This whole book leans on a quiet assumption: that somewhere out there, people whose job is to find out what is true are still doing it. When the manuscript checks a government claim, cites a court ruling, or tells you a corporate disclosure is technically honest but selectively scoped, it is standing on the shoulders of reporters, editors, and outlets who did the original digging. Journalism is not just another industry AI happens to touch. It is one of the mechanisms the rest of the book relies on to check AI's claims in the first place. That mechanism is under more financial strain than at any point in living memory, and AI arrived in the middle of the strain, hitting from two directions at once. Inside the newsroom it shows up as a tool that can genuinely help and can also be used as a reason to employ fewer people. Outside the newsroom it shows up as a flood of synthetic "news" that competes for the same attention and the same trust, costs almost nothing to make, and does not employ anyone at all. You will leave this chapter able to separate three things people constantly blur: the long financial collapse of the news business (which started before generative AI), what AI is actually doing inside newsrooms (less than the hype, more than nothing), and what AI is doing to the information supply around journalism (a real and growing problem). And you will be able to ask the question this book keeps asking: of the mechanisms meant to hold any of this accountable, which are enforceable, and which are just promises?
The floor was already gone
Start with the money, because almost every confused argument about "AI killing journalism" is really an argument about a business that was already in free-fall when AI showed up.
Statistics Canada's accounting is blunt. In its release on newspaper publishers covering 2024, total operating revenue was $1.6 billion, down 17.9% from 2022. Advertising sales fell 26.1% to $722.8 million in two years, with print advertising down more than a third. This is the financial base of a lot of Canadian reporting, and it is shrinking fast. None of that decline is generative AI. It is the long migration of advertising to platforms that do not produce journalism, playing out on a Canadian balance sheet.
Then came a specifically Canadian twist that the rest of the world watched with interest. In 2023 Parliament passed Bill C-18, the Online News Act, which received royal assent on June 22, 2023, with final regulations released that December. The law's logic was that the big platforms profiting from news links should pay the outlets that produce the news. Meta's answer was to stop carrying Canadian news entirely. Since August 2023, links to Canadian news have been blocked on Facebook and Instagram, and as of late 2025 the regulator confirmed the block was still in place with no enforcement action planned. Canada is the country where a platform looked at a payment law and chose to remove the news rather than pay for it. That is a globally distinctive data point, and it is ours.
Google made the other choice. Rather than block news, it negotiated an exemption: roughly $100 million a year, indexed to inflation, paid into a single collective (the Canadian Journalism Collective) that distributes the money to outlets, in exchange for a five-year pass from the Act's mandatory bargaining. Whether that is a fair price for the value of Canadian journalism, or a modest sum that buys a large company out of a larger obligation, is exactly the kind of question this book wants you to hold open rather than answer by reflex.
Underneath the platform fights, the local map keeps going dark. The Local News Research Project, which has tracked Canadian outlet closures since 2008, counted 603 local news operations closed in 388 communities by late 2025, against only a fraction relaunched and surviving. "News desert" is the term for a community with no dedicated local coverage, and the deserts are spreading. This matters for AI specifically, because a thinned-out local press is a thinned-out supply of the verified local facts that everything downstream, including AI systems trained on the web, depends on.
🧌 GOBLIN CHECK — blame the wound, not the bystander. When a newsroom cuts jobs in 2026 and mentions AI in the same breath, it is tempting to read it as "AI took the jobs." Usually the deeper cause is older and duller: advertising left a decade ago, the platforms restructured the audience, and the local map was already emptying. AI is arriving as the newest pressure on an industry that was already bleeding, and sometimes as the convenient name for a cut that the balance sheet had already made inevitable. Watch which one you are actually looking at.
AI in the newsroom: less than the hype, more than nothing
Inside Canadian newsrooms, the real story is more boring and more honest than either the boosters or the doom-criers want. AI is in use, but mostly at the edges, and the serious outlets have written rules to keep it there.
The common uses are unglamorous: transcribing interviews, translating copy, searching large document dumps, summarizing for internal research. These are real time-savers and they are not, on their own, a threat to anyone's byline. The published guidelines tell you where the lines are drawn, and they are worth reading as what they are: corporate self-disclosure, useful but also a form of reputation management, so read them for what they promise and for what they carefully avoid promising.
