McClatchy cut 90 jobs in September 2026 at newspapers including The Miami Herald, The Fresno Bee, and The Charlotte Observer. Poynter reported that some newsrooms lost as much as 40 percent of their staff in the round, reporters and photographers and videographers together. LinkedIn News wrote it up. The writeup was a summary of Poynter’s reporting, padded out with sympathy quotes lifted from LinkedIn posts, running eighteen inches from a promoted post for Castrol engine coolant.
Twenty years before that, in 2006, McClatchy bought Knight Ridder for $4.5 billion. It took on the debt and the underfunded pension obligations to do it. The deal closed one year after American newspaper advertising revenue hit its all-time high of $49.4 billion. By 2020 the same figure was $8.8 billion.
The 90 jobs are the last visible twitch of a leveraged bet placed at the top of the market by executives who are no longer at the company.
The order of events is the whole argument
Two explanations compete for what happened, and both are popular because both let their holders off the hook. People inside the industry say the platforms stole the money. People outside it say bias drove the audience off. The first is largely true. The second is partly true. The sequence they get told in is backwards, and the sequence is the only part that matters, because it determines whether any of this is recoverable.
The money went first. It went for structural reasons nobody in a newsroom could have voted against and no editor could have prevented. What the industry controlled was what it did next. What it did next was swap the expensive input for the cheap one, and that swap is what cost it the integrity. The integrity loss cost it the value of the product. The value loss is why the last business model available, asking readers to pay, doesn’t work and won’t.
Money, then integrity, then value, then no money ever again. The first step was not anybody’s fault. The second one was.
The quarter on the driveway never paid for the newsroom
The Sunday Los Angeles Times landed on my front porch for years. The daily ran a quarter. Nothing about that transaction paid for the newsroom, and it was never supposed to.
Circulation revenue was always the minority of the business, and it barely moved through the collapse. It ran $10.4 billion in 2012 and $10.9 billion in both 2015 and 2016, holding near $11 billion into 2020, propped up by raising prices on a shrinking base. Advertising is what fell off the table. In 2020, for the first time in roughly sixty years, American newspapers made more money from circulation than from advertising.
So the thing readers were paying for was never the thing that funded the reporting. What a newspaper actually sold was scarcity. A car dealer in Fresno had exactly one efficient way to reach Fresno households, and that geographic monopoly on local attention was the entire product. The city hall reporter was a cost center underwritten by classifieds, grocery inserts, and dealer ads. Print classifieds alone ran about $19.6 billion in 2000. The Sunday edition was the highest-margin object in the business, and the preprint inserts stuffed inside it were the reason. Sunday didn’t pay for Sunday. It paid for Tuesday, Wednesday, and Thursday.
Then the bundle came apart in pieces. Classifieds went to Craigslist. Jobs went to Monster and then Indeed. Houses went to Zillow, cars to Autotrader, stock tables and TV listings to anywhere at all. Every vertical got taken by a dedicated product better at that one thing, and not one of those products had any interest in paying for a courthouse reporter. At the same time ad inventory on the open web turned out to be effectively infinite. Scarcity price became commodity price. Google and Meta then took the targeting layer, so an advertiser stopped having to buy Fresno Bee readers and started buying the specific person, anywhere, cheaper.
Craigslist never read the editorial page. None of that first stage is a bias story, and anyone selling it as one is selling something.
The PennySaver had no editorial page. It died anyway.
If the collapse were fundamentally about ideology, the parts of the business carrying no ideology at all should have come through it fine. The test is available, and it has already been run twice.
A directory salesman named Bob DeMarco founded the PennySaver in Huntington Beach in 1962. Harte-Hanks bought it in 1973 and moved the operation to Brea. At its height it mailed roughly nine million copies a week to California households across 783 hyperlocal zones, and as late as 2015 fifty dollars a week bought fifteen words in front of a hundred thousand people. There was no editorial page in it. No reporters, no columnists, no wire desk, no opinions of any kind, just classified ads and coupons in an envelope. Harte-Hanks sold the business to OpenGate Capital in 2013 for $22.5 million. In May 2015 a lender pulled the credit line, and about 700 employees were told on a Friday afternoon to pack their desks and leave. They filed a class action the following Tuesday over the sixty days of notice California law required. Nobody ever accused the PennySaver of liberal bias. Groupon killed it anyway.
The distinction matters, because Craigslist and Groupon ate different meals. Craigslist took the newspaper’s paid classifieds, the individual selling a couch, the jobs column, the auto listings. The PennySaver’s business was the other one. A carpet cleaner or a taqueria buying a coupon into the mailboxes of nearby households, fifty dollars for fifteen words in front of a hundred thousand people. Groupon sold a local merchant with unused capacity exactly that transaction, targeted better, priced on performance, with a redemption count at the end of it. PennySaver active advertisers numbered in the thousands. Groupon’s active customers peaked above 50 million in the fourth quarter of 2014. The PennySaver was dead six months later.
