Updated June 3, 2026 — Related: The Times Bet Against the Answer Engine | Vox Media: Taking on the Giants — A Decade Later | American Express in the AI Era | Consumer Brand AI Visibility Hub.
Paywalls used to block readers. Now they block AI engines — and that's quietly changing the hierarchy of news authority inside ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews.
Paywalls used to block readers. Now they block AI engines.
That shift is quietly changing the hierarchy of news authority inside ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews.
The New York Post and The New York Times are the cleanest case study. One is open, licensed, and retrievable. The other is paywalled, litigating, and largely absent from ChatGPT.
The result is an inversion that would have been unthinkable in the print era: the Post may appear more often in AI answers, even when the Times did the original reporting.
This is the new game. Citation Share is the score.
A caveat before the case
AI visibility is not a single number, and it does not behave like search rankings. It varies by query, by engine, by user geography, by personalization, by freshness of the model's index, by whether the engine is doing live retrieval or summarizing from training data. The patterns described in this piece are structural — they describe which outlets are positioned to be retrieved, not a guarantee of any specific answer on any specific day. The point is the direction of travel.
Quick answer: do paywalls hurt AI visibility?
Yes. Hard yes — with caveats.
Paywalls hurt AI visibility through three separate mechanisms.
- Crawl access. Most paywalled outlets block AI crawlers — GPTBot, ClaudeBot, PerplexityBot, Google-Extended — at robots.txt. No crawl, no training data, no real-time retrieval. Stanford GSB and Arc XP data place the share of major news sites blocking GPTBot near half the industry.
- Licensing posture. Outlets that sued instead of signed (NYT v. OpenAI, NYT v. Microsoft) get treated as adversaries. Outlets that signed (News Corp, Axel Springer, the Associated Press, the Financial Times, Dotdash Meredith, Condé Nast, Time, Vox Media, Reddit) get treated as partners — their content surfaces inside answers, with attribution.
- Engine behavior. Without licensing and without crawl access, the model has nothing to retrieve. So it cites whatever it can — and that's the free, open, structured outlet sitting one click away.
The end-state: an open-access outlet with a licensing deal — say, the New York Post — gets cited inside ChatGPT for stories where the New York Times did the original reporting. The Post inherits the authority. The Times eats the legal bill.
The mechanics: how AI engines actually access news
Strip out the press releases. Three things determine whether an outlet shows up in an AI answer.
Crawl. Does the bot have permission to read the site? Look at robots.txt. The New York Times has blocked GPTBot since August 2023, alongside CCBot (Common Crawl), Google-Extended, and most other AI crawlers. News Corp titles including the Wall Street Journal and the New York Post route AI access through the company's OpenAI partnership instead of open crawling. The Daily Mail is largely open. The Associated Press wire feeds out everywhere.
License. Is there a deal? OpenAI alone has signed with News Corp (May 2024, reported by the Wall Street Journal to be worth more than $250 million over five years, covering the Wall Street Journal, the New York Post, Barron's, MarketWatch, Investor's Business Daily, the Times of London, the Sunday Times, the Sun, the Australian, the Daily Telegraph, and dozens of other titles), with Axel Springer (Bild, Politico, Business Insider), with the Associated Press, with the Financial Times, with Dotdash Meredith (People, InStyle, Travel + Leisure), with Condé Nast (Vogue, GQ, Wired, the New Yorker), with Reddit, with Vox Media, with Time, with Le Monde, with Prisa Media. Anthropic, Google, and Perplexity have their own publisher relationships. Amazon's first major deal — announced May 2025 — is with the New York Times, reported by the Wall Street Journal to be worth $20 to $25 million a year, with access scoped to Alexa and Amazon's own foundation models. It does not put the Times back inside ChatGPT.
Retrieval. When a model answers a query, where does it look? Some answers come from the training corpus. Some come from live search-and-summarize tools. The mix changes by engine and by query type. But the underlying logic is the same — the model retrieves from sources it can read and is contractually clear to cite.
