I pulled three days of attribution data and nearly published the wrong headline. The 30-day view told the truth: a chatbot is my number three acquisition channel, and the pages it cites are not the ones I polished.
I run AetherWave Studio solo - an AI music and creative platform, every commit shipped through AI coding agents, every marketing channel run from the same chair. Last week revenue jumped and I assumed it was me. I had been loud on social media for three days straight, so obviously my posting caused it.
Then I pulled the attribution data for those three days, and ChatGPT was my top channel at 36% of signups. I wrote a whole draft of this post around that number. Before publishing, I widened the window to 30 days:
- Google organic: 354 signups (37%)
- ChatGPT: 151 (16%)
- True direct: 151 (16%)
- Everything else, including all my social posting: the long tail
The three-day number was real, but it was a spike, and worse, it moved depending on where I cut the window. Shift the boundary by half a day and Google was back on top. The 30-day number is the honest one: ChatGPT sends me 1 in 6 signups. That is still wild. A language model is recommending my site to strangers, around the clock, for free, and it outranks every channel I actively work except search itself. But it is not "my biggest channel," and I came one publish button away from claiming it was. Lesson zero: widen your window before you brag.
## How I could even see this
ChatGPT appends utm_source=chatgpt.com to links it cites. That is a gift, but there is a trap: its in-app browser strips the referrer header while keeping the UTM in the URL. If your analytics buckets by referrer, ChatGPT traffic shows up as "direct" and you will credit your own charisma for a machine's work, like I did. The fix is one line of logic: channel = UTM source first, referrer host second. Never the other way around.
And these are not idle visits. Some ChatGPT arrivals pulled out a card within minutes: one bought a credit pack four minutes after landing, another inside the first half hour.
## The part that changed how I write content
I have a free album art generator - a proper tool page, my best lead magnet, ranked in Google, polished. I also have a plain knowledge-base article that just explains how one of the music features works. No design. No hero section. Just an explainer.
In 30 days, ChatGPT sent 17 signups to the boring explainer and 4 to the shiny tool page.
LLMs do not recommend pages that sell. They recommend pages that explain. When someone asks ChatGPT how to do something, it wants to cite a page that answers the question, and an article structured as an answer beats a landing page structured as a pitch, every time. My knowledge base, which I built for customer support, turned out to be my storefront for AI referrals.
## The kicker
ChatGPT's favorite page on my entire site turned out to be a music editing page. 29 signups in 30 days landed there, and every single one came from ChatGPT. But that page is for working on audio you already have: WAV conversion, stem separation, sound effects. The people ChatGPT was sending wanted to create music - which happens to be my best-converting intent - and the page gave them no path to do it.
Here is the move, and it is the opposite of what instinct says. I did not try to correct ChatGPT. You cannot file a ticket with a language model about where it sends people, and why would I want to? The misinterpretation is delivering customers. So I adapted the page to the traffic instead: one Generate New Music card at the top, sitting above the editing tools, linking straight to the full creation workflow where album art gets made too. An AI sending you the "wrong" traffic is only wrong until you change what the page offers. Go with the flow.
## The search side tells the same story
Over the last three months, 77% of my Google clicks came from static prompt-library and explainer pages, not from product pages. One page about fight scene video prompts has more clicks than my homepage got from anything except my brand name. These pages cost nothing to serve, they compound monthly, and they are the exact pages LLMs cite. Search and AI referrals are not two channels. They are one channel with two front doors, and the same content feeds both.
## What I would tell any creator or founder doing this
1. Write explainers for every intent your product serves. One honest article per question. Support docs count double - they answer questions in exactly the shape LLMs want to cite.
2. Check your attribution logic today. UTM first, referrer second. If ChatGPT is sending you users, there is a decent chance your analytics is filing them under "direct" right now.
3. Measure landings, not just clicks. Which page did the user enter on? That first-touch page tells you which content actually recruits users. Mine surprised me twice.
4. When an AI sends traffic to the "wrong" page, do not fight it. Put the right next step on that page and let the traffic keep coming.
5. Sanity-check your best number before you publish it. A three-day window made me a better story than a thirty-day window made me. Ship the thirty.
6. Do not stop posting on social. My posts did move something real - dormant users from months ago came back and converted. Social wakes up the people who already know you. Explainers recruit the people who do not.
I spent three days believing my posting drove a revenue bump, and the data said a chatbot and a search engine were quietly doing the heavy lifting through pages I wrote months ago. That is the strangest part of building in 2026: your best salesperson might be someone else's AI, and the only way it can pitch you is if you wrote something worth citing.
More from the solo build as it happens. Every number above is real, pulled from production this week.