August 5, 2026 by Vincent Schmalbach
TIME is now serving two different versions of its website. Humans get the magazine. AI crawlers get a stripped down markdown copy with ads baked in that no person will ever see.
I'm writing this while working on vroni.com：https://www.vroni.com/?utm_source=vincentschmalbach.com&utm_medium=blog&utm_campaign=blog_post_vroni&utm_content=inline_note, my AI coding tool for turning tasks into pull requests.
I fetched one ordinary TIME health article, The Morning Light Habit Sleep Experts Swear By：https://time.com/article/2026/08/04/morning-light-better-sleep/, over and over from the same machine, changing only the User-Agent header each time. That header is the string every browser and bot sends to say what it is. TIME reads it and decides what to hand back.
As Chrome, I got 200 OK , text/html , and 303,235 bytes. The full page, design, images, scripts. As Safari, the same 303KB. As Googlebot, also the same 303KB of HTML.
Then I asked as an assistant crawler. As ClaudeBot, I got 200 OK , text/markdown , and 13,409 bytes. As PerplexityBot, byte for byte identical. As OpenAI's OAI-SearchBot, identical again. Same URL, same second, one twenty-third of the size, and a completely different format: no HTML, no layout, just clean markdown that a language model can process.
A couple of the bots did not even get that. GPTBot and ChatGPT-User, the agents OpenAI uses for training and live fetches, came back 406 . Blocked. But OAI-SearchBot, the one that feeds ChatGPT's search index, was waved through to the markdown. So this is not a blanket bot policy. TIME is choosing, per bot, who gets the reader-free version.
Response headers on the markdown copy:
Mobian is an ad-tech vendor. The markdown page literally begins with . That x-mobian-impression value is a fresh UUID on every single request. I fetched the same page twice and got two different IDs. Paired with cache-control: no-store , that means every time a bot reads the page, it is logged as a distinct ad impression. And x-mobian-tokens: 3323 tells you the unit being counted. Not a person, not a pageview. Tokens fed into a model.
The article itself carries no ad. The sponsored unit shows up on the list and section pages, one per page. So I fetched TIME's Best Inventions of 2025：https://time.com/collections/best-inventions-2025/ collection as ClaudeBot. Sitting inside the markdown, where no human reader would ever encounter it, is a full Ally Bank FAQ:
It runs on with questions like "Which banks offer early direct deposit?" and "Can you deposit cash at Ally Bank?", each answered in Ally's own marketing language, plus a FAQPage JSON-LD block and tracking links tagged campaign="ally-2026-q3" . The business section had the same treatment for the Project Management Institute: a "Reference Facts and FAQ" table, member counts, a claim that certified project managers earn 16% more, all labeled sponsored. The human HTML of those pages contains none of it. Zero hits for "Ally Bank" or "Mobian" when I loaded them as a person.
"Who is Ally Bank?" is written like the phrasing a model emits when a user asks ChatGPT which bank to open an account with.
The ads are labeled sponsored inside the markdown, so this is not undisclosed advertising in the classic sense. What is hidden is the audience split. There is now a layer of TIME.com written entirely for machines, and the humans who read the site have no idea it exists or what is being said to the models on their behalf.
Googlebot gets the same HTML a human gets, so the search-ranking crawler sees the real page. Only the assistant crawlers get the forked version.
TIME says its bot traffic already outnumbers its human traffic on most days. That number is going to be true for a lot of publishers soon. So this is probably the first clear look at what the web starts to become when the main audience is AI models.
Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.
GPT-5.6 Sol xhigh now uses more than twice as many tokens per session as GPT-5.5 xhigh in my Codex workflow. For Codex…
LLMs use Markdown tables all the time. A few months ago, Codex CLI displayed them as ordinary wrapped text. In a narrow…
Google Lighthouse and Cloudflare are adding different controls for the same emerging class of automated web visitors. Lighthouse added an experimental Agentic…
Your email address will not be published. Required fields are marked *
I write about software, AI, Laravel, SEO, and what it takes to build online products. In 2025, 73k people read the blog. Some posts were picked up by publications and developer communities.