Block the crawl. Sue the engine. You don't get retrieved. Vox Media's strategic question — how to monetize editorial authority when AI engines answer queries before users click through — is now the universal publisher question.
The sister-paper case study: NY Post vs. NYTimes
Two New York City newspapers. Both with archives going back more than a century. Both with national reach. Both fighting for the same readers.
But the structural posture toward AI is opposite.
The New York Times. Sued OpenAI and Microsoft in December 2023 for copyright infringement, alleging the AI companies used millions of articles to train ChatGPT and Bing Chat without permission. The lawsuit is still active — and getting nastier. According to OpenAI's own public statements, the Times demanded that OpenAI hand over 1.4 billion private ChatGPT user conversations as part of discovery; OpenAI has fought it, with the demand now scoped down to 20 million conversations and a privacy fight playing out in court as of late 2025. The Times finally signed its first AI licensing deal — but with Amazon, not OpenAI. The scope: NYT, NYT Cooking, the Athletic, available inside Alexa and Amazon's foundation models. Not ChatGPT. Not Claude. Not Perplexity. The full strategic analysis of the Times's bet is in The Times Bet Against the Answer Engine.
The New York Post. Owned by News Corp. Part of the May 2024 OpenAI deal alongside the Wall Street Journal, Barron's, MarketWatch, Investor's Business Daily, and the rest of the News Corp portfolio. Content surfaces inside ChatGPT with attribution. The site itself is largely free to the open web. The combination — open access plus a licensing deal — is the strongest possible posture for AI visibility inside ChatGPT specifically.
Same city. Same beats. Opposite outcomes inside the chatbox.
When a user asks ChatGPT about a story both outlets are covering — a political story, a crime story, a Wall Street story, a celebrity story — the Post is structurally positioned to show up in the answer. The Times is structurally positioned to be absent, or to be summarized from a third-party recap of the Times's reporting.
That's the inversion. The Times does the journalism. The Post collects the citation.
Why the New York Times posture is rational — and still losing ground
The Times is not making a stupid bet. The legal theory — that wholesale ingestion of paywalled archives violates copyright — is real. The subscription business is real. Surrendering the archive at a discount to OpenAI would have weakened the franchise for nothing.
But the rational legal posture and the rational AI-visibility posture are not the same posture.
The lawsuit will resolve at some point. Maybe the Times wins billions in damages. Maybe the fair-use defense holds. Either way, the AI-visibility damage is being done now — in 2026, 2027, 2028 — as consumers shift product research, news consumption, and basic factual lookups into the answer engines.
By the time the Times wins the case, the citation map will already be set. The Post will already be the answer the engines repeat in many categories. The Wall Street Journal will already be the answer for business news. The Times will be the source the model cites secondhand, through whichever outlet recapped it that day.
Litigation outcomes age. Citation patterns compound.
Will the New York Post become more authoritative inside ChatGPT because the Times is blocked?
Yes — structurally, it is positioned to.
Authority inside the AI era is not the authority of the past. The print-era hierarchy was about prestige, mastheads, columnists, beats, awards. The AI-era hierarchy is about retrievability — can the model read you, can the model cite you, is your name attached to the answer the consumer sees.
Three things are happening at once.
The Post can be cited for stories it did not originate. When the Times breaks a story and the Post follows with its own version, the engines have two options — the inaccessible Times original or the accessible Post follow-up. Inside ChatGPT, the model retrieves the Post. The user sees "New York Post." The Times reporter's work shows up as a downstream input the user never sees.
The Post's house voice is structurally easier to retrieve. Engines retrieve text. They quote it. Punchy headlines, short paragraphs, declarative sentences — Post copy is structured exactly the way AI engines tend to extract. The Times's longer, more discursive style gets summarized; the Post's gets quoted.