CBC's guidelines state the principle plainly: "CBC News employees, not AI tools, create and are responsible for our journalism," and the broadcaster says it will not use AI at the first stage of composing anything meant for the audience. It carves out a narrow exception for transforming existing content, such as text-to-speech or closed captions, where a human need not be in the loop before publication. The Canadian Press frames any AI output as "unvetted source material" that must be checked word by word, and bars using AI to rewrite or edit its copy. The Globe and Mail says it will not use AI to write or edit stories or for photojournalism; its head of newsroom development, not the policy page, supplied the blunt version that the paper is "not going to use it to write our stories." In Quebec, the Conseil de presse adopted a deontological principle in 2024 requiring that a human exercise editorial control over any AI-generated content before it reaches the public.
Notice what these policies share: a human-responsibility line at the point of publication, and near-silence on the business decisions upstream. That gap is where the honest worry lives. The same outlet that promises a human will always be responsible for the journalism can still decide, for reasons that have nothing to do with any guideline, that it needs fewer humans to produce it.
And the automation is already deeper than most readers think on the distribution side. The Globe and Mail's in-house system, Sophi, places more than 99% of the content on the paper's digital pages, deciding what you see where. That figure comes from the system's operator, not an independent audit, so treat the number as a self-report. But even discounted, it tells you that the editorial judgment of "what leads, what gets buried" is already substantially automated at a major Canadian paper, and has been for years, well before the generative-AI panic.
The labour pressure is real and mostly arrives through the front door, not the algorithm. When Bell Media announced sweeping cuts in February 2024, Unifor described it as roughly 4,800 jobs company-wide, with around 100 of its media members affected; a smaller round in early 2026 cut about 20 more Unifor members, including 11 journalists. (The 4,800 figure is the union's characterization of Bell's announcement; no Bell press release confirming it was located, so attribute it to Unifor.) When Corus restructured Global News in 2024, Unifor counted 35 members cut and warned of "bigger swaths of news deserts." Most of these cuts are about debt, advertising, and consolidation, not robots. But they hollow out the same newsrooms that AI is now being pitched into, which is why workers are right to ask what incentive a new "efficiency" tool is walking into.
ALIGNMENT — efficiency, or evacuation? When AI gets pitched to a newsroom, the sell is always speed: file faster, cover more with less. The question the pitch skips is whether "less" means a journalist freed to do harder reporting, or a journalist who is no longer there. Same tool, two completely different outcomes, and the slide deck rarely says which one is funding the purchase.
The training-data fight
There is a second front, and it runs straight into the copyright contest from Chapter 11. The models that now summarize the news back to readers were trained, in part, on the news. In November 2024 a coalition of Canada's largest news organizations, including CBC/Radio-Canada, The Canadian Press, Torstar, The Globe and Mail, Postmedia, and Metroland, sued OpenAI in the Ontario Superior Court of Justice, alleging it scraped their journalism to train ChatGPT without permission or payment, and seeking damages.
In November 2025 the publishers won an early and important round: the court ruled the case can proceed in Ontario rather than being pushed into a US forum. OpenAI has appealed that jurisdiction ruling, and the merits, whether the training was lawful, remain undecided. Chapter 11 handles the underlying copyright doctrine, the text-and-data-mining argument, and where Canadian law actually stands. The point here is narrower and about power: a newsroom's own work can become the training fuel for a system that then answers the questions readers used to come to that newsroom to ask, sending no traffic, and no money, back. When the summary of the story competes with the story, the economics of finding things out get worse.
The flood
The third front is the one most people mean when they say "AI and the news," and it is the one where the alarm is most justified, as long as we are precise about it. Generative tools have made it cheap to manufacture things that look like news and are not.
During the 2025 federal election, the Canadian Digital Media Research Network documented AI-generated ads impersonating real Canadian news brands, including CBC and CTV, to push investment and crypto scams; one French-language ad used the CBC/Radio-Canada logo. The same researchers identified seven deepfake videos of Mark Carney mimicking news interviews, spread across a network of fake accounts. Separately, researchers traced an "AI slop" operation of dozens of YouTube channels posing as Canadian news outlets; one analysis (DFRLab) counted 42 channels and 771 videos with millions of views before the platform removed most of them, pushing narratives about election fraud and Alberta separatism. CBC and Canadian researchers gave the Alberta-separatism strand its own name, "slopaganda," for the inauthentic network selling secession to Albertans. (This connects to the Cipher AI tracking work in Chapter 13: foreign and domestic actors amplified the separatism narrative, but, by the researchers' own account, did not manufacture the underlying movement.)
