The internet did not arrive as a single event. It arrived as a series of small, dismissible changes. Classifieds moved. Then audiences. Then the ad buy. By the time the industry agreed something structural had happened, the structure belonged to someone else.
Then streaming did the same thing to broadcast. Viewers drifted to Netflix and YouTube a show at a time. Ad dollars followed. Ottawa passed the Online Streaming Act in 2023, a decade after the audience had left, and is still arguing over what the streamers owe.
Both times, policy arrived to negotiate terms for a market that had already moved.
The signs are visible again. This time they are moving faster: every six months, a new layer of how people find, read and buy shifts underneath us.
The new layer is the answer.
Search is becoming answers
Ask Claude, ChatGPT, Gemini or Google a question today and you increasingly get a paragraph, not a page of links.
Most of these systems now run a web search in the background, read the top sources and summarize them. The reader gets the answer. The source gets a footnote, if that.
The click is the unit the whole publisher economy runs on. It is disappearing. Ahrefs found that when an AI Overview appears, the top-ranked page gets 58% fewer clicks than it would have. Eight months earlier, the same study measured 34.5%.
The answers also change. Ask the same question twice and you get different answers. Memory, location and phrasing shift them further. There is no top position to hold anymore.
The platforms now own the measurement, access and the payment
In less than a year, the three things a publisher needs in this new layer moved into someone else’s systems.
Measurement. Microsoft launched an AI Performance report in Bing Webmaster Tools in February, showing how often Copilot cites your pages. Google followed in June with generative AI reports in Search Console. Both are free. Both show only their own platform. Neither show clicks.
Access. On Sept. 15, Cloudflare began blocking AI crawlers by default on ad-supported pages for new domains and free plans. A publisher can now become invisible to AI answers without anyone in the building making that decision.
Payment. Cloudflare is turning its Pay Per Crawl toll into Pay Per Use, which pays publishers when their content shapes an AI answer, not when a reader visits.
Each of these is useful. None of them is ours. The data, the switch and the cheque all sit on someone else’s servers.
That is flow, not stock. A publisher that depends on platform dashboards to know whether it exists is renting its own visibility.
Canada is regulating the last internet
The Online News Act was built for links and feeds. It was designed around platforms displaying news and readers clicking through.
It produced one deal. In October 2024, the CRTC granted Google a five-year exemption in exchange for $100 million a year, indexed, paid to the Canadian Journalism Collective. Meta blocked news instead, and has never paid.
That number was set when the click still worked. It runs to 2029. The click is collapsing now.
Now look at Meta.
In 2023, Meta pulled news off Facebook and Instagram in Canada rather than pay for it. Its argument was simple: news has little value to its platforms. It has held that line for three years. Local independent TV stations asked the CRTC in February to force it to the table.
Outside Canada, Meta has been paying for news the whole time, just not for feeds. It has signed AI licensing deals with CNN, Fox News, USA Today and People, following earlier deals with Reuters and Le Monde, so its chatbot can serve real-time news.
News has no value for the feed. It has real value for the answer.
Canada is now on the menu. Canadian Heritage’s own briefing note says Meta has signalled interest in deals with Canadian publishers for its AI products, and possibly joining the Online News Act framework, on one condition: an exemption. Ottawa has confirmed it is in talks with Meta about bringing news back, while Washington presses on the Act in trade talks.
In October, Meta revised its business terms in a way that led publishers to worry it was reserving rights to crawl their sites. Meta says the terms apply only to its ads and business products, and that publishers can use robots.txt. The Canadian Association of Broadcasters (CAB) called it an “obvious contradiction”: Canadian journalism blocked on Meta’s platforms while Meta seeks broad rights to the content.
Read that sequence again. Walk away from the feed. Pay for the answer elsewhere. Offer Canada an AI deal in exchange for an exemption from the law.
Meta is not confused about where the value is. Canada is.
That is the pattern. Ottawa negotiates the terms of the last platform shift while the next one sets its own terms.
When the source gets paid
Paying for citation sounds like the fix. It changes what an answer is.
