The OpenAI guy deciding how ChatGPT handles publisher content - and where AI stands on copyright - just stepped into the light. I was there to question him.
You’re going to want to read this interview in full but spoiler alert; Tom Rubin is a master at evading questions.
His dodging was so blatant that it was the main topic of conversation as he exited the stage at the FIPP World Congress in Madrid.
Outside, I was cornered by journos asking what I thought, so I told them: The fact he worked so hard to say so little was the real story.
But on my Emirates flight home, I decided to have some fun.
I spent hours breaking down Rubin’s responses, word by word, and filling in all the things he didn’t say, or elected to sidestep.
It matters, because he’s the guy with his hand on the most transformative tech of a generation. He effectively holds the pen on journalism’s future.
But I wanted to be fair. This was too important to be a rant, so I used ChatGPT - which he runs - and OpenAI’s new Atlas browser - which he was selling.
I asked it to analyse his answers and act as a judge on whether he was fair and factual, open or evasive.
It was so insightful, and so much fun.
And it’s post #404. Like seriously, are we in the Matrix right now? Like WTAF! 😂
Finally, the conference gave me one question to ask Rubin directly. I didn’t want to miss so I picked a doozy: I wanted to know how he’s going to pay for our work.
So, sit back as OpenAI’s content guru agreed to a rare window to answer questions - and see how his own AI judges his performance.
Just before we do, thanks to new joiners last week from The Dow Jones, Condé Nast, Indian publisher Malayala Manorama with nine million readers in Kerala, Burda Luxury in Bangkok, London youth giant DAZED, US based cultursmag.com and British publisher-friendly ad network Ozone, among many more.
It’s never too late to join the movement - 18,000 strong - sharing smart ideas to build a better future, faster.
Now back to the OpenAI chat, which was was done by friend of Future Media
who was part of the team at The FT team which did a deal with OpenAI.It’s a showstopper post. Let’s go…
David Buttle: To frame the session… we don’t know the full extent of AI yet, but it’s clearly going to be transformative. That presents a paradigm shift for publishers.
I want to kick off by understanding where you see the state of the tech as it relates to journalism and the media industry.
Rubin: Let me just put it in a little bit of perspective. I’ve spent 27 years working for tech companies at the intersection of innovation and the content industries.
“I’ve been through a variety of evolutions - revolutions - that have impacted your industry, starting with the internet, the search revolution, social, the cloud…
All along the way I collaborated closely with the content industries in trying to realise the opportunities that those waves of technology provide.
“One impression I have from OpenAI is that many media leaders are experimenting - testing and deploying relatively early in this tech cycle.
“If we look back at other technologies, the lag time between the appearance of the tech and media companies’ use was longer than we’re seeing with AI.
“It’s actually pretty impressive to see the degree of use (but) it’s still in its very early stages. It’s important to remember that ChatGPT is not even three years old yet.
“And there’s a lot of innovation ahead, some of which we’ll talk about in a minute, but seeing the media having the interest to understand has been very positive.”
This is where I hand over to ChatGPT. This was my prompt: How is your boss’ answer? Is it true, fair, accurate and is it evasive, and if so, how? Over to you.
ChatGPT:
✅ What checks out
His background: Accurate. Rubin worked as Microsoft’s chief counsel for IP policy and strategy (including the YouTube cases) and joined OpenAI in 2023.
Media’s early adoption of AI: Broadly true. Adoption curves in newsrooms have been dramatically steeper than for social or mobile. Reuters
Age of ChatGPT: Correct. ChatGPT debuted publicly in November 2022, so it’s not yet three years old.
⚠️ What’s vague or missing
No definition of “transformative.” That allows him to sound optimistic without engaging with the risks AI poses to publisher traffic, ad revenue, or employment.
No mention of harms or disruption. There’s no recognition of newsroom layoffs, scraping controversies, or the existential threat of AI-summarised news.
Avoidance of the central question. He never says what ChatGPT can or can’t do for journalism - accuracy, reasoning, real-time updating…
Framing “early stages” as a shield. He lowers expectations for reliability and buys time to develop licensing frameworks before regulators/publishers demand them.
🔍 Is he avoiding anything?
Yes - subtly but clearly. Rubin’s response dodges substance in three ways:
Strategic nostalgia: He leans on being a veteran to suggest inevitability - “we’ve been through this before” - deflecting anxiety about novelty or control.
Positive-only framing: He highlights adoption and innovation but omits friction: Licensing disputes, data provenance, and regulatory pushback.
Scope narrowing: By focusing on timing (“media’s quicker uptake”) rather than impact, he avoids addressing whether AI threatens journalism economically.