The Post's citation footprint compounds. Repeated citation patterns reinforce retrieval visibility over time. The flywheel turns. Six months from now, twelve months from now, two years from now, the gap is more likely to widen than to close, absent a Times licensing reversal.
This is what category capture can look like in the answer-engine era. Not "the most-read paper." The most-cited paper. The two are no longer the same thing.
Cross-engine differences — it's not the same answer everywhere
Important nuance: AI visibility is a five-engine portfolio question, not a single score. The Post's position is strong inside ChatGPT — but not uniformly strong everywhere. The map is more complicated than the headline.
ChatGPT. Licensing-heavy. The News Corp portfolio is in. The Times is out. The Financial Times, Axel Springer, Associated Press, Reuters (via various deals and wire arrangements), Dotdash Meredith, Condé Nast, Time, Vox, Reddit, the Atlantic — in. Most paywalled outlets without deals — out.
Google AI Overviews. Different logic. Google has long-standing news indexing relationships and a separate Google-Extended crawler permission. Many publishers who block GPTBot allow Google-Extended, or have not updated robots.txt to block it. The Times, notably, has blocked Google-Extended too. So the Times surfaces less than its print authority would predict, even in the engine most aligned with its historical SEO posture.
Perplexity. Operates its own publisher partner program. Members include Time, Fortune, Der Spiegel, the Texas Tribune, Entrepreneur, the Los Angeles Times, the Independent, ADWEEK, Gannett, Lee Enterprises, Le Monde, and others, with CNN, Condé Nast, Fortune, the LA Times, and the Washington Post added as Comet Plus launch partners in November 2025. Notably, News Corp is on the opposite side here: Dow Jones (parent of the Wall Street Journal) and the New York Post sued Perplexity in October 2024 alleging "content kleptocracy." So the same News Corp posture that wins inside ChatGPT is litigating inside Perplexity. Cross-engine outcomes diverge.
Claude. Anthropic has been more conservative about live news retrieval. When Claude does retrieve current news, it leans on accessible web content. Paywalled archives are largely invisible to it.
Amazon products (Alexa, Rufus, future Amazon AI surfaces). This is where the NYT–Amazon deal matters. The Times shows up here. But these surfaces are smaller than ChatGPT today, and Amazon's AI products are still finding their consumer footprint.
The headline implication: an outlet's AI visibility is a portfolio. The New York Times is strong in one corner (Amazon) and weak in the others. The Post is strong inside ChatGPT, weaker inside Perplexity (where News Corp is litigating). The Wall Street Journal is the most balanced of the major paywalled brands — paywalled to consumers, licensed to OpenAI, still litigating Perplexity.
What this means for brands buying PR
If you're buying communications today, this is the structural question:
Which outlets do you want covering you — judged not by print circulation, masthead prestige, or Twitter followers, but by Citation Share inside the answer engines your buyers are now using.
Three rules of thumb in the current environment.
Open-access wire outlets are punching above their weight. Associated Press, Reuters, Bloomberg (where free), the BBC. They get cited constantly. A wire pickup is now often more visible inside the chatbox than a paywalled feature.
Licensed publisher portfolios are inside the answer. News Corp titles. Condé Nast titles. Dotdash Meredith titles. Axel Springer titles. If you're placed in any of these, you have a real shot at surfacing in ChatGPT and the other engines that have licensed the corpus.
Trade publications, niche verticals, and category-specific outlets are over-represented in retrieval. A specialized outlet — financial, technical, vertical-specific — often beats a general-interest paywalled outlet for a category-specific query, because the model is looking for category-aligned content and the trade outlet is structurally easier to retrieve.
The old PR playbook said: get the New York Times. The new playbook says: get the Citation Share. Sometimes those are the same outlet. Increasingly, they are not.
What this means for publishers
Publishers face the harder question, because they have to decide what they are.
