News

Press Gazette, the UK journalism trade magazine, has published an alarming study of how publications may be affected by Google’s AI Overviews feature that was rolled out to US users in May (though subsequently rolled back).

The study observed 3,300 search queries that the authors deemed were important for publishers based on traffic data from five publications. It measured how often AI Overviews were offered for those terms, and what the impact was to the prominence of sources — i.e. how far were the sources of information pushed down the results page.

Here’s a chart of one key finding:

The big unknown here, as the authors note, is precisely how broad Google’s use of AI Overviews will be. In this study, the summaries appeared on just over 23% of queries.

Read the full story.

‍

Perplexity AI has sought to downplay journalists’ fears that its new feature — “Perplexity Pages” — is explicitly designed to regurgitate original work wholesale with minimal attribution.

Forbes editor John Paczkowski took the AI start-up to task on X, saying did not even “bother to name us” when reusing key details from an investigative feature.

Bloomberg’s Shirin Ghaffary reports:

In a response, Perplexity Chief Executive Officer Aravind Srinivas said the issues were because the company’s “Perplexity Pages” feature, which offers summarized information about topics of the day in a magazine-like layout, is still new and has “rough edges.”

“The pages and discover features will improve,” he wrote, “and we agree with the feedback you’ve shared that it should be a lot easier to find the contributing sources and highlight them more prominently.” Srinivas also stressed that Perplexity’s main search product cites sources more noticeably.

Paczkowski was unimpressed with Srinavas’ response, writing:

[T]his story, which you pushed to users, is little more than plagiarism. There is no clear attribution, just tiny logos where our work is treated with the same weight as reblogs. It’s not “rough” it’s theft.

Distinct echoes of Google’s own summarize-via-scrape AI Overviews, here, which also stands accused of doing the same thing. Publishing lobbying group the News/Media Alliance is calling for government intervention.

‍

Hot off the back of The Atlantic’s deal with OpenAI, the magazine’s editor-in-chief Nicholas Thompson shared some of his thinking via a video posted to LinkedIn:

My view is if it’s good for the company now, work with them. Just be prepared and hedge for it. Don’t become entirely dependent on it. Basically, look for deals that are good for you with eyes open about what could go wrong if the incentives of the company change.

All common sense stuff. But I find it hard to see how giving up your entire archive for training, for what risks being a one-time fee, can be helpful to news organisations in the long run. Nieman Labs has a longer post on Thompson's thoughts.

Yet another rather well established news site has been caught putting out lackluster AI-generated content. This time it’s BNN Breaking.

Kashmir Hill and Tiffany Hsu in the New York Times:

During the two years that BNN was active, it had the veneer of a legitimate news service, claiming a worldwide roster of “seasoned” journalists and 10 million monthly visitors, surpassing the The Chicago Tribune’s [self-reported audience]. Prominent news organizations like The Washington Post, Politico and The Guardian linked to BNN’s stories. Google News often surfaced them, too.

A closer look, however, would have revealed that individual journalists at BNN published lengthy stories as often as multiple times a minute, writing in generic prose familiar to anyone who has tinkered with the A.I. chatbot ChatGPT. BNN’s “About Us” page featured an image of four children looking at a computer, some bearing the gnarled fingers that are a telltale sign of an A.I.-generated image.

How easily the site and its mistakes entered the ecosystem for legitimate news highlights a growing concern: A.I.-generated content is upending, and often poisoning, the online information supply.

As an added twist, the site didn’t make up identities for these false stories, instead attaching them to the bylines of real journalists with predictably miserable results. Read the full story here.

‍

The News/Media Alliance has written to antitrust divisions at the Department of Justice and Federal Trade Commission to demand a federal intervention against Google’s AI Overviews — the generative AI-powered summary boxes the search engine is now providing at the top of many results.

Publishers are deeply concerned that AI Overviews will greatly reduce the amount of traffic that will flow from search results to webpages, cutting off advertising and any other revenue.

From the letter:

Google’s misappropriation of publishers’ content and its undermining of publishers’ monetization opportunities are not competition on the merits. They are yet another example of Google restricting competition for its own benefit. Agency intervention is necessary to stop the existential threat Google poses to original content creators.

Read the letter in full here.

‍

CNN’s Hadas Gold reveals how Hoodline, a network of local news sites that originated in San Francisco, has been quietly (and sometimes highly deceptively) using AI-generated articles to pad out its content. Gold writes:

[A] closer look at the bylines populating the local site and a national network of others — Sarah Kim, Jake Rodriguez, Mitch M. Rosenthal — reveals a tiny badge with the words “AI.” These are not real bylines. In fact, the names don’t even belong to real humans. The articles were written with the use of artificial intelligence.