Chapter 13 covers whether people can actually detect this material (mostly, they can't reliably) and the regulatory tools aimed at it. The angle that belongs here is the economic one. A synthetic news channel costs almost nothing to run and competes for the same attention and trust as a real newsroom that has to pay reporters, lawyers, and editors. When the fake is free and the real thing is expensive and shrinking, the information commons fills up with the cheap stuff. That is not just a misinformation problem. It is a market problem, and the market is tilting against the people doing the verified work.
EXAMPLE — Gresham's law for the feed. There is an old idea in economics that bad money drives out good: if two coins are worth the same at the counter but one is debased, people spend the bad one and hoard the good. Synthetic news works a little like that. When a real article and a convincing fake compete for the same click at the same price (free), the cheap-to-produce fake floods the channel, and the expensive-to-produce real reporting gets harder to find and harder to fund. The feed doesn't reward truth. It rewards volume.
The ladder
Put the accountability mechanisms in a row and run them up this book's enforceability ladder, the one that runs from promised but not funded through funded but not regulated to regulated but not enforceable.
Bill C-18 is real law and it moved real money, but only after one platform left rather than pay and another bought a five-year exemption, so its reach is partial and time-limited. The newsroom AI guidelines are genuine commitments, but they are voluntary self-policing: an outlet can rewrite or ignore them tomorrow, and they say almost nothing about the staffing decisions that worry workers most. The OpenAI lawsuit is an accountability mechanism still working its way through the courts, with the central question unresolved and under appeal. And against the flood of synthetic news, the live tools are thin: platform moderation that acts after the fact, and the deepfake provisions of Chapter 13's Bill C-16, which reach sexual deepfakes but not the impersonation-for-scams or the slop networks. Almost everything here sits at promised, funded-but-partial, or contested. Very little is enforceable.
That is the synthesis the evidence supports, and it is not a counsel of despair. Chapter 20 lays out concrete mechanisms Canada could actually reach for, and several of them, mandatory AI-use disclosure, real teeth on synthetic-content labelling, durable funding models for local news, would land directly on the problems in this chapter. The point of naming the gaps is not to mourn. It is to know where the load-bearing walls are missing before someone tells you the house is fine.
The working test
For any claim about AI and the news, ask:
Cause. Is AI actually responsible for this cut or closure, or is it the older collapse (advertising, platforms, consolidation) wearing AI's name?
Inside or outside. Is this about AI used by a newsroom (a tool, governed by a guideline) or AI used against the information supply (synthetic news, impersonation)? They are different problems with different fixes.
Self-report or audit. Is the reassuring number (a usage stat, a "human always in the loop" promise) coming from the company itself, or from someone independent?
Enforceable or promised. Is the accountability mechanism a binding requirement, or a voluntary commitment that can change tomorrow?
Who pays to find out. Follow the money to the verified facts. If the real reporting gets harder to fund while the synthetic stuff stays free, the problem is structural, not a vibe.
CHAPTER RECAP — you now have: - The financial floor mapped: a news business already in free-fall (StatCan revenue down 17.9%, ad sales down 26.1% over two years) before generative AI arrived. - The Canadian platform story: Bill C-18, Meta's still-standing news block since 2023, and Google's ~$100M/year exemption deal — with 603 local outlets closed since 2008 as the backdrop. - AI inside the newsroom: real but bounded use, the CBC/CP/Globe/Conseil de presse guidelines read as self-disclosure, Sophi's 99% automated placement, and layoffs that are mostly about debt, not robots. - The training-data fight: Canada's news coalition v. OpenAI, the 2025 jurisdiction win now under appeal, handed to Chapter 11 for the copyright doctrine. - The flood: AI-impersonation scams and "slop"/"slopaganda" networks in the 2025 election, framed as an economic attack on the information commons, with detection handed to Chapter 13. - The enforceability ladder applied: most mechanisms sit at promised, partial, or contested — very little is enforceable — with the constructive options waiting in Chapter 20.
The next chapter (Chapter 15) turns from the institution that reports the facts to the question that haunts every AI system underneath them: bias. Not the statistical kind the vendors love to benchmark, but the structural kind — how a system can be mathematically "fair" and still reproduce the power it was built inside. The news chapter showed you an information supply under economic pressure. The ethics chapter asks what the machines do with the information they are fed.
