On August 6, 2026, we launched svetstylist.com — the website of a personal stylist in Lisbon. A new domain, zero history, not a single inbound link.
On August 23, the first visitor arrived from ChatGPT. Seventeen days.
By mid-September, ChatGPT and Perplexity were citing the site, and Google was showing it in AI-generated answers. Two of the visitors who came from ChatGPT became paying clients.
- 17 days to the first visit from ChatGPT
- 5 visits from ChatGPT in the first month
- 2 of them became clients
- 7 impressions in Google’s AI answers
Below: what we did, how it differs from regular SEO, and how to check it on your own site. And, up front, a caveat about scale — without it, this piece wouldn’t be honest.
First, About Scale
In the month since launch, the site had 161 visitors. From Google Search: 43 clicks on 1,050 impressions, average position 11.1. From ChatGPT: five visits. In Search Console’s generative AI features report: seven impressions.
This isn’t a story about ten thousand visits. It’s a story about speed and quality.
If I showed you a screenshot with tens of thousands of sessions, the honest question would be: how many of them bought anything? Here I can answer precisely: two out of five. And that’s exactly why this case is worth taking apart — a visitor who arrives from an assistant’s answer arrives with the decision already half made.
Ranking and Being Cited Are Not the Same Thing
Here’s the main thing we learned on this project.
A search engine ranks pages. An assistant cites sentences.
A real example. The first client who came from ChatGPT messaged on WhatsApp: how much does it cost, and when are you available? Even though the prices were on the site. Right on the service page, in large type, on the pricing cards.
The assistant couldn’t quote them. The price lived in a badge, the duration in a line next to an icon, the format in a column header. A human reads that grid in a second. A machine sees a set of numbers with no statement it can lift whole and drop into an answer.
We rewrote it so that every fact also existed as a sentence:
“A wardrobe edit costs €226 and takes three to four hours — at your home in Lisbon or over video call. You’re left with a lookbook of ready-made outfits from the clothes you already own and a prioritized shopping list.”
That’s quotable. A grid of icons isn’t. Both versions live on the page at the same time: cards for people, sentences for machines.
What We Did
Not a step-by-step recipe, but the substance. Five things.
1. The Technical Basics Everyone Finds Boring
Fast page delivery, a correct sitemap, hreflang between language versions, instant notification of search engines about new pages. Nothing inventive. But until this is done, nothing else matters: an assistant won’t cite a page it hasn’t seen.
2. Markup That Describes the Business, Not the Page
Most sites mark up articles and breadcrumbs. We described the service itself: what it is, what it costs, where it’s provided, in which languages, and by whom. An assistant doesn’t need a document — it needs a subject it can make a statement about.
3. Questions Worded the Way People Actually Ask Them
Not “Services and Prices” but “How much does a stylist in Lisbon cost?” Not “Ways to Work Together” but “I’m not in Lisbon — does this work?” We took clients’ real first messages and turned them into headings. One detail from there: that same first client called the service by a word that didn’t appear on the site even once. Now it does.
4. An Unambiguous Identity
Before recommending anyone, an assistant has to be sure it understands who it’s talking about. Who this is, where they work, in which languages, what exactly they do — and all of it has to match everywhere there’s data about the person. Mismatches between the site and external profiles are expensive: they’re a reason not to name you in an answer.
5. A Plain-Text Version for Machines
A separate file where everything important about the business is laid out in plain text with links. Cheap to make, easy to read, and it doesn’t go stale, because it’s generated from the same database as the site.
And separately, what we did not do: we didn’t buy links, didn’t write articles “for keywords,” and didn’t wait for the domain to “age.” That last one matters most — in classic SEO a new domain sits in the sandbox for six months. Assistants don’t have that mechanism.
How to Check This on Your Own Site
Three places worth looking at today.
- Search Console, Performance → the generative AI features report. It’s in beta, and most site owners don’t know it exists. It shows how many times your site appeared in Google’s AI-generated answers.
- A referrer filter in your analytics.
chatgpt.com,perplexity.ai,copilot.microsoft.com. If it’s zero, you’re not being cited. If it’s a handful, things are moving. - Ask the assistant itself. Give ChatGPT the question your client would ask and see who it names. It’s a rough check, and the answer changes from query to query, but it’s more honest than any forecast.
The Trap You’ll Fall Into When Measuring
And now for something almost nobody writes about.
Here’s what ChatGPT traffic looks like in this project’s analytics: five visits, 100% bounce rate, average visit duration zero seconds.
By any dashboard, that’s the worst traffic on the site. Junk you should get rid of.
Two of those five became paying clients.
The explanation is simple: the person has already read the answer in the chat. They come to the site not to explore but to verify — to make sure the business is real, to see a face and a price. Then they close the tab and send a message. Analytics sees one page, zero seconds and a bounce. The business sees a client.
If you judge AI traffic by the usual engagement metrics, you’ll switch it off as useless. Measure it by inquiries, not by pages per session.
What This Means for Your Business
The window is open right now, and it works differently from the search one.
In regular search, a small business has to spend years building domain authority to compete with aggregators and directories. An assistant needs domain authority far less. What it needs is to be able to make a verifiable statement about you — and to find that statement as a ready-made sentence.
Which means a small, honest site with clear facts can beat a big site with beautiful landing pages today. Not for long. In a year, everyone will be doing this.
Want the Same?
I do projects like this end to end: the technical foundation, structure, markup, copy that can be quoted, and measurement — so that a month later you see not “traffic went up” but exactly where your inquiries came from.
The case above isn’t my biggest, but it’s the most telling: it shows that getting into assistants’ answers takes neither a budget nor an aged domain. It takes work that almost nobody is doing yet.
Write to me — we’ll look at your case, and I’ll tell you honestly whether there’s something to gain.
Data in this article: Google Search Console and the project’s analytics, August 5 – September 10, 2026. Case published with the client’s consent.