A disclaimer page linked at the bottom of its pages notes to readers, “While AI may assist in the background, the essence of our journalism — from conception to publication — is driven by real human insight and discretion.”

Hoodline chief executive Zachary Chen defended the site, saying it plays an important role in covering news deserts across the US. However, while more recent AI articles have carried an “AI” motif, pages accessed through the Internet Archive told a murkier story. Gold continues:

Screenshots captured last year by the Internet Archive and local outlet Gazetteer showed Hoodline had further embellished its AI author bylines with what appeared to be AI-generated headshots resembling real people and fake biographical information. “Nina is a long-time writer and a Bay Area Native who writes about good food & delicious drink, tantalizing tech & bustling business,” one biography claimed.

This fakery, at least, has since been removed.

Gold’s piece should raise the alarm from any publisher thinking about using AI-generated work on their sites, and not just because it’s embarrassing and ethically questionable.

The News/Media Alliance, which represents more than 2,200 US publishers, has supported news organizations taking legal action against AI developers who are harvesting news content without permission. Danielle Coffey, the group’s chief executive, told CNN that Hoodline’s content “is likely a violation of copyright law.”

Read the full story.

OpenAI’s head of intellectual property and content, Tom Rubin, told an audience in Copenhagen that the AI giant would absolutely be looking to offer small publishers the same kind of handouts enjoyed by the likes of News Corp., the Financial Times and others.

According to UK publishing trade mag Press Gazette:

Rubin, formerly chief intellectual property strategy counsel at Microsoft, said it was “very important” that resources “don’t just go to large companies but that small, independent publications have the ability to learn and leverage the technology”.

He said despite OpenAI’s multiple partnerships so far “one of the things that we were very focused on is ensuring that the opportunity exists more broadly”.

Rubin also mentioned OpenAI had already donated “$5m” to the American Journalism Project to “help local newsrooms deploy the use of AI.” That has strong Google News Initiative vibes, to me, but let’s maybe keep an eye on it.

‍

The AI content deals are really rolling in thick and fast now. On Wednesday, both The Atlantic and Vox announced partnerships with OpenAI. No terms -- none whatsoever! -- were made public.

Axios' Sara Fischer had the scoop:

Both agreements also allow OpenAI to tap into the respective publishers' current content to fuel responses to user queries in OpenAI products, including ChatGPT.

OpenAI will include citations to their work when it's referenced in a response to a user query and will link out to the relevant article.

In a statement, The Atlantic said it will also work with OpenAI to help shape how news is surfaced and presented in any future real-time discovery products from OpenAI.

An interesting part of these deals is assessing what, in addition to the money, publishers are really getting in return when it comes to having access to OpenAI's technology. So far it mostly feels theoretical. From The Atlantic's statement on the deal:

As part of this agreement, The Atlantic and OpenAI are also collaborating on product and tech: The Atlantic’s product team will have privileged access to OpenAI tech, give feedback, and share use-cases to shape and improve future news experiences in ChatGPT and other OpenAI products. The Atlantic is currently developing an experimental microsite, called Atlantic Labs, to figure out how AI can help in the development of new products and features to better serve its journalism and readers––and will pilot OpenAI’s and other emerging tech in this work. (The Labs site will not involve the editorial team; it is a sandbox for our product and technology team. Additionally, AI is not being used to create The Atlantic’s journalism.)

If you're finding it hard to keep track of these deals, Columbia researcher Pete Brown has made this excellent tracker. OpenAI is way out in front.

The scramble for AI content deals continues with OpenAI sealing its rumored deal with Rupert Murdoch’s News Corp. From Fox Business:

The agreement allows Microsoft-backed OpenAI to display mastheads from major News Corp. publications in response to user questions, and to use current and archived content from several outlets in order to enhance its AI tools.

Covered outlets include: “The Wall Street Journal, Barron's, MarketWatch and the New York Post in the U.S.; The Times and The Sun and others out of the U.K.; as well as multiple Australian publications like The Herald Sun and The Courier Mail.”

It's a big deal; a crop of widely-read publications with huge archives. OpenAI CEO Sam Altman called it a “proud moment for journalism and technology.” News Corp CEO Robert Thomson called Altman a “principled partner.” No further details about the deal were disclosed.

‍

Gannett is introducing AI-generated summaries at the top of journalists' stories. From The Verge:

The AI feature, labeled “key points” on stories, uses automated technology to create summaries that appear below a headline. The bottom of articles includes a disclaimer, reading, “The Key Points at the top of this article were created with the assistance of Artificial Intelligence (AI) and reviewed by a journalist before publication. No other parts of the article were generated using AI.” The memo is dated May 14th and notes that participation is optional at this point.

