Scott Purcell, co-founder, Man of Many
What Time and Axios just taught us about selling visibility in AI search, why most of the GEO industry is yesterday’s snake oil in a new bottle, and what brands should actually do about it.
Last week Press Gazette reported that Time and Axios have started selling their prominence inside AI answers as an advertising product. Axios is sitting down with partners to explain “why we’re currently popping in the LLMs” and charging for it. Time’s COO went further, describing a machine-readable version of the site built to feed the bots, then saying the quiet part out loud: Time is building “a pure data product that doesn’t even get published to the web, it’s just marketing to the bots”.
Marketing to the bots. Sit with that phrase, because it is the entire strategy in four words, and it is the wrong one. The telling detail is the demand behind it. Time’s COO says that the moment they raised it, brands replied “that’s our problem, help us”. Every CMO can feel the same gap opening. Customers are asking AI assistants what to buy, and most brands have no idea whether they are in the answer or invisible. That anxiety is real, and it is about to move a lot of budget. The only question is whether it gets spent well or set on fire.
Start with what nobody selling this wants to admit: the technology is a switch. Serving a lean markdown version of your articles to AI crawlers is a toggle inside Cloudflare’s AI Crawl Control. We turned it on at Man of Many in June. It uses content negotiation, so a verified AI crawler that asks for markdown gets markdown, while a human and Googlebot get the identical HTML article. It took an afternoon. So can anyone with a Cloudflare account. The pipe is not a competitive advantage. It is plumbing, and paying a premium for it is a category error.
The genuinely dangerous move is the next one, the one Time is bragging about: content that only a machine ever sees. Serving different content to a crawler than you serve to a human has a name that predates ChatGPT by two decades. It is called cloaking, and Google has penalised sites for it since the early 2000s. There is a clean line most coverage walks straight past. Content negotiation that returns the same article in two formats is fine. A bot-only marketing payload no reader ever sees is cloaking with a media kit attached, and it collides with both Google’s spam policies and its helpful content guidance. Google’s John Mueller put it plainly: “Why would they want to see a page that no user sees?” When we put the cloaking question to our own Google contact directly, the answer was a pointed silence and a link back to Google’s “focus on people” guidance. Read the silence. And a warning for anyone on WordPress: popular SEO plugins now auto-generate these markdown and llms-full.txt files by default, switched on out of the box, so plenty of brands are quietly publishing a parallel machine-only version of their site they have never once looked at. That is how you sleepwalk into cloaking without deciding to.
Which brings me to the part of this industry I keep losing patience with. There is now a thriving trade in Generative Engine Optimisation, a discipline sold as the urgent new thing your brand cannot survive without. It sounds terribly sophisticated, particularly when it is being pitched to you from a yacht at Cannes with an Aperol Spritz sweating in hand and a five-figure monthly retainer on the table. Strip the acronym back, though, and the receipts are brutal. Ahrefs tracked 1,885 pages that added schema markup and found AI Overview citations actually fell 4.6 per cent. SE Ranking tested llms.txt across roughly 300,000 domains and found no relationship with citations whatsoever, which is unsurprising given Mueller has confirmed Google does not read the file. Lily Ray monitored 220-plus sites scaling AI content and found 54 per cent lost 30 per cent or more of their peak traffic, with 22 per cent down 75 per cent or worse. As she put it, “the packaging is new, but the pattern is not”. And SparkToro ran 2,961 prompts with 600 volunteers and found less than a one in a hundred chance that two runs of the same prompt return the same brand list, prompting Rand Fishkin’s verdict that “any tool that gives a ranking position in AI is full of baloney”. The tactics being sold hardest are, in many cases, the same ones that got sites flattened by the 2024 Helpful Content Update. The dashboard is new. The grift is not.
Here is the clarifying part, and it comes from Google itself. In its guidance on the AI search era, Google VP Brendon Kraham writes that “generative AI features like AI Mode are built directly on top of our core ranking systems”. Your edge, he says, is “authenticity and expertise”, and the line that should end the debate is his instruction to stop trying to “optimise your content for bots” and instead “focus on people”. Good SEO is good GEO. The qualities that earn a citation in an AI answer are the same ones that earned a ranking in blue links: real expertise, first-hand experience, a named human who knows the subject, a point of view a model cannot get anywhere else. There is no separate game to hack. There is only the original game, played well. And it matters more than the trackers admit, because the surface everyone ignores is the biggest one: Peec AI’s analysis of 500,000 prompts found Google’s AI Overviews now appear in roughly 86 per cent of searches, rising to 88.5 per cent on commercial, bottom-of-funnel queries, reaching some 2.5 billion users a month and dwarfing ChatGPT and Perplexity combined.
So what does a brand actually do, when the customer is asking a machine what to buy and the shortcuts are a trap? Look at where the citations come from. AirOps analysed 21,311 brand mentions across GPT-5, Claude and Perplexity and found 85 per cent came from third parties, not the brand’s own website, with listicles and editorial round-ups driving roughly 90 per cent of those. Read that twice if you sell something. The route to AI visibility runs almost entirely through other people’s credibility, not your own domain. The branded mention in trusted editorial is the new backlink. You cannot fabricate it in a markdown file at the edge. You earn it by being genuinely worth recommending, and then being written about by a publication a model already trusts.
That, finally, is why the panic is a gift to anyone doing real work. At Man of Many we test what we recommend, our editors put their names on the work, and we publish points of view a content farm cannot. When Parallel Web Systems launched Index, a platform that measures how AI agents actually use publisher content, we ranked first across our competitive set on impressions, citations, value and, most tellingly, uniqueness, the score for how hard our work is to substitute. AI agents are not rewarding the sites that tried hardest to optimise for AI. They are rewarding the ones with something only they could have said.
And here is the good news, because this is no longer a theory we are hoping plays out. It is already in our numbers. More than 1.3 million AI-crawler requests now hit Man of Many every week, the bots steadily reading, surfacing and learning from what we publish. That machine attention is not vanishing into a void. It is converting into real people: our referral traffic from AI assistants is up 38 per cent year on year, with ChatGPT alone now sending three in every four of those visitors. The model reads us, the model cites us, and the reader follows the citation back to the source. That is the entire flywheel, working and measured, in the only currency that counts.
So here is the lesson for any brand staring down the AI visibility question. You cannot fake your way into that flywheel with a markdown trick on your own domain, and 85 per cent of the brand mentions AI serves up are coming from third-party editorial anyway. The route runs through publishers a model already trusts. Working with reputable publishers does not just sound nice. It pays, and now we can prove it pays.
By all means turn the switch on. We did. Serve the bots a clean copy of your journalism, because it is cheap, honest and makes you easy to cite. But understand what is being sold when someone offers to help you market to the machines. It is a shortcut around the one asset that matters, and Google has told you in writing that the shortcut does not work. The brands and publishers who win the AI search era are the ones who never stopped doing the unglamorous thing. The authority that AI rewards is already out there, earned, named and on the record, and for us it is already paying its way back in readers. The only question worth asking is whether you are part of it.