SEO nearly destroyed journalism. Could AI bring it back?

AI destroys the business of repackaging information that already exists. Whether that saves journalism or finishes it depends on something nobody has solved.

I worked in digital journalism during the period when publishing increasingly became a game of feeding Google.

Looking back, I think it hollowed a lot of journalism out. Not all at once, and SEO wasn’t solely responsible. But the incentives became brutal.

Find something another publication had reported. Rewrite it quickly. Get the keywords in. Publish. Chase the traffic. Repeat.

You could spend hours speaking to people, digging through documents and trying to find something genuinely new. Or you could rewrite something that already existed and probably get more traffic for it.

The economics favoured the second option, increasingly and then overwhelmingly.

That’s a large part of why I left journalism relatively quickly. The part of the job I found interesting was finding the story — talking to people, spotting the angle, working out what was actually going on. Digital publishing was rewarding speed, volume and aggregation. I didn’t have the language for it then, but I could see where it was heading.

I’ve written about the SEO era’s effect on content generally, so I won’t repeat that argument. This is the version specific to reporting, and there’s a part of it I’ve never written down.

The question I never resolved

Was I investigating technology companies, or promoting them?

As a tech journalist I was supposedly there to report on the industry. In theory that means asking awkward questions, investigating, finding things companies would rather you didn’t write, challenging the story you’re handed.

But almost all the machinery pulled the other way.

You needed access to the companies you covered. You needed executives to agree to interviews. You needed PR teams to answer your calls. The publication needed advertisers, events, traffic and a constant supply of stories.

So it became very easy to spend your time rewriting announcements, interviewing executives about whatever they wanted to announce, and publishing essentially the version of the story that had already been prepared for you.

At some point I started wondering whether I was reporting on the technology industry or working as an unpaid part of its communications function.

That never sat comfortably. It still doesn’t, and I’d note that I went on to work in exactly the industry that was doing the preparing — which makes me a poor candidate for outrage about it.

What distribution was hiding

Here’s the uncomfortable question underneath all of that.

If you’re not producing genuinely new information, not investigating anything and not challenging the organisations you cover, what is the defensible value of the article?

For a long time, distribution hid the problem. Google could send traffic to the rewrite. Social platforms could send traffic to the announcement. A publication could publish enough of them for the economics to work.

The value being captured wasn’t the reporting. It was the position between the information and the audience.

AI removes that position. And when it goes, the question that distribution was covering up becomes very difficult to avoid.

AI can’t retrieve what doesn’t exist yet

A machine is extremely good at turning a press release into an article. Extract the key points, add background, produce the conventional 500 words. Give it information that already exists and it will summarise, restructure and synthesise a hundred versions of the same story.

That is precisely the work journalists shouldn’t have been spending their time on.

What a model can’t do is retrieve information nobody has created.

An interview nobody else has done. A document somebody obtained. A dataset somebody interrogated. A contradiction somebody noticed. A source who knows something. An investigation that connects several apparently unrelated pieces of evidence.

Which is where AI becomes genuinely interesting as a reporting tool rather than a threat to reporting. A journalist can use it to work through thousands of pages of documents, compare company filings, analyse datasets, transcribe interviews, build timelines and look for inconsistencies. Work that used to consume weeks can take an afternoon.

The journalist doesn’t disappear. The journalist gets to spend more time being a journalist — curiosity, sources, verification, judgement, noticing when something doesn’t add up.

Maybe AI doesn’t save journalism. Perhaps it forces journalism to become more journalistic. If the machine can handle the information companies actively want published, the human value moves towards finding the things they don’t.

The part that makes me much less optimistic

I’ve just described a comforting story, so here is the objection to it, which I think is the strongest one available and which the optimistic version tends to skip.

The aggregation was paying for the journalism.

The rewrite, the listicle, the SEO explainer, the fourth version of somebody else’s scoop — that volume business generated the traffic that sold the advertising that funded the newsroom. Investigative reporting has almost never paid for itself. It was cross-subsidised by exactly the cheap repetitive work AI is now absorbing.

So “AI destroys the aggregation business and frees journalists to do real reporting” gets the mechanism right and the economics backwards. Removing the volume business doesn’t liberate the reporting. It removes what was quietly funding it.

There’s a version of the next decade where newsrooms shrink further, publishers fail, AI companies consume the output without meaningfully paying for it, and the amount of original reporting in the world falls sharply — not because anybody stopped valuing it, but because the thing subsidising it went away.

I don’t know how that resolves. Attribution, copyright and payment for original work are all unresolved, and none of them are technical problems that get solved by better models. The honest position is that AI improves what a reporter can do and simultaneously attacks what pays for reporters, and I can’t tell you which effect is larger.

Journalism got there first

The reason I keep returning to this is that journalism is the clearest early example of something now happening in B2B content.

Digital publishing gradually separated producing articles from the harder work of finding things out. SEO rewarded volume, repeatability and coverage of subjects people were already searching for. B2B content adopted the same habits: take an existing topic, write the definitive guide, optimise it, turn one report into five blogs and ten social posts.

None of that is inherently wrong. But eventually the production system becomes the strategy — and production is exactly what AI is now extremely good at.

I’ve made the case for where that value moves elsewhere, so the short version: customer conversations, original research, product data, sales intelligence, subject-matter expertise, experiments, market signals — and the editorial judgement to work out which of those contains an argument worth making.

Which is why I think good B2B content increasingly looks a bit like journalism. Not because every company needs a newsroom. Because companies need people who can find things out before they start producing things.

The B2B version has one significant advantage over the journalism version, and it’s the cross-subsidy problem in reverse: a company funding original research isn’t relying on that research to generate advertising revenue. It has another reason to want to know things.

That doesn’t make it easy. It does mean the economics aren’t obviously against it, which is more than journalism can currently say.

SEO pushed publishing towards producing more. AI might push it back towards finding things out.

For journalism that could mean going somewhere, talking to somebody and uncovering something new. For B2B it means getting much closer to customers, research, products and expertise, then using the machines to make sense of what you find.

I’d like to be confident that’s how it goes. What I’m actually confident about is narrower: the era where you could build a publishing business on repackaging what already existed is ending, and nobody has yet worked out what pays for the alternative.