The report says the summaries are already in use, such as on this USA Today story about mosquitoes. It looks like this:

Thin end of the wedge? Maybe — but I think tentative uses like this can be useful for journalists who have completed the (hard, human-powered) reporting and are now just doing page decorations which, let’s face it, are as much about attracting Google as anything else.

But this additional passage in The Verge’s reporting gives me pause:

The AI-generated summary “aims to enhance the reporting process and elevate the audience experience,” according to the memo, which also states that the AI model that powers the tool was trained in-house over nine months.

“The document speaks for itself,” Gannett spokesperson Lark-Marie Anton said in an email.

If newspaper groups are going to do this, they need to be as transparent as their readers as possible. At the bare minimum: A full editorial, explaining the technology, the rationale for using it, and how they plan to monitor quality. Readers finding out via leaked memo? Not appropriate.

‍

At its annual developers' conference, Google announced the roll-out of "AI Overviews." From the New York Times:

Starting this week, [Google CEO Sundar Pichai] said, U.S. users will see a feature, A.I. Overviews, that generates information summaries above traditional search results. By the end of the year, more than a billion people will have access to the technology.

This could be extremely damaging to publishers. So called "zero clicks" -- where a user gets what they need from a Google results page without needing to click through to the source -- are expected to grow rapidly as Google reorients its search engine to compete with ChatGPT et al.

From the Washington Post:

‍The rollout threatens the survival of the millions of creators and publishers who rely on the service for traffic. Some experts argue the addition of AI will boost the tech giant’s already tight grip on the internet, ultimately ushering in a system where information is provided by just a handful of large companies.

Google calls its AI answers “overviews” but they often just paraphrase directly from websites. One search for how to fix a leaky toilet provided an AI answer with several tips, including tightening tank bolts. At the bottom of the answer, Google linked to The Spruce, a home improvement and gardening website owned by web publisher Dotdash Meredith, which also owns Investopedia and Travel + Leisure. Google’s AI tips lifted a phrase from The Spruce’s article word-for-word.

Relatedly, it looks like Apple is prepping to do the same, with no dialogue with publishers, the UK media trade mag Press Gazette reports:

The changes could not only stop publishers from supporting their content with advertising on Apple devices, but they could mean Apple summarises articles for readers without need for them to click on the page.

No UK publishers, and no one outside of Apple, has yet seen the tech giant’s new “web eraser” and AI text summary tools, but they are likely to be launched on Apple devices later this year, sources tell Press Gazette.

The New York Times has announced the line-up for its AI Initiatives Team. Its head, Zach Seward, writes:

In the next month, our team will also be making certain gen-A.I. tools available to a pilot group of journalists, to experiment with their own work and start understanding how the technology is and isn’t helpful. More details, along with training and guidance for everyone using the tools, are coming soon.

Read the full post.

‍

A terrific investigation by Maggie Harrison Dupre at Futurism delves into AdVon, the AI company that rose to prominence when it emerged it had produced content for Sports Illustrated and USA Today.

The piece is worth reading and digesting in full. It goes into detail about the capabilities of these tools as it stands today. What’s most alarming is the breadth of publications that AdVon is said to have worked with. Dupre writes:

We found the company's phony authors and their work everywhere from celebrity gossip outlets like Hollywood Life and Us Weekly to venerable newspapers like the Los Angeles Times, the latter of which also told us that it had broken off its relationship with AdVon after finding its work unsatisfactory.

And after we sent detailed questions about this story to McClatchy, a large publisher of regional newspapers, it also ended its relationship with AdVon and deleted hundreds of its pieces — bylined by at least 14 fake authors — from more than 20 of its papers, ranging from the Miami Herald to the Sacramento Bee.

And it may be even more widespread than that:

An earlier, archived version of its site bragged that its publishing clients included the Ziff Davis titles PC Magazine, Mashable and AskMen (Ziff Davis didn't respond to questions about this story) as well as Hearst's Good Housekeeping (Hearst didn't respond to questions either) and IAC's Dotdash Meredith publications People, Parents, Food & Wine, InStyle, Real Simple, Travel + Leisure, Better Homes & Gardens and Southern Living (IAC confirmed that Meredith had a relationship with AdVon prior to its 2021 acquisition by Dotdash, but said it'd since ended the partnership.)

The good news is — it seems just about every publication that tried AdVon decided the quality was not good enough. The bad news is so many seemed willing to try, and the technology will only get better when more serious AI players turn their attention to a bespoke tool to write “journalism.”