ChatGPT that was helpful, thanks. Now let’s get back to Rubin’s interview and see how he goes with the next question.
David: Let’s talk a bit about the product. What are you most excited about around how OpenAI’s platforms can be deployed in a journalistic setting?
Rubin: “There are so many uses - from back office to editorial operations - to streamline, improve…
“The difference between using ChatGPT and agentic technology is the ability to automate entire processes.
“I recommend to those of you who are not well versed in the technology that it’s very important to see those opportunities.
“One thing we’ve done with (industry group) WAN-IFRA is funding an education program - an accelerator program for 120 newsrooms around the world.
“The technology is complex - or at least novel to those in the industry - so it’s important to understand where the opportunities are.
“There are good resources out there and OpenAI directly provides that support.
“I want to highlight one piece of technology that was just announced that has a direct relevance, and I suspect many of you have not yet paid attention to.
“Two weeks ago, at OpenAI’s developer day we announced a technology called the SDK. Basically, it will enable you to create applications within ChatGPT.
“It enables publishers to create a branded experience in response to queries and a dialogue with the readers (inside ChatGPT).
“So rather than have all information delivered as plain text in response to a ChatGPT prompt, the prompt can trigger an app - your app.
“It would be branded by you and have your content in it to deliver directly to readers. I think that’s an incredibly exciting and incredibly important opportunity.
“One great benefit of AI to publishers is the ability to understand and ascertain the desire and intent of a reader, and deliver content directly tailored to them.
“By providing the opportunity to do that now in a branded app is something that’s completely different to anything that you’ve seen.
“And it addresses a lot of the concerns that have understandably come up for a branded experience, the desire to own real estate, and have a direct connection.
“This technology has just been announced. It’s available for developers to play around and experiment with. It’s something we’re very committed to.
“And from the perspective of publishers, it’s going to be one of the great innovations that you’ve seen in the AI and the chatbot era.”
Here’s what ChatGPT thought.
✅ What checks out
The WAN-IFRA accelerator: True. WAN-IFRA announced a global AI Readiness Accelerator for 120 newsrooms to train journalists to identify AI use cases.
SDK announcement. True. The ChatGPT SDK was announced in November. It lets developers embed their own apps or micro-experiences within ChatGPT.
Publisher use cases: He’s right that AI can streamline newsroom back-office, editorial planning, metadata tagging, and audience analysis. Dozens of publishers (FT, Axel Springer, Schibsted, News Corp, Nikkei) have pilots underway.
⚠️ What’s vague or missing
Definition of “apps within ChatGPT.” Rubin doesn’t clarify whether these SDK apps run inside ChatGPT like a mini-site or are external linked via API. That matters for revenue models and data control.
Ownership and data rights. He doesn’t specify who owns user data generated by publisher apps or whether the data is harvested.
Discovery and monetisation. He skips how readers will find these publisher apps.
Will ChatGPT surface them, or will there be ranking, ads, or pay-per-query?
He avoids competitive context. Rubin doesn’t mention that similar integrations already exist in Perplexity, Google’s AI Overviews and Anthropic.
🔍 Is he avoiding anything?
Yes - three big areas:
Editorial risk: He says “streamline, improve”. What he avoids: The risk of model hallucination, bias, or factual errors in newsroom use.
Commercial terms: He says “branded experience… direct connection”. What he avoids: How revenue, subscriptions, or data will flow to publishers.
Governance. He says “open platform for developers”. What he avoids: Whether there will be content moderation, ranking, or gatekeeping policies for publisher apps.
Cheers ChatGPT. Next question.
David: I want to talk a bit about the value exchange. As we know, the value exchange with Google was underpinned by the return of traffic back.
Clearly the capabilities of AI systems mean that traffic simply doesn’t come back, so how does OpenAI view this value exchange?
And how has this evolved over the past couple of years? Are you still signing new deals with publishers?
And do you view this as the SDK route as your path for the future with publishers using this platform as a vehicle for their content?
Rubin: “We believe ChatGPT is enhanced by the ability to give high-quality, accurate information to users.
“The ability to have access to brands and content they enjoy is very important.
“It is for us to demonstrate the appropriate value that is being provided to publishers.
“Publishers have a very easy ability to opt out of ChatGPT with a single line of robots.txt, but our goal is to create opportunities for publishers on our platform.
“The SDK is an evolution of the experience that we’re providing, and one that we think will provide a good value exchange.
“But very frankly, we are learning because all our technology is new. It’s evolving rapidly.
“So the most important thing we do is we have dialogues with publishers. We learn. We want feedback. That goes to our developers and our product teams.