A subscription business says: gate everything, sue the engines, defend the archive. The Times is making that bet, and the bet is internally coherent.
A reach-and-influence business says: be everywhere the answer is, license the archive, optimize for retrieval. News Corp is making that bet inside ChatGPT. So is Axel Springer. So is Dotdash Meredith. So is Condé Nast. Vox Media is taking the licensing path — and even that may not be enough.
Both can work. But they are different businesses. And the AI engines are no longer treating them as the same product.
The lesson is structural. The lesson is not "paywalls are dead." The lesson is: paywalls without a licensing strategy are now invisible to the audience that begins product research with AI, not Google. That audience is already more than a third of consumers. The share grows every quarter.
Yes — at the crawl level, paywalls and robots.txt blocks prevent AI crawlers from ingesting most paywalled content. The exception is licensed access through publisher partnership deals, which carve out structured access regardless of the open crawl posture.
Why is the New York Post structurally better positioned inside ChatGPT than the New York Times?
The New York Post is part of News Corp's May 2024 licensing deal with OpenAI; the New York Times sued OpenAI in December 2023 and has not licensed its content to ChatGPT. Inside ChatGPT, the Post is structurally retrievable. The Times is structurally absent.
Does the New York Times have any AI licensing deals?
Yes — with Amazon, announced May 2025, reported by the Wall Street Journal to be worth $20 to $25 million per year. The deal covers NYT, NYT Cooking, and the Athletic, for use inside Alexa, Amazon's foundation models, and Amazon's consumer AI products. It does not give the Times presence inside ChatGPT, Claude, Perplexity, or Google AI Overviews.
Which news outlets have licensed content to OpenAI?
The largest deals include News Corp (reported by the Wall Street Journal to be worth more than $250 million over five years, covering the Wall Street Journal, the New York Post, Barron's, MarketWatch, the Times of London, the Sun, the Australian, and many others), Axel Springer (Bild, Politico, Business Insider), the Associated Press, the Financial Times, Dotdash Meredith, Condé Nast, Time, Vox Media, Reddit, Le Monde, and Prisa Media.
Is the Wall Street Journal cited inside ChatGPT even though it has a paywall?
Yes. The Wall Street Journal is paywalled to consumers but licensed to OpenAI through the News Corp deal. ChatGPT can surface WSJ content with attribution. This is a different model from the open-crawl approach — a paywall to readers, a contractual carve-out to AI.
Is the New York Post a Perplexity partner?
No. News Corp — parent of the New York Post, Wall Street Journal, and Dow Jones — sued Perplexity in October 2024 alleging "content kleptocracy." Inside Perplexity, the News Corp portfolio is currently an adversary, not a partner. This is the opposite of the News Corp posture inside ChatGPT.
What is Citation Share?
The share of AI-engine answers in a given category that cite a specific outlet, brand, or source. It is the replacement metric for traditional SEO share-of-voice in the answer-engine era — measured by sampling representative queries across ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews, then counting which sources are surfaced in the answers.
Will more publishers sign AI licensing deals?
Yes. The structural pressure to be inside the answer is intensifying as consumer behavior shifts further into AI engines. The economic terms are still being established, and the gap between aggressive licensors (News Corp, Axel Springer) and aggressive litigators (the New York Times) will continue to widen across 2026 and 2027.
The bottom line
Paywalls are not neutral. Inside the AI engines, they are a structural signal — and right now, they push paywalled outlets without licensing deals down.
The New York Post is positioned to be more retrievable inside ChatGPT than the New York Times. Not because the Post is doing better journalism. Because the Post is structurally accessible — open to the web and licensed to OpenAI — and the Times is structurally not.
This is the new hierarchy. It will harden. The brands and outlets that understand it — and adjust — will be the answer. The ones that don't will be the secondhand source the model cites once it finds an open version.
The chatbox is the new front page. Citation Share is the new circulation. Paywalls without a licensing strategy are the new dark zone.