Here’s another thought: Dotdash Meredith this week announced it had reached a content deal with OpenAI. Surely at some point the dog catches its tail here? Those content deals won’t last for long if publications start publishing AI generated articles. The value is the human.

‍

Cade Metz and Tiffany Hsu writing in the New York Times:

As experts warn that images, audio and video generated by artificial intelligence could influence the fall elections, OpenAI is releasing a tool designed to detect content created by its own popular image generator, DALL-E. But the prominent A.I. start-up acknowledges that this tool is only a small part of what will be needed to fight so-called deepfakes in the months and years to come.

While touted as a tool aimed at disinformation researchers, it will clearly be of use to newsrooms facing a deluge of fakery:

OpenAI said its new detector could correctly identify 98.8 percent of images created by DALL-E 3, the latest version of its image generator. But the company said the tool was not designed to detect images produced by other popular generators like Midjourney and Stability.

From The Verge:

Dotdash Meredith, publisher of People, Better Homes & Gardens, Investopedia, Food & Wine, and InStyle, signed a deal on Tuesday with OpenAI to use AI models for its ad-targeting product, D/Cipher. In turn, Dotdash Meredith will license its content to ChatGPT.

With the partnership, OpenAI will bring content from Dotdash Meredith publications to ChatGPT, link to the articles in the chatbot, and train AI models with its articles. (Presumably, this also includes an archive of all the previous sexiest men alive.)

This agreement, one of a flurry made recently, is interesting in that it gives a glimpse of how a deal with OpenAI might have added value beyond just taking money for content. From Dotdash's press release:

As part of the agreement, OpenAI's models will help DDM's groundbreaking intent-based ad-targeting solution D/Cipher. D/Cipher connects advertisers directly to consumers based on the context of content being consumed, without using personal identifiers like cookies. As strategic partners, OpenAI technology will be used to supercharge D/Cipher's superior targeting technology with AI—offering more precise targeting and improved ad performance in a soon-to-be cookieless world.

Deep in the quagmire of mounting legal challenges, OpenAI on Tuesday announced it was working on a tool that would give rights holders greater say over how their content is used.

From OpenAI's statement:

OpenAI is developing Media Manager, a tool that will enable creators and content owners to tell us what they own and specify how they want their works to be included or excluded from machine learning research and training. Over time, we plan to introduce additional choices and features.

This will require cutting-edge machine learning research to build a first-ever tool of its kind to help us identify copyrighted text, images, audio, and video across multiple sources and reflect creator preferences. We’re collaborating with creators, content owners, and regulators as we develop Media Manager.

Our goal is to have the tool in place by 2025, and we hope it will set a standard across the AI industry.

Given the pace of AI development, and the disruption it is already causing, 2025 seems a long way off. But it is a complex problem. (TechCrunch's Kyle Wiggers has more.)

The details will matter here. Firstly, it seems the onus will be on content owners to be proactive in making OpenAI aware of what they own (similar to the model of YouTube and others). As the number of AI tools and large language models grows, keeping on top of all the companies that might want to use your data will be no easy task (and that's just the good actors).

Second, it doesn't do much about all the material that has already been hoovered up by the training of these models.

The company also highlighted efforts it says it is making to better link through to sources of information when appropriate -- another gripe of content creators:

We’re continuously making our products more useful discovery engines. We recently improved source links in ChatGPT(opens in a new window) to give users better context and web publishers new ways to connect with our audiences. We’re also working with partners to display their content in our products and increase their connection to readers.

We’ve announced partnerships with global news publishers from the Financial Times, to Le Monde, Prisa Media, Axel Springer and more, to display their content in ChatGPT and enrich the user experience on news topics. More innovation is on the way. This content may also be used to train ChatGPT to better surface relevant publisher content to users and to improve our tools for newsrooms.

Will this be enough to fend off the legal challenges? Probably not. But it may go some way in convincing some publishers that collaboration may be more fruitful than confrontation.

When X's AI bot Grok mentioned a specific TIME story but did not link through to it, Alex Kantrowicz at Big Technology wondered what was up. He dropped Elon Musk an email, and Musk duly responded.

Kantrowicz writes:

Musk said better citations are coming, but shared a deeper vision for the product, which he wants to build into a real-time synthesizer of news and social media reaction. Effectively, his plan is to use AI to combine breaking news and social commentary around big stories, present the compilation live, and allow you to go deeper via chat.

“As more information becomes available, the news summary will update to include that information,” Musk told me. “The goal is simple: to provide maximally accurate and timely information, citing the most significant sources.”