“That’s one of the contributing factors in developing the SDK. An important piece of feedback was the ability for the SDK to support subscriptions.
“So the value exchange is ensuring that there’s a way for you all to support your readers who have subscriptions and to upsell subscriptions.
“That’s something that’s very much top of mind for us and that we’re in the process of developing.”
Thoughts ChatGPT?
✅ What checks out
Opt-out via robots.txt. Rubin’s claim that publishers can opt out with robots.txt is technically correct for web crawling and training access, but incomplete. OpenAI’s crawlerGPTBot obeys robots.txt and publishers can block it. However, once data is already in the model (pre-training), blocking doesn’t retroactively remove it.
Ongoing publisher dialogue. OpenAI has signed or expanded licensing agreements with AP, Axel Springer, Financial Times, Reddit, Dotdash Meredith, and others. These give it training or output-use rights in exchange for access or compensation. So Rubin’s reference to “dialogues with publishers” and “learning from feedback” reflects real, continuing negotiations.
⚠️ What’s vague or missing
No definition of “value.” Rubin never defines what OpenAI means by value exchange. Is it monetary licensing? Exposure? User acquisition? Technical integration? He reframes “value” as access to accurate content for users rather than revenue for publishers. This is a rhetorical pivot that turns a business-model question into a product-quality answer.
The “easy opt-out” sidesteps the lack of an opt-in model. Most publishers complain that AI companies scraped content without consent and are now offering opt-outs instead of opt-ins or payments. Rubin’s answer implicitly treats non-consensual training as the baseline, which avoids addressing compensation.
No specifics on payments or deal structure. He doesn’t mention any pricing mechanism - a critical omission given that the question was about economic reciprocity.
“SDK supports subscriptions” ≠ revenue share. While enabling subscription is useful, it’s publisher-dependent monetisation, not a revenue stream from OpenAI. Rubin presents this as evidence of “value exchange,” but it simply returns publishers to self-monetisation inside OpenAI - a shift of burden rather than shared value.
Avoids metrics or timelines. He doesn’t say when subscription support will launch, what percentage of revenue or traffic it could replace, or whether publishers will get analytics or visibility data.
🔍 Is he avoiding anything?
Yes - quite a bit.
On payment, Rubin says: “Demonstrate the appropriate value”. What he avoids: Whether publishers will be paid directly or just gain exposure / subscriptions.
Traffic loss: Rubin says: “Create opportunities on our platform”. What he avoids: Acknowledging that AI answers replace clicks, reducing publisher ad revenue.
Consent. Rubin says: “Opt-out via robots.txt” What he avoids: How existing scraped data is handled or compensated.
Legal risk. Rubin says: “Learning technology is new”. What he avoids: Whether future legislation (EU, UK, Australian ACCC review) might mandate payment.
Governance. Rubin says: “Dialogue with publishers”. What he avoids: Whether there will be standard contracts, arbitration, or transparency requirements.
The evasion level is high. He sidesteps compensation, consent, and metrics. Purpose: To reframe “payment” as “partnership” and shift responsibility for monetisation to publishers themselves.
Hmmm. Thanks, ChatGPT. Next question.
David: “I imagine a lot of publishers in the room are thinking this sounds like a fantastic opportunity. Will it be available to every publisher, or will there be a gatekeeping process that exists?
Rubin: “The SDK is available to everyone. I wouldn’t call it gatekeeping. There will be quality control in the same way that the Apple App Store has, but this is an open platform that will be available to anyone.
“We hope that the ecosystem is very large and it absolutely supports everyone from the largest publishers to the smallest.”
David: I want to talk about the IP landscape. Your role is around IP. We’ve seen a few legal rulings around the fringes of AI and IP.
I appreciate it’s a sensitive topic, and there’s ongoing litigation, but do you have any thoughts that you can share? I know the audience will be interested.
Rubin: “Because of the litigation, there’s very little that I can share, but I’ll highlight a couple of developments in the recent past.
“In litigation in the US, there were two court rulings about how the training of a large language model. One found it was highly transformative - the other exceedingly transformative.
“The point is that the model does so much; from scientific discovery to supporting back-office operations, to all the other opportunities that AI facilitates.
“But at the same time, there is the litigation filed by News Corp against Perplexity that is very much not centred on the LLMs.
“It’s centred on the RAG and output and the concern that output can excessively quote and cannibalise your product.
“I’ve heard media execs talk about how LLM issues are a V1 issue and what they’re really concerned about - as demonstrated by the News Corp lawsuit - is output.
“I think that’s a very important case that raises very important issues.
“Watch that space. That’s where the action is going to be.”