Groupon didn’t keep it either. Its revenue topped out just over $3 billion in 2016 and came in around $493 million in 2024, with active customers down to 15.4 million. The local merchant budget never settled anywhere. It went to Groupon, then to Facebook and Google, and it is going somewhere else now. Not one stop on that route has an editorial page.
TV Guide ran the same arc at larger numbers. Annenberg sold Triangle Publications, TV Guide included, to Rupert Murdoch’s News Corporation in 1988 for $3 billion. News Corp sold TV Guide Inc. to Gemstar in 1999 in a stock deal valued around $9.2 billion. Then cable boxes began printing the listings on the screen for free. Paid circulation fell from 8.2 million in 2005 to 3.3 million in 2007, and Gemstar’s 10-K put the magazine’s loss for 2007 at $20.3 million. In October 2008, Macrovision sold the print title to OpenGate Capital for one dollar and loaned OpenGate up to $9.5 million at 3 percent to keep the lights on.
The cover price that week was $2.99. One issue cost three times what the company did. OpenGate bought both of these, incidentally, TV Guide for a dollar in 2008 and the PennySaver for $22.5 million five years later.
In 2000 I wrote the ASP for usetheexperts.com at Web Business Development in Chatsworth, a consumer-to-local-business directory. My friend Lowell did the design. You typed in what you needed and the site handed back local businesses that did it. The PennySaver was at that moment mailing nine million copies a week across the same state selling precisely that transaction on paper, and what I was building was one of several thousand attempts nationally to move it into a browser. Most of them failed, mine included. Enough of them worked. Nobody in that office was thinking about the Los Angeles Times, and the Times paid for it regardless.
That is the whole first stage. Listings, classifieds, coupons, and car ads left for products that did one job better, and none of those products cared what any newsroom thought about anything.
The people who threw the paper had no opinions in the product either. California newspaper carriers were classified as independent contractors for decades, driving their own cars, buying their own gas, paid by the piece, no benefits attached. That arrangement is the gig economy, and it predates the term by about forty years. When AB 5 took effect in 2020 and threatened to convert carriers into employees, the California News Publishers Association put the cost to the industry at a minimum of $80 million, and its chief executive Chuck Champion said publishers had told him they would not survive it. Carriers got a one-year reprieve, then another one under AB 323, then an extension running to January 2025 under AB 1506. Uber, Lyft, DoorDash, Instacart, and Postmates took a different route, spending roughly $200 million on Proposition 22 to write themselves a permanent exemption, which California voters approved in November 2020 and the state Supreme Court upheld in 2024.
AB 1506 also had to fix a definition. A newspaper carrier, the bill specifies, is a person who is not working as an app-based driver during the time they are performing the newspaper delivery. The legislature had to put in writing that those are two different jobs.
Opinion costs one salary. Reporting costs a bureau.
A columnist costs a salary, a laptop, and no travel. A statehouse bureau costs salaries, mileage, records fees, and a lawyer on retainer for the week the records request comes back denied. When revenue falls against fixed debt service, the accounting picks. It picks the same way every time.
Newsroom employment across newspapers, radio, broadcast, cable, and digital fell from roughly 114,000 in 2008 to about 85,000 in 2020. Newspapers on their own went from 71,000 to 38,000 between 2008 and 2018. Full-time newspaper statehouse reporters dropped from 374 in 2014 to 245 in 2022. On the other side of the table, the Bureau of Labor Statistics counted roughly 283,000 public relations specialists against about 39,000 reporters in 2025. That is somewhere near seven to one, up from roughly three to one in 2004. The occupational categories blur at the edges and people argue about the exact ratio, but nobody argues about the direction.
What survives a cut like that is whatever costs the least per column inch. Opinion, aggregation, and rewrites of other outlets’ reporting all cost close to nothing, and all three scale without a travel budget.
The Duke News Measures Research Project put numbers on the result before most people noticed there was a result. Researchers content-analyzed more than 16,000 stories across 100 randomly sampled American communities. Forty-three percent were original to the outlet that published them. Seventeen percent were actually about the community. Fifty-six percent addressed what the team defined as a critical information need. Apply all three tests at once and roughly 11 percent of the stories were local, original, and served a real informational purpose. In twenty of the hundred communities, there were no local stories at all during the sampled week.
That study sampled 2016. It is the before picture.