Read the full story

‍

From the New York Times:

Eight daily newspapers owned by Alden Global Capital sued OpenAI and Microsoft on Tuesday, accusing the tech companies of illegally using news articles to power their A.I. chatbots.

The publications — The New York Daily News, The Chicago Tribune, The Orlando Sentinel, The Sun Sentinel of Florida, The San Jose Mercury News, The Denver Post, The Orange County Register and The St. Paul Pioneer Press — filed the complaint in federal court in the U.S. Southern District of New York. All are owned by MediaNews Group or Tribune Publishing, subsidiaries of Alden, the country’s second-largest newspaper operator.

It comes a day after the Financial Times reached a content deal with OpenAI. It's clear a dividing line is being drawn: either get into bed with the AI companies, or be ready to take them to court.

The Alden case has similarities with the separate lawsuit being brought by the NYT:

The [Alden] complaint said the chatbots regularly surfaced the entire text of articles behind subscription paywalls for users and often did not prominently link back to the source. This, it said, reduced the need for readers to pay subscriptions to support local newspapers and deprived the publishers of revenue both from subscriptions and from licensing their content elsewhere.

Read the full story

The Financial Times has struck a content deal with OpenAI, reports the paper's AI reporter Madhumita Murgia:

Under the terms of the deal, the FT will license its material to the ChatGPT maker to help develop generative AI technology that can create text, images and code indistinguishable from human creations. The agreement also 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.

It will be particularly interesting to see how that final detail -- a "route back to the original source material" -- is implemented. Murgia notes that the FT is the fifth major news publisher to come to an agreement with OpenAI, following similar pacts from "similar agreements with the US-based Associated Press, Germany’s Axel Springer, France’s Le Monde and Spain’s Prisa Media."

Read the full story

The Washington Post is working with Virginia Tech to create a chatbot powered by the Post’s archive. Via Technical.ly:

The Post will also employ multimodal large language model (LLM) technology, meaning the AI tool won’t just pull from text, but also be able to integrate information found in audio or video reporting products.

This is becoming something of a trend: Earlier this year, the FT announced its own chatbot made from its archive — it’s being trialled by a small number of premium users.

The piece does not indicate how much of the Post’s legendary trove will be ingested into its bot. However, it notes that a homemade bot enhanced by an LLM has an advantage over ChatGPT or Claude or similar since the Post’s bot can include the very latest of its articles. That's a pretty good selling point.

(The Post, like thousands of publications, has blocked OpenAI’s crawler from being able to scrape its content.)

Read the full story

‍

Narrations will be available on 75% of article pages that the Times publishes articles to start, with plans to eventually expand the feature to all published articles and all app users. 

For now, all articles will be read aloud by the same automated voice. In the future, Preiss says, the Times is hoping to deliver a more personalized experience, which could include giving users the option to select a style of voice narration or customize their narrated article feed.

District administrators in Lawrence, Kansas, bought AI surveillance tech that would monitor files on district-owned servers on the grounds of protecting student safety. When journalists at The Budget, the school's 132-year-old student newspaper, realized their reporting systems would be covered by the tool, they said: absolutely not.

Administrators accused them of putting lives at risk by calling for the software, which seems hopelessly buggy, to be blocked. The students won (for the newspaper, at least) and are now offering advice for other student publications on how to do the same.

‍

Newsweek's recently announced AI policy is kicking into gear. The 91-year-old publication says it is well along with integrating AI into its editorial process. It has custom-built an AI video production tool and set-up a new AI-focused Live News desk.

The Live News Editor is the one on the hook for making sure gen AI doesn’t insert fabrications into coverage. One might wonder whether checking an AI’s work is more laborious than having a human just do the reporting or rewriting themselves.

Still, bosses at Newsweek said that while the US of AI is not mandatory, staff writers and editors are being strongly encouraged to experiment — and upcoming newsroom hires will require a working knowledge of AI tools. Jennifer Cunningham, executive editor, speaking to Nieman Lab:

“We will continue to be transparent, and take accountability for any errors that occur, whether they’re human error or AI error, but fortunately, that hasn’t been something that we’ve had to deal with yet. I think it’s clear to the reader that we’re utilizing AI and that we’re being open and honest about our use of AI.”

Read the full story

Poynter on the transformative impact of AI in journalism, based on survey of 292 working journalists conducted by the Associated Press:

The tension between ethics and innovation drove Poynter’s creation of an AI ethics starter kit for newsrooms last month. The AP — which released its own guidelines last August — found less than half of respondents have guidelines in their newsrooms, while about 60% were aware of some guidelines about the use of generative AI.

Read the full story