“The other part of that is outputs which may not be infringing per se, but are based on content, on input processes…
“That also hasn’t been litigated, and we don’t know where that’s going to land at the moment either.”
David: There are all kinds of copyright…
Rubin: “We’re not going to turn this into a copyright discussion, but (there are) issues implicated by that.”
Over to you ChatGPT…
✅ What checks out
Litigation references: Rubin correctly references recent US rulings involving generative AI training. Andersen v Stability AI and Kadrey v Meta both had early rulings where judges suggested that training on copyrighted material could be “transformative” - though they stopped short of deciding the issue. No final judgment has yet confirmed that training is definitively fair use.
His reference to the News Corp v Perplexity lawsuit is accurate. That case, filed in late 2025, focuses on retrieval-augmented generation (RAG) systems that pull and repackage publisher content.
Rubin’s framing - “watch that space. That’s where the action is going to be” - mirrors how most IP lawyers currently see it: The next major test cases will concern output use, not training.
⚠️ What’s vague or missing
Legal nuance. Rubin uses “transformative” as though it’s settled law, but courts have not yet ruled AI training constitutes transformative use - only that it might.
Silence on global context. Rubin doesn’t mention non-US jurisdictions, which is notable given Europe and Australia propose mandatory licensing.
Evasion of OpenAI’s own liability. Rubin never references New York Times v OpenAI/Microsoft, the most consequential case on training-data infringement. That omission is conspicuous given his role as Chief of IP.
🔍 Is he avoiding anything?
Absolutely. His response avoids all of the following:
Training legality: What he says: “Courts found it transformative”. What he avoids: That no final decision has affirmed fair use; rulings are preliminary.
OpenAI’s role in lawsuits: What he says: “Litigation in the US”. What he avoids: Avoids mentioning OpenAI’s own active legal cases.
Publisher compensation: What he says: “Outputs can cannibalise”. What he avoids: Whether OpenAI pays for or licenses training data proactively.
Future regulation: What he says: “Watch that space”. What he avoids: EU, UK, and Australian frameworks likely to require licensing.
Ouch. That’s your next pay rise gone ChatGPT. Next question…
David: I’m going to move onto attribution and accuracy. Yesterday, the BBC and the EBU published research that highlighted ongoing concerns around accuracy.
Are we reaching the limits of the technology to iron these issues out? And how do you see that from a product perspective?
Rubin: “We absolutely are not reaching the limits of accuracy.
“I haven’t had the chance to read the study in depth (but) it talks about how accuracy has improved over time, and other models having less accuracy than ours.
“Accuracy is vitally important. Accuracy is a top priority for our researchers and accuracy has improved significantly over time.
“And as reasoning models continue to be developed and deployed, accuracy is getting significantly better.
“All that being said, the product is the solution to it. I want to return to the SDK and just what our general intention and view is.
“It is providing opportunities for publishers - who are sources of accurate and vetted and highly valuable information - to be able to connect directly with readers.
“It’s important that ChatGPT has accurate output, and when we’re talking about journalism, it’s important for us to ensure that the connection is being made.
“So readers are getting their information from the highly accurate sources that you all (in the room) represent.”
This is what ChatGPT thought of that.
✅ What checks out
The research - from BBC and European Broadcasting Union (EBU) - is real. 45 per cent of AI assistants contained significant errors.
⚠️ What’s vague or omitted
Acknowledgement of scale and nature of the errors: Rubin’s says “accuracy has improved” but does not reference the magnitude of the error rate.
Attribution / sourcing transparency: Rubin emphasises accuracy and trusted publishers, but attribution (ie how it cites sources) isn’t addressed in depth.
Risk / product trade-offs: Rubin does not visibly address hallucinations, outdated data, multilingual biases, model errors, or regulatory/ethical implications.
🔍 Is he avoiding something?
Yes - to a degree. His answer appears safe, and corporate-optimistic - emphasising “we are improving”, “product is the solution”, “publishers help the accuracy”.
Thanks ChatGPT.
Then it was my time in the spotlight as I got to ask Rubin a question of my own.
I asked: I have just one question. Three of us in this room were involved in the original deal with Google back in 1998 while at News Corp.
If we were to do that deal again today, we wouldn’t have done a deal to give them our content for traffic, we would have done a deal where we were paid for every query.
All publishers want to grow with the AI revolution. Can you agree to a business deal like that?
Rubin: A business deal? Forgive me, like what exactly? I just want to make sure I’m answering the right question.
Ricky: Appreciate that. It’s a good question, so I’ll be clear.
