Twenty-six days old, and ChatGPT out-referred Google

We keep being asked whether the AI channel is real for a brand new product or just something consultants talk about. Here is the smallest honest answer we have: one domain, one month, two numbers.

Run the free check

Kurli.me was registered on 27 July 2026, which is verifiable in public WHOIS records. In the first twenty-five days of that domain's existence, ChatGPT referred 56 visitors and Google Search referred 23. Both figures are small. The finding is the ratio, at the exact point in a domain's life when Google has the least reason to trust it and an AI assistant has no comparable reason to hesitate.

The product and the situation

Kurli is an AI hairstyle advisor. It reads a single selfie, maps the face in the browser to work out its shape and proportions, and renders the cuts that suit that face onto the person's own photo. It is free and requires no sign-up.

The relevant fact for this page is the domain. Kurli.me was registered on 27 July 2026, which anybody can confirm in a WHOIS lookup. That means during the window described here it was a domain with no history, no accumulated links, and no record with any search engine of what it was or whether it was any good.

This is the condition every new product starts in, and it is the condition under which conventional search advice is least useful, because all of that advice assumes a year of patience.

The numbers

In the first twenty-five days of the domain's life, measured from registration:

ChatGPT referred 56 visitors. Google Search referred 23.

That is the whole finding. We are not going to inflate it, because the inflation is what makes this kind of claim useless. Fifty-six visitors is a small number in absolute terms. Nobody builds a business on it, and anybody presenting it as a growth story is selling something.

What makes it worth publishing is the comparison and the timing. At twenty-five days, a domain is at its weakest relative to Google. That is precisely the moment when the AI channel was contributing more than twice as much.

What the numbers do not show

Being explicit about the limits is the only way this is worth anything to you.

This is one domain, in one category, over one short window. It is not a study, it does not establish a rate, and it should not be used to predict what your product will get.

It also does not measure quality of traffic, conversion, or retention. Fifty-six visits is too few to say anything reliable about what those people did.

And the window is short enough that ordinary variation could move either number noticeably. We are reporting what the analytics recorded, not claiming it would replicate exactly.

What survives all of those caveats is the direction, and the fact that the mechanism explains it.

Why it happened

The explanation is not mysterious, which is part of why we trust the result more than we would trust an unexplained one.

Google, faced with a domain registered three weeks ago, has no evidence to work with. It has no history of user satisfaction, no established pattern of citation from elsewhere, and no way to distinguish this site from the very large number of low-quality domains registered every day. The reasonable policy is caution, and caution shows up as very few impressions and fewer clicks.

An assistant answering a specific question is doing something different. It is finding passages that address the question and composing an answer from them. A page that clearly and specifically answers a question about which haircut suits a face shape is a reasonable source for that question regardless of when the domain was registered.

So the asymmetry is a consequence of how the two systems decide, rather than a quirk of this one product.

What Kurli did that helped

Nothing exotic, and all of it is available to any builder.

  • A real custom domain from the start, rather than a generated platform subdomain.
  • Content that answers specific questions people actually ask about face shapes and haircuts, written as answers rather than as product marketing.
  • Pages that a crawler can read, with the substance in the HTML rather than assembled later by script.
  • No blanket robots.txt block on AI crawlers, which sounds trivial and is the step that most often silently prevents all of this.
  • A product that can be tried immediately, with no account, so an assistant recommending it is recommending something the reader can actually use.

What we are still measuring

This page reports a twenty-five day window ending in late August 2026. The domain has continued to exist since then, and the sensible thing is to keep watching rather than to freeze one snapshot and treat it as permanent.

The questions worth answering next are whether the ratio holds as the domain ages and Google begins to trust it, whether the AI referrals convert differently from the search referrals, and whether other assistants beyond ChatGPT begin contributing.

We will update this page with the longer window rather than quietly leaving a month-old figure standing as though it were current. A number with a date attached is worth something. A number that has silently gone stale is worse than none.

Part of a larger guide

This page is one part of The AI channel. The other parts:

Questions people ask

How old was the domain when these numbers were recorded?
Kurli.me was registered on 27 July 2026, which is verifiable in public WHOIS records, and the figures cover the first twenty-five days of the domain's existence from that date.
How many visitors did ChatGPT actually send?
56 in that twenty-five day window, against 23 from Google Search. Both are small numbers in absolute terms, and the finding is the ratio at the point in a domain's life when Google has the least reason to trust it.
Does this prove AI search is better than Google?
No. It is one domain, in one category, over one short window, which establishes no rate and predicts nothing about another product. Google remains the larger channel over any long horizon. What the case shows is that the early months look different from what conventional advice assumes.
Why did ChatGPT send more than Google to a new site?
Because the two systems decide differently. Google has no evidence about a three-week-old domain and cautiously waits, which shows up as very few impressions. An assistant composing an answer selects passages that address the question, and a clear specific answer is a reasonable source regardless of when the domain was registered.
What did Kurli do to earn the citations?
Nothing exotic: a real custom domain from the start, content written as direct answers to specific questions rather than as product marketing, substance present in the HTML rather than assembled by script afterwards, no blanket robots.txt block on AI crawlers, and a product usable immediately without an account.
Are these numbers still current?
They cover a twenty-five day window ending in late August 2026 and are reported with that date attached deliberately. The page will be updated with a longer measurement window rather than leaving a stale figure standing as though it described the present.

Find out what the assistants say about you

The free Get Found Check reports whether AI assistants can currently reach, read and cite your domain, which is the precondition for any of this.

Run the free check Have Licheo do it for you