LLM SEO is real: How ChatGPT is driving recurring traffic to my side project (without an llms.txt)
Hey builders,
I just wrapped up a 3-month traffic experiment with my tool, Convertino, and the results were unexpected.
For the past 90 days, I completely stopped all social sharing. No posts, no launches, no manual distribution. I wanted to see how much traffic the tool could pull entirely on autopilot through traditional SEO and LLM discovery.
Here is what I learned:
Google is steady, but ChatGPT is stickier: While organic Google search traffic is growing, a surprising chunk of our user base is coming directly from ChatGPT suggestions.
LLM traffic retains better: The crazy part? The visitors coming from ChatGPT aren't just one-and-done clicks. They are turning into recurring users. Once an LLM recommends a tool to solve a specific workflow, users seem to trust it and keep coming back.
You don't need
llms.txtto start: I didn’t even have anllms.txtfile deployed until a few days ago. While it's definitely a great practice to help bots index your product correctly, ChatGPT was already finding and recommending us without it.
If you're building a micro-SaaS right now, optimizing for AI crawlers might be just as important as traditional keywords.
Have any of you noticed recurring traffic from LLMs or Claude/ChatGPT wrappers? How are you optimizing your landing pages for AI discovery?
The llms.txt part matches what I found. I went looking for evidence that any engine actually fetches it and couldn't find any. No vendor documents reading it, and in my own logs the file gets hit by scrapers and curiosity, not by OAI-SearchBot or PerplexityBot. Worth having, but it's probably not what got you recommended.
One thing that might change your ChatGPT number though. Last week I clicked through to my own site from inside each assistant and read the requests off the server. Desktop ChatGPT, Perplexity and Gemini send a referrer. Claude sends nothing at all. All four iOS apps strip the referrer completely, but ChatGPT and Perplexity make up for it by appending utm_source=chatgpt.com and utm_source=perplexity to the link itself.
So if your analytics groups by referrer only, your mobile ChatGPT traffic is sitting in Direct and your real number is higher than what you reported. Claude and Gemini on iOS send nothing either way. Those ones are just gone.
The retention claim is the more interesting one and I don't have data to argue with it. If you've got the cohorts, ChatGPT vs Google organic at day 30 would be a post I'd read.
The part that caught my attention wasn't that ChatGPT sent traffic—it was that the behavior after the click looked different.
A new acquisition channel only becomes interesting once it starts bringing users who actually return, not just visitors who arrive once.