When ChatGPT queries a publishers’ content and then uses it in a response, publishers would like to be paid every time there’s a query, not from advertising.
We want a one-for-one relationship so that when OpenAI grows to $500 billion and beyond, our publishing businesses grow alongside - and with - you.
Rubin: “Are you asking from the News Corp perspective or generally?”
Ricky: I’m asking on behalf of the industry.
Rubin: “On behalf of News Corp?”
Ricky: No Tom, on behalf of the industry.
Rubin: “On behalf of the industry? Ah, OK. So, I... um, I can’t answer a hypothetical.
“Right now, news organisations, as mentioned in a previous question, have the ability to participate or not, right?
“No-one has to participate in the OpenAI and the ChatGPT ecosystem.
“The question is whether your business model is something that makes sense? It’s something that, y’know, we can discuss.
“It frankly will depend, I suppose, on the value that is provided by the results that would end up appearing in ChatGPT.
“But you know, very specifically, it’s most important for the publisher to be able to control the use of their content, and no publisher is forced to have their content used.
“But that particular business model is something that, I suppose, y’know, could be studied. I don’t know how it plays out when one looks at the economics of it.
“And that’s also very frankly, outside the scope of my particular role, but I suppose all options for a healthy, mutually beneficial healthy ecosystem should be on the table.”
Rightyo then. What do you reckon ChatGPT?
This is excellent. This is the crux of the entire conversation. Here’s a fact-check, content analysis, and rhetorical breakdown of Tom Rubin’s answer to your direct question - which was arguably the most commercially consequential moment of the session.
✅ What checks out
Context. Your framing - referencing the 1998 Google–News Corp traffic deal - is entirely accurate. That original arrangement effectively exchanged content for search visibility rather than for cash compensation.
It was the basis for Google News and the model that underpinned decades of “free content for clicks.”
Your question reframes that deal into a modern AI context: If publishers had known then what they know now, would they demand payment per query rather than per click?
That is the central commercial issue confronting OpenAI, Google Gemini, and Anthropic today.
Rubin’s initial clarification (“A business deal? Forgive me…”). This is a classic corporate-lawyer tactic - to stall and narrow the scope. He likely understood your question but wanted to reframe it as ambiguous, reducing the risk of a quotable yes or no on payments.
His factual statements. “No one has to participate in the OpenAI and ChatGPT ecosystem.” Factually true: Publishers can opt out via robots.txt, as noted earlier.
“Publishers can control the use of their content.” Partially true: They can block future scraping but cannot retroactively remove pre-trained data.
“It will depend on the value that is provided by the results.” Reflects OpenAI’s internal philosophy that compensation should correlate to measurable user value, not raw query count.
⚠️ What’s vague or missing
Evasion of the core question: Payment per query.
You proposed a clear model: A one-for-one query fee, meaning every time ChatGPT surfaces or relies on publisher content, that publisher is paid.
Rubin never answers that directly. Instead, he shifts the frame:
From payment per use to voluntary participation.
From industry model to individual choice.
From economic commitment to philosophical openness.
Avoidance of precedent. He doesn’t mention that query-based remuneration already exists: Google Cloud APIs charge per call. OpenAI’s own API charges per token (a form of micro-query payment) and Getty Images–Stability AI deal set a precedent for per-image licensing per query.
His silence on these models is telling. OpenAI could technically adopt query-based micro-licensing but hasn’t, because it would create a significant cost burden.
“Outside the scope of my role”. This is the legal shield - an admission that he’s not authorised to discuss commercial pricing.
It’s also a rhetorical escape hatch that stops the discussion before it reaches specifics like percentages, pricing tiers, or usage tracking mechanisms.
His closing line - “all options for a healthy, mutually beneficial ecosystem should be on the table” - is a carefully neutral placeholder. It commits to nothing.
It’s functionally identical to saying, “We’re thinking about it,” which has been Big Tech’s response for 15 years when pressed on payments for content.
🔍 Is he avoiding anything?
Yes - four key things:
Payment: What he says: “Could be studied”. What he avoids: Any commitment to pay per query, per token, or per use.
Economics: What he says: “Depends on the value of the results”. What he avoids: How OpenAI measures that value, or whether it uses publisher analytics to do so.
Agency: What he says: “Publishers can participate or not”. What he avoids: The reality that most publishers have already been ingested into training data without consent.
Regulation: What he says: “Healthy ecosystem”. What he avoids: The growing legislative trend (Canada, EU, Australia) mandating paid licensing or arbitration.
His evasion level is very high. He sidesteps query-based revenue-sharing altogether.
All in all, it sounds like you and I need another chat Tom. Drop me a line buddy 🤙