Once sources are licensed, citation stops being purely editorial. It becomes partly commercial. The outlets with deals become the default voices. Everyone else becomes optional.
The reader cannot see the difference. A search ad is labelled. A paid source inside an answer is not. Nobody tells you whether the machine leaned on a source because it was best, or because it was paid. No one in Ottawa is asking that question yet.
The price is set by the platform, again. Pay Per Use pays when content shapes an answer. The platform defines what shaping means and what it is worth. That is not an asset. It is a meter someone else reads.
The incentives drift. If inclusion pays, newsrooms write for what machines can lift. Quotable facts over analysis. Volume over reporting. The cheapest content to cite is not the content most worth publishing.
And Canada faces a choice it has not named. The Online News Act model is one cheque, through one collective, for everyone. AI licensing runs the other way: private deals, confidential terms, a few large outlets at a time. If Meta trades AI deals with the biggest players for an exemption from the law, the collective model breaks. The small outlets that need it most get nothing.
Being paid to be cited is not leverage. It is a usage fee on content someone else has already captured.
The only escape is reporting nobody else has. A machine can find a substitute for everything else.
Where this goes next
You can see the path. Over the next 12 to 24 months:
- The ad moves inside the answer. Every one of these platforms will monetize, and sooner rather than later. When they do, the sponsored recommendation sits in the same paragraph as the editorial one.
- Measurement gets tied to money. Reporting will spread from Google and Microsoft to the rest. It will come bundled with ad and publisher programs, each built on the platform’s own data, each showing you only its own slice.
- Citation becomes the currency. Payment shifts from visits to inclusion. Being the source an answer is built on is what gets paid. Being summarized by someone who was the source is not.
- Small outlets disappear by default. Not by policy. By a crawler setting nobody checked, an unmarked paywall, a page the machine could not parse.
None of this needs a new law to happen.
Metadata is the new distribution
In the answer layer, your first reader is a machine. It decides what to retrieve, what to trust and what to quote before any human sees a word.
That machine reads metadata before it reads journalism: Who wrote this? When it was published and last updated? Who the publisher is, and whether that publisher is the same entity it has seen cited elsewhere? Whether the text is actually in the page or loaded later by script it never runs? Whether the paywall is marked, or looks like a page hiding something?
Get that wrong and the reporting underneath does not matter. The machine can’t find it, can’t date it, and cannot tell who stands behind it. So, it reaches for someone it can parse.
Broadcasters have seen this before. On streaming platforms, a show is only as discoverable as its catalogue data: title, genre, cast, origin. The fight over whether Canadian content gets surfaced has always been, underneath, a fight over metadata someone else controls.
The answer layer is the same fight, one level down. The difference is that this time the metadata sits on our own sites. It is the one part of the stack Canadian media fully owns.
Most of it is still badly built. That is the cheapest leverage in this entire story.
What operators should do this quarter
The structural move is to become the source the machines can’t route around. That is stock. Five things, none of which require Ottawa:
- Decide your crawler policy on purpose. Check your robots.txt and your Cloudflare settings today. Know which AI crawlers you allow, which you block, and why. A default is not a decision.
- Take the free data. Verify your sites in Bing Webmaster Tools and Google Search Console. Read the AI reports monthly. It is incomplete, but it is real usage, not a simulation.
- Make your content machine-readable. Article and publisher schema with authors and dates. Text in the raw HTML, not loaded by script. Have accurate sitemaps and feeds. This is table stakes, not a growth plan.
- Open the headline, sell the analysis. If your reporting sits fully behind a paywall, the machine cites whoever summarized you. Leave the headline and key facts crawlable, mark the paywall properly, and keep the analysis paid.
- Measure yourself. Pick 20 questions your audience actually asks. Run them through the major AI tools every month. Log who gets cited. That is your own visibility score, and you own the data.
Underneath all five: publish reporting nobody else has. Structure gets you read. Originality gets you cited.
The source or the summary
Twice now, we waited until the structure belonged to someone else.
This time the window is shorter and the warning is clearer. The answer layer is being built right now, and Canadian content is going into it whether Canadian media is at the table or not.
Being the source is leverage. Being summarized is not.