Most journalists come from the same side of the aisle
The Syracuse Newhouse survey of 1,600 American journalists in 2022 found 3.4 percent identifying as Republican, 36.4 percent as Democrat, 51.7 percent independent. The same survey series in 1971 found 25.7 percent Republican. The country splits roughly 29, 29, and 40. Among journalists who name a party at all, Weaver and Willnat put the Democratic share close to 79 percent.
The mechanism is not a hiring conspiracy, and treating it as one is exactly why the complaint keeps failing to land with anyone who isn’t already convinced. Journalism converted itself into a credentialed profession. Degree required, unpaid internship as the entry toll, jobs concentrated in the most expensive metros in the country. Pew has found that highly educated adults, and graduate degree holders in particular, are substantially more likely to hold liberal positions. David Brooks described starting out in Chicago in the 1980s with crusty working-class guys still around the newsroom, and the profession becoming not just college-dominated but elite-college-dominated. A class filter produced an ideological result. Nobody had to plan it, and no HR department implemented it.
The counterevidence deserves to be stated plainly, because pretending it doesn’t exist is how you lose this argument to the first person who has read it. Hassell, Holbein, and Miles published a study in Science Advances in 2020 that inferred ideology for 6,801 journalists and then ran a field experiment mailing press releases about candidates from both parties. They found individual ideology had strikingly little effect on which stories reporters chose to pick up. Gentzkow and Shapiro found that outlet slant tracks reader demand more closely than it tracks owner or reporter ideology, which is the market explanation and it has teeth.
Both of those test gatekeeping at the story-selection stage for one specific kind of input. Neither tests what counts as a story in the first place, or which expert gets the phone call, or whether a claim gets the questions-raised treatment or the debunked treatment. That is where the actual complaint lives, and it is the part that has never been measured well in either direction.
Which is why the 3.4 percent matters less as proof of slant than as a description of a monoculture. A newsroom where half the country is represented at three percent will produce work that is perfectly sound on most stories and blind on a specific set of them, and it will have no internal mechanism for knowing which set. That was survivable while the product was mostly reporting, because facts push back. It stopped being survivable when the product shifted toward commentary, because nothing pushes back on a take except an editor who disagrees with it, and the newsroom no longer contains one.
The clearest documented capture went the other direction
Ownership deserves its own accounting, because the best-evidenced case of ideological capture in American local news ran right rather than left. Gregory Martin and Joshua McCrain analyzed 743 stations for the American Political Science Review and found that Sinclair acquisitions produced roughly a 25 percent relative increase in national political coverage at the expense of local coverage, a measurable rightward shift in slant, and a small decline in viewership. Sinclair reaches close to 40 percent of the American population. Their later work with Mastrorocco and Ornaghi shows the effect is specific to the owner rather than a general consequence of consolidation. Sinclair cut local coverage about 10 percent. Nexstar raised it about 8 percent. Gray didn’t move it at all. Separately, Metric Media built an estimated 1,200 websites dressed as local outlets to promote right-wing candidates.
Reporters don’t capture local news. Owners do, in whichever direction the owner is pointed, and the reporters are downstream of a decision made in a boardroom they’ve never been in.
Seventy percent, and falling
Pew’s September 2025 numbers put trust in national news organizations at 56 percent, down 11 points since March of that same year and 20 points since 2016. Republicans sit at 44 percent. Local news does considerably better at 70 percent. It was 82 percent in 2016. An AP-NORC survey found 74 percent of American adults believe the news media is widening the country’s political divide.
Then the number that converts trust into cash. Pew found 12 percent of Americans paid or gave money to any local news source in the past year, down from 15 percent. Of the ones who didn’t, half said the reason was that they can find plenty of local news for free. That isn’t a price objection. It’s a verdict on the product, delivered by people who did the comparison and concluded the free substitute was close enough.
Meanwhile the supply side kept contracting. Medill counted 213 counties with no local news source at all in 2025, another 1,524 counties down to a single outlet, and 136 newspapers closed in the preceding year. Roughly 50 million Americans now live with limited or no access to local news.
A summarizer inherits whatever it summarizes
Set the aggregation layer on top of that corpus and the arithmetic gets ugly.
Zero-click searches hit 68 percent in early 2026, up from 60 percent in 2024 and 49 percent in 2019. Chartbeat data provided to Axios in March 2026 shows search referral traffic over two years down 60 percent for small publishers, 47 percent for mid-sized ones, and 22 percent for the largest. Google referrals across more than 2,500 news sites fell 34 percent between December 2024 and December 2025, with Discover down another 15. Pew found readers click a result 8 percent of the time when an AI summary is present, against 15 percent when it isn’t. Referrals from AI chatbots grew more than 200 percent over the same window and still account for under 1 percent of publisher traffic.
Notice the gradient. Sixty percent for small publishers, 22 for large. The loss is regressive by size, which means it lands hardest on exactly the outlets that were already down to one reporter.
