given disputes over copyright infringement.

If, unlikely as it may seem, those rich links translate into something meaningful for publishers, will FOMO increase the likelihood of other news organizations striking similar deals? Alternatively, if summaries from chatbots decimate search traffic, will partnerships with AI providers take on a new appeal for otherwise reluctant publishers?

What lessons have news organizations learned from their past dealings with technology companies, and will they put those learnings into practice? Will the relationship be any less fraught? To what extent will news organizations’ prior experience with technology companies inform their approach to potential AI partnerships?

The reshaping of this relationship is something that University of Oxford researcher and Tow Center fellow Felix Simon has been exploring.

Outlining one of the ways in which tensions are likely to emerge, Simon argues, “News organizations have good reasons for increasingly relying on platform companies’ AI… But there is no such thing as a free lunch. AI reshapes the dependency of publishers on platform companies by increasing their control over technological infrastructure, exacerbating existing dependencies in distribution, and introducing new dependencies in production, especially as generative AI is making inroads”

One school of thought says news organizations have a strong negotiating hand because AI platforms will require high quality, up-to-date information. Ingesting archives alone won’t be enough. From this perspective, journalism would seem an extremely valuable commodity.

This, too, has certain echoes of the earlier period of courtship. Back in 2016, a senior editor interviewed for our original study warned, “Publishers have more leverage than they think.”

As talks ramp up and new deals emerge, we edge closer to knowing whether that proves any more prescient this time around.