Skip to main content
Yes, people arrive from ChatGPT. Here's how many.

Yes, people arrive from ChatGPT. Here's how many.

Publié par Elena-Octavia Mateescu, 2026-11-16

English

Eleven. Between 18 August and 9 November, eleven conversions on TOKERO came from people who clicked a link inside ChatGPT. Ten different people. In the same weeks, the site recorded 4,421 registrations.

If you've read a headline lately saying AI assistants are "the new search", that number is probably smaller than you expected. It was smaller than I expected too. I still think it's the most interesting line in our data.

How I found it

I'm a developer at TOKERO, a social network about money. On 9 November I was pulling figures for a list of article topics: one figure per article, each one something only our own data could show. Most of them came from the obvious places. Summit sign-ups, which became my last article. Reels scheduled across networks. Accounts opened per month.

Then I sorted every conversion by where it came from.

Some background first. Since 18 August, every conversion on the site has been saved with the tags that came in on the link (source, campaign, medium). A conversion here means one of three things: someone registers, signs up for a webinar, or buys something. The tags are the usual utm_ parameters you've seen at the end of links a thousand times.

Most rows carried the sources we put there ourselves. Then, further down, one I hadn't: chatgpt.com.

Nobody on our team had tagged a link that way. We never ran a campaign in ChatGPT. The tag was added by ChatGPT itself, which marks the links in its search answers with utm_source=chatgpt.com. So these were people who asked an assistant something, got an answer that mentioned us, clicked, and then did something on the site.

What eleven actually means

My first reaction was to dismiss it. Eleven conversions out of thousands is a rounding error. If you're planning a budget, it isn't a

My second reaction was that I was reading it wrong.

Think about what had to happen for each of those eleven rows to exist. Someone had a question. They didn't type it into a search box; they asked it as a sentence. The assistant answered, and somewhere in that answer it decided our site was worth linking. The person clicked. And then, instead of reading and leaving, they registered, or signed up for a webinar, or bought.

We didn't buy a placement or pitch anyone. These came from answers, not ads. The way in was to be the page an assistant thought answered the question.

And eleven is only what we can see. The tag only survives when someone clicks the link straight out of the chat. If they copy the addrer the name and type it in later, or open the app on a phone and come back on a laptop, the visit arrives with no source at all. In the sameperiod, 1,710 of our 4,421 registrations carried no source. I can't tell you how many of those started in a chat window. Nobody can, from this data. But some of them almost certainly did.

So the honest version is: at least eleven, from ten people, in twelve weeks, without a campaign behind it.

Why I'd rather watch this than ignore it

Plenty of channels start out looking like this. A strange source in a report, too small to matter, easy to scroll past. The ones that grew were the ones somebody noticed early enough to understand.

There's also something different about how these visitors arrive. A visitor from an ad was interrupted. A visitor from a feed was scrolling. A visitor from an assistant asked a question first, and the page they land on either answers it or it doesn't.

That changes what I look for on our pages: whether a page answers one clear question well enough that something summarising the web would point to it. A feature page that says plainly what the tool does and who it's for. A help article that answers the question in its first paragraph instead of its fifth.

I don't know yet which of our pages the assistant is linking to. That's the next thing I want to find out, and it's the first thing I'd suggest you check too.

Check your own site in five minutes

You probably already have this data. Here's where to look:

  1. In your analytics, filter traffic or conversions by source and search for chatgpt.com. Do the same for other assistants you think your readers use.
  2. Look at the landing page, not just the count. Which page did the assistant send people to? That page is doing something right.
  3. Look at what they did next. A visit is one thing; a sign-up from a visitor who asked a question first is another.
  4. Write the number down with its date. At eleven, a single month tells you nothing. The trend over six months will.
  5. Don't build a strategy on it yet. Watch it for a few months first.

If the result is zero, that's useful too. It tells you that right now, when people ask an assistant about your field, your site isn't the answer it gives.

The number I'll be checking in February

Eleven conversions is too few to plan around. It is enough to keep watching. Some people now find a site by asking a question and following the answer, and the ten who found ours acted on what they saw.

I'll pull the same figure again in three months and write down what changed. If you want to see what those visitors found when they la.com/">TOKERO is here.

Have you looked for chatgpt.com in your own analytics? What did you find, even if it was zero?

Octavia Mateescu is a software developer at TOKERO, where she builds the tools creators use to publish and share their work across networks.

Figures: TOKERO production data, ConversionEvents, read on 9 November 2026. Conversions are registrations, webinar sign-ups and purchases recorded since 18 August 2026. Registration totals (4,421; 1,710 without a source) are from the same data, same dates.