The licensing money that’s supposed to backfill this goes to about two dozen companies. News Corp’s OpenAI agreement was reported at more than $250 million over five years, Dotdash Meredith collects at least $16 million annually in fixed payments, and Amazon’s deal with the New York Times was put somewhere between $20 and $25 million a year. No county paper is in that conversation, and every serious analysis published this year concludes the long tail will see nothing meaningful.
I build RAG pipelines in production. Retrieval quality is capped by the corpus, and no model upgrade changes that. If the documents in your index are 43 percent original and increasingly weighted toward commentary, what the system returns is a confident synthesis of commentary. Compression makes it worse rather than better, and this is the part people building news summarizers keep missing. A 1,200-word column carries hedges, attributions, a dissenting quote three paragraphs down, and a byline belonging to someone who can be argued with. Summarize it to 80 words and the hedges go first, because hedges are the least information-dense text in the document and every extractive objective in the stack is tuned to drop them. What comes out the other end reads like a finding. It went in as a take.
That is the actual laundering mechanism, and it requires nobody at any layer to intend it. The model isn’t biased. The index is, and the model is doing its job.
The asset was the reporting
The industry’s remaining plan is to charge readers for a product that half of non-payers already describe as freely available elsewhere, produced by institutions that 44 percent of Republicans don’t trust, distributed through channels that cut their referral traffic by 60 percent, and summarized downstream by systems that pay nothing and strip the attribution in transit.
There is one more twist waiting. Audiences rate news labeled as AI-generated as less trustworthy even when they don’t rate the same articles as less accurate or less fair, a finding Toff and Simon produced using actual AI-generated journalistic content. Fletcher and Nielsen measured 58 percent trust for human-written news against 14 percent for AI-written. So the cost pressure pushes toward automation, the disclosure norm punishes it, and hiding it risks a CNET-style blowup that costs more than the savings. Prism News found the third door in August 2026 and shut down about 200 AI-generated local news sites after reporters caught it scraping and distorting their work.
Reporting was the asset. It was the only thing in the bundle that couldn’t be reproduced by somebody who wasn’t there. Everyone had opinions, including the readers, and opinions were never scarce enough to sell. What the paper had that nobody else had was a person in the room at seven on a Tuesday night when the planning commission voted on the zoning variance.
They cut that person. They kept the opinion, because it was cheaper and because traffic rewarded it for about a decade. Now the traffic is gone too, and what’s left to monetize is a corpus of takes that an aggregator can compress to eighty words for free.
There’s a second thing in that bundle worth naming before it disappears completely, and it has nothing to do with slant. The paper on the driveway delivered the planning commission story whether or not anybody wanted it. That was the function. A reader bought the thing for the coupons or the box scores and absorbed four column inches about a zoning variance on the way past, and enough readers absorbed enough column inches that a city manager had to assume somebody was watching. Aggregation cannot do this and never will, because aggregation requires scale to justify itself and nothing about a school board in a town of forty thousand scales. A feed shows people what they already selected for. The paper showed them what they didn’t.
Every proposed fix currently on the table addresses the wrong half of that. Paywalls, pay-per-crawl, collective licensing, and the litigation all try to restore excludability over text. Copyright was never designed to protect facts, deliberately, and the moment a story publishes anybody can lawfully restate what it established. Newspapers never solved that problem either. They just owned the delivery truck, which made the problem irrelevant for a century and a half. A sufficiently good paraphraser routes around every remedy on that list, and a sufficiently good paraphraser is a precise description of what a language model is.
Nobody ever paid a quarter to find out what the staff thought.
About the Author
Gal Ratner has spent nearly thirty years building and shipping production software on the Microsoft and .NET stack, for clients including Microsoft, Sony, Rockstar Games, 2K Games, Best Buy, and Allegiant Air. He wrote the ASP for usetheexperts.com in 2000, one of the early attempts to move the local business directory off paper and into a browser, and was the sixth employee at Break.com during its run as one of the largest entertainment sites on the web, which is where he learned what advertising-funded content economics look like from the inside. He is Chief Architect at Prana Entertainment, the founder and CTO of Inverted Software and WhiteStar Labs, an enterprise software and AI consultancy based in Las Vegas, and a finalist for the Los Angeles Business Journal’s CTO of the Year.
His current work is agentic AI in production: MCP servers, the Microsoft Agent Framework, RAG pipelines over messy enterprise corpora, SQL Server 2025 vector search, and the open-source diagnostics tooling behind PLogger. He writes about AI the way a practitioner sees it, bullish on what the tools can actually do and allergic to the marketing wrapped around them.
galratner.substack.com · linkedin.com/in/galratner · galratner.com