Bias label for this chapter: structural analysis of the Canadian news and information economy as AI enters from both inside the newsroom and outside it. Author lean: written sympathetic to working journalists and the public value of verified reporting, and skeptical of both the "AI will save the newsroom" efficiency pitch and the "AI killed journalism" panic that lets a decade of advertising collapse off the hook. Government and regulator sources (Statistics Canada, the Online News Act and its regulations, the CRTC) treated as primary on the economic and legal framework. Newsroom AI guidelines treated as corporate self-disclosure — read for what they do not commit to. Union sources (Unifor) treated as primary on layoffs and advocacy on framing. Research-network and think-tank sources (CDMRN, DFRLab) treated as method-dependent findings on synthetic-media operations. Platform statements (Meta, Google) treated as self-interested.
Primary sources cited or relied on in this chapter: Statistics Canada, "Newspaper publishers, 2024" (The Daily, November 10, 2025); Bill C-18 (Online News Act), royal assent June 22, 2023, and final regulations (December 2023 / Canada Gazette, Part II, January 3, 2024); Canadian Heritage releases on the Online News Act and the Google exemption; the CRTC's December 2025 statement on the Meta news block; the Toronto Metropolitan University Local News Research Project / Local News Map closure data (October 2025); CBC News, The Canadian Press, and The Globe and Mail published AI guidelines, and the Conseil de presse du Québec AI principle (2024); Mather Economics / Sophi disclosure on automated content placement at The Globe and Mail; Unifor releases on Bell Media (February 2024) and Corus/Global News (2024) layoffs; the Canadian news coalition's statement of claim against OpenAI (Ontario Superior Court, November 2024) and reporting on the November 2025 jurisdiction ruling and appeal; and the Canadian Digital Media Research Network and DFRLab analyses of AI-impersonation and "slop"/"slopaganda" networks during the 2025 federal election. Detailed citations in the Sources appendix.
---
🧌 GOBLIN CHECK — blame the wound, not the bystander. When a newsroom cuts jobs in 2026 and mentions AI in the same breath, it is tempting to read it as "AI took the jobs." Usually the deeper cause is older and duller: advertising left a decade ago, the platforms restructured the audience, and the local map was already emptying. AI is arriving as the newest pressure on an industry that was already bleeding, and sometimes as the convenient name for a cut that the balance sheet had already made inevitable. Watch which one you are actually looking at.
Recap
- The financial floor mapped: a news business already in free-fall (StatCan revenue down 17.9%, ad sales down 26.1% over two years) before generative AI arrived.
- The Canadian platform story: Bill C-18, Meta's still-standing news block since 2023, and Google's ~$100M/year exemption deal — with 603 local outlets closed since 2008 as the backdrop.
- AI inside the newsroom: real but bounded use, the CBC/CP/Globe/Conseil de presse guidelines read as self-disclosure, Sophi's 99% automated placement, and layoffs that are mostly about debt, not robots.
- The training-data fight: Canada's news coalition v. OpenAI, the 2025 jurisdiction win now under appeal, handed to Chapter 11 for the copyright doctrine.
- The flood: AI-impersonation scams and "slop"/"slopaganda" networks in the 2025 election, framed as an economic attack on the information commons, with detection handed to Chapter 13.
- The enforceability ladder applied: most mechanisms sit at promised, partial, or contested — very little is enforceable — with the constructive options waiting in Chapter 20.
Sources
- Statistics Canada, "Newspaper publishers, 2024" (The Daily, November 10, 2025)
- Bill C-18 (Online News Act), royal assent June 22, 2023, and final regulations (December 2023 / Canada Gazette, Part II, January 3, 2024)
- Canadian Heritage releases on the Online News Act and the Google exemption
- the CRTC's December 2025 statement on the Meta news block
- the Toronto Metropolitan University Local News Research Project / Local News Map closure data (October 2025)
- CBC News, The Canadian Press, and The Globe and Mail published AI guidelines, and the Conseil de presse du Québec AI principle (2024)
- Mather Economics / Sophi disclosure on automated content placement at The Globe and Mail
- Unifor releases on Bell Media (February 2024) and Corus/Global News (2024) layoffs
- the Canadian news coalition's statement of claim against OpenAI (Ontario Superior Court, November 2024) and reporting on the November 2025 jurisdiction ruling and appeal
- and the Canadian Digital Media Research Network and DFRLab analyses of AI-impersonation and "slop"/"slopaganda" networks during the 2025 federal election.