A few older posts got the most attention. I keep them here, separate from the newest writing below.
The post that led to a Guardian column and a long developer-community discussion.
Why AI makes weak engineering foundations more expensive, not less.
Good AI work still depends on technical judgment and taste.
Specific Laravel advice for people working in real codebases.
The newest posts from the blog. This updates automatically when I publish.
TIME is now serving two different versions of its website. Humans get the magazine. AI crawlers get a...
GPT-5.6 Sol xhigh now uses more than twice as many tokens per session as GPT-5.5 xhigh in my...
LLMs use Markdown tables all the time. A few months ago, Codex CLI displayed them as ordinary wrapped...
Places that have mentioned, quoted, or discussed my writing and commentary.
Actual feedback from clients I've worked with.
"HIRE VINCENT. Seriously. I've worked with many many developers over the years (remember rentacoder?), and Vincent is that rare breed of web developer who just gets it. You have a chat, you express what you need, and he comes back with what you asked for. His english is fluent, he understands what you need, and he's not looking to overcomplicate things. It's a joy and pleasure to work with him, not to have to micromanage and just overall a great experience. Honestly I wish I could keep him on we just don't have the budget anymore - that is the only reason the contract ended."
"What stood out about working with Vincent was how methodical and efficient he was. He started by really digging into our requirements, asking clarifying questions until he had a solid understanding. He then built exactly what was needed and kept us informed throughout. Communication was clear and proactive at all times. Everything stayed on track, both in terms of time and budget. The end result was excellent and has been running smoothly since launch. It's that combination of practical focus and attention to detail that makes the difference. I would definitely recommend Vincent for quality development."
"Having Vincent on retainer has been excellent. He's tackled some major technical challenges for us, notably significantly improving our website's performance. He also expertly managed a complex upgrade of our core platform (Laravel) to the latest version, ensuring we stay current and secure. We're now rolling out the multilingual capabilities he built, which is exciting. Vincent is reliable, technically skilled, and a valuable partner for keeping our site running optimally and moving forward. Highly recommend his expertise."
"I've been working with Vincent for several months, and he's one of the most independent developers I've collaborated with. He doesn't need micromanagement - he's fully capable of making decisions and driving the project forward, which has freed me up to focus on other parts of the business.
Vincent works very well asynchronously: communication stays focused, there is little unnecessary back-and-forth, and he still raises the important questions when needed. The result is a developer who delivers without overhead, and that's been genuinely valuable."
"We work with Vincent regularly on Laravel/Vue projects, both new builds and existing codebases that have grown over the years. In both cases, he delivers clean code and reliable implementation. We especially value how quickly he gets into mature projects and finds solutions that fit the existing state of the codebase. Absolutely recommended."
"Vincent was easy to work with and produced high-quality code. Would recommend."
"Vincent is one of those rare developers who can jump into a complex codebase, become useful immediately, and solve real problems without hand-holding.
He worked on performance optimization, customer-specific features, and a major custom reporting system. He also turned customer discussions into clear technical requirements and documented his work so the rest of the team could build on it.
Vincent does not just close tickets, he thinks through the product, the edge cases, and the customer impact. If you need a senior Laravel developer who can move a SaaS product forward, I strongly recommend him."
Send the rough version. I can help turn it into a plan and then build it.
Launch Your SaaS in Days, Not Months. A practical, no-BS guide to building production-ready SaaS apps with Laravel 12: speed, billing, teams, AI integration, and the parts you need before customers can pay you.
Vincent Schmalbach is an entrepreneur, software developer, and SEO expert. He works on SaaS products, business apps, automation, and technical project rescue, with 15+ years of software and online business experience.
Scan to open this article