How a cycling website turned its newsroom tools into the product

Hi, I’m media innovation journalist Ulrike Langer and you’re reading my weekly AI in media newsletter. Peter Stuart of Velora Cycling and I both spoke at the European Publishing Congress in Vienna on June 18. I missed his session because it ran parallel with one of mine but we chatted for a while at the event and it immediately became clear to me that Velora Cycling just had to be a case study for News Machines. Peter and then I had an hour-long conversation via Meet on June 29. This piece is based on our talk.

Peter Stuart launched Velora Cycling in late 2025 as a one-person specialist publication built on an AI-heavy workflow: tools that draft articles, surface news leads, generate social posts, select images, transcribe interviews, and trace every quote back to an original source. Stuart, formerly editor of Cycling News, edits everything himself. His co-founder Danny Bellion, former head of AI at the fintech provider Capital on Tap, built the platform.

That workflow enables a one-person publication to operate at the speed of a small team. Seven months in, Stuart says that Velora has 10 publisher clients paying to license it. The cycling website is the showcase. The platform is the actual business.

"The site's probably worked better for us as a showcase for the platform that other people have started to use and pay for on a license basis," Stuart told me in an hour-long Q&A in late June. "The site is not monetized to a point that it would pay my salary and my partner's salary in its isolation."

The publication still does what Stuart says it does — it uses AI to escape commodity coverage and chase distinctive cycling journalism — but it now does that work in service of a software business sitting beneath it.

The original plan was efficiency. “Initially, we just thought, how are we using AI to help our workflows and do things quicker?” Stuart told me. “But the creation of commodity content is something that's quite easy to automate."

The market shifted on them within months. By Stuart's account, the value of commodity coverage collapsed sometime around November 2025 — derivative news lost its monetization floor, especially for new domains without an inherited audience. "Nobody really cares if you're talking about the same thing that was reported by two other people. Even if you were the first to report it, it doesn't matter because you haven't got access to that audience before they do."

So Velora pointed the AI capacity in a different direction. Tools that flag when a draft isn't distinct enough and prompt for added perspective or lived experience. Transcription and drafting pipelines built around interviews rather than rewrites. A source-mapping research function that tries to find the original source of every claim rather than the nearest available secondary one. Asked whether he has cracked the code of being efficient and distinctive at the same time, Stuart declined to go that far. "I think that's a very hard one to crack."

Velora is currently approaching 50,000 sessions (visits) a month and targeting 100,000 — the threshold Stuart says historically unlocks first-party display monetization. One Oxford-based freelance reporter is on contract, filing 15 to 20 pieces a week. 

Stuart deliberately chose quality over quantity. He could push to 20 or 30 stories a day with AI but he doesn't, because one editor can't personally fact-check that many stories. "Even with automations, you need somebody working at 8 p.m. UK time. You need someone working at the weekend. That's one thing that's very hard for one individual to do."

The cycling site is not economically viable on its own, by Stuart's own account, and it doesn't have to be. It demonstrates what the platform can do for the publishers paying to license it.

Stuart and Bellion charge around £300 per site per month (about $400) for a workflow supporting roughly 150 articles, with multi-site and enterprise tiers below that on a per-site basis. Ten publishers currently use it, some with multiple sites — mostly B2B and consumer-direct titles, some financial. Stuart declined to disclose the names of his clients for now. "I think by the time we get to 30 clients or 30 sites we'll probably be sustainable." Bootstrapped, with some external investment.

Stuart is candid about the cost of being public about AI use. A vocal minority of maybe 20 or 30 cycling readers, he says, treats his disclosure as reason to write off Velora Cycling as AI-generated. He suspects competitors use AI more heavily than he does but say less about it. Stuart also believes Google penalized on-page AI disclosures at one point as thin-content signals, which further discouraged transparency. His approach: Treat the hostile audience as sunk cost and compete on content quality.

The case for Velora's editorial defensibility is that one editor sits close enough to every output to feel every rough edge — a tight feedback loop AI can't replicate. That's true today. It also has a planned expiration date. Stuart says he'd like to hire "one or two editor-journalists" in the long term. And the platform is built to scale exactly the kind of one-editor-feels-everything workflow Velora demonstrates — for buyers who won't necessarily run it the way Stuart does.

Velora, the publication, sits in the distinctiveness lane. Velora, the platform, is category-agnostic. The clients get to decide whether they point the freed AI capacity at distinctive work, at scale, or at the same commodity volume Stuart says isn't worth producing anymore

Velora's hallucination-detection workflow shows what the platform actually does. Its research agent traces every quote and claim back toward an original source (source-mapping) and flags anything it can't anchor.

It catches hallucinations that are already cascading through journalism. Stuart described one case where a quote about an athlete's injury had propagated across cycling media; Velora's source map traced it back to a Danish-language article that contained no such quote. Someone's AI tool had likely inferred it from context. The fabricated quote then got picked up by outlets that didn't verify it. Across all the detection problems Velora catches — parody, parasite SEO (low-quality content published on high-authority domains that Google ranks based on the parent site's reputation), hallucinations — fewer than five cases ever reached the drafting stage, and none reached publication. But the pattern is the point. Hallucinated quotes are entering journalism through loose AI workflows and propagating outward through outlets that don't check. Without source-mapping or equivalent verification, every publisher using AI tools becomes a potential amplifier of misleading content.

This is the part of the platform Stuart can sell with the fewest caveats. Distinctiveness is a claim. Source verification is a process — and one that's increasingly necessary regardless of how the editorial work gets done.

Asked whether his stack is a rat race against the frontier models — whether the next version of ChatGPT or Claude will flatten his advantage — Stuart said that Velora provides coordination of AI models, tools and agents, not any single one.

"Someone will always have the better research agent or the better image selection agent or the better writing tool.” Velora’s USP is picking the right model for each job and testing the picks against each other as new models ship. That advantage holds until a single model gets good enough at everything.

Stuart doesn't pretend that moment is far off. If a single model gets good enough at every job, he says, the question isn't what happens to Velora — it's what happens to news publishing at all when readers can just ask an AI agent.

Stuart has already tested where the line sits. Early on, Velora ran an AI agent against him — machine editor versus human editor, same site. "It was abysmal," Stuart said. "It just couldn't do it at all. The stories it decided to plan, the stories that it decided to publish just didn't make sense." The agent surfaced leads but it couldn't make the calls. That gap is Velora's business. Publishers buy the coordination layer. They keep the editorial decisions themselves. For now.