Last November, a developer asked Tailwind to publish its documentation in a format built for AI assistants to read. Tailwind is one of the most-used tools for styling websites. It was a small, reasonable request.
In January, the founder, Adam Wathan, replied. The idea was good. It would have to wait. The day before, three-quarters of Tailwind’s engineering team had lost their jobs. Traffic to the docs was down about 40% since early 2023, “despite Tailwind being more popular than ever.” Revenue was down close to 80%.
Tailwind did not lose to a competitor. Nobody stopped using it. The AI kept the user and took the visit — and the visit was the business.
Most companies with a website have the same shape. Theirs is just smaller, and they haven’t checked yet.
More users
Tailwind’s model was one that thousands of businesses run without naming it. The framework was free. The documentation was free, and excellent. Somewhere in those docs, a developer would notice that Tailwind Labs also sold ready-made interface components and templates, and some of them bought. Wathan put it plainly in the same reply: “The docs are the only way people find out about our commercial products.”
Then the assistants arrived. A developer asks for a pricing page, and the assistant writes the Tailwind for them. The framework gets used more than ever. The docs never get opened. The paid product never gets seen.
Wathan’s own summary is unusually honest for a founder: “there’s just no correlation between making Tailwind easier to use and making development of the framework more sustainable.” Making Tailwind more useful had stopped paying for Tailwind.
On 9 September, Tailwind Labs announced it was joining Shopify. The framework stays free and maintained. The company closed sign-ups for new customers — including ui.sh, a newer product built for people who design with AI agents. “On the commercial side,” Wathan wrote, “we’ll no longer be trying to grow the business around Tailwind.” The same post notes the framework is installed more than 110 million times a week, and that it styles ChatGPT.
Fewer visitors
This is not a developer problem. Developers just got there first.
Pew Research tracked 900 American adults through a month of Google searches in 2025. When an AI summary sat at the top of the results, people clicked through to a website on 8% of visits. Without one, 15%. They clicked a link inside the summary itself on 1%. And they were more likely to simply stop — ending the session on 26% of pages with a summary, against 16% without.
Now picture the pages behind those numbers. An accountant’s explainer on the new filing deadline. A clinic’s page on what to expect before a procedure. A law firm’s plain-English guide to tenancy disputes. Each was written for one reason: the person who reads it might call. The explanation was free because the phone call wasn’t.
The answer engine reads the explanation for them. The person gets the answer and never sees the phone number.
The better your page answers the question, the less reason anyone has to visit it.
Here’s the cruel part. Quality used to be the defence. A thin page lost to a thorough one, so businesses invested in thorough pages. Now the thorough page is the one an assistant can answer from most completely. The work you did to earn the visit is exactly the work that lets a machine skip it.
The visit, for sale
If the visit no longer comes free, someone will sell it to you. September settled who.
On 16 September, OpenAI began testing Sponsored Agents with a small group of US advertisers. A business builds its own agent; after seeing an ad, a ChatGPT user can start “a clearly labeled conversation” with it, ask questions, and follow a link to the business’s site. On 24 September, ChatGPT ads reached seven more Asian markets, including Singapore and Taiwan — more than 60 countries in all. The ad business passed a US$1 billion annualised run rate in under 200 days, with tens of thousands of advertisers.
OpenAI says ads don’t influence ChatGPT’s answers. Take it at its word. That means the unpaid answer stays exactly as it is — built from what the web published for free — and the only reliable way back to the customer is the paid slot beside it.
For twenty years, being useful online was paid in visitors. That payment is now a product, and you are the buyer.
What to count
None of this means the website is dead. A small site can cost nothing to run and still earn its keep. It means the job the site was doing has changed, and most owners are still measuring the old one.
Count the path, not the traffic. Take your last twenty customers and find out how each one first reached you. If most came through a page that answered a question, that is your exposure — whatever the traffic chart says.
Watch revenue, not visits. Tailwind’s docs traffic fell about 40%. Its revenue fell about twice as far. A traffic dip looks survivable right up until the invoice numbers arrive.
Sell what an answer can’t deliver. Information travels without you now. The doing, the judgement, the accountability, the signature at the bottom of the page — those still need a person, and they are what the free content should point towards.
Decide what the free part is for. If the explainer’s job was to attract strangers, it is losing that job. Its new job may be to reassure someone who already found you by another route — a referral, a repeat customer, a name they heard. That is a smaller audience. It is also one that no answer engine can intercept.
Wathan built something installed more than a hundred million times a week, and still couldn’t make the arithmetic work. The usage was real. The visits were the business.
Being useful is no longer the same as being found.