Originally published on my own site. Numbers are unedited; the uncomfortable ones are still in there.
Most write-ups about AI-driven marketing show you what worked. This one shows what didn't, because I kept the analytics either way and the negative result is the part that cost me a year.
What I ran
One person, three hours a day, no marketing budget. Over roughly a year I ran an automated posting system on social, generated blog articles with AI on a daily schedule, and built out a programmatic SEO layer of 118 pages. All of it was live. None of it was a prototype.
What the analytics said
| What I ran | Volume | Result |
|---|---|---|
| Automated social posting | ~1 year, 1-3 posts/day | 0 sessions. Not "low" - the referrer list for a 120-day window contains no entry for it at all. |
| AI-generated blog posts | 100+ articles | 2 visitors from search in 30 days, across the entire site. |
| Programmatic SEO pages | 118 pages | No measurable organic traffic in the same window. |
For the 30-day window I analysed, the whole site drew 93 users and 234 page views. 75 of those users arrived as direct traffic - which, honestly, includes my own testing. Search brought 2.
I am not claiming this generalises. It is one site, one operator, one year. What it does rule out is the version of the story where volume alone is enough.
The part that is harder to write
I sell software that measures whether AI assistants recommend a business. So I pointed it at myself. Across 10 questions my own site appeared zero times, and was cited as a source zero times. I re-ran it three days later and got the same result.
Publishing that is uncomfortable, and it is also the strongest argument I have that the measurement is real: a tool that always flatters its owner is not a measurement.
One caveat I found while writing this up, and it belongs in the article rather than in a footnote: my own robots.txt disallows the training crawlers - Google-Extended, GPTBot, ClaudeBot, CCBot - while allowing the answer-time ones. My product measures through Google's grounded search. So I cannot yet rule out that part of that zero is self-inflicted. I am leaving the number exactly as measured and testing the setting separately, rather than quietly re-running it until it looks better.
The bug I only found by paying
Separately, I spent 50 yen buying my own subscription - a real card, the real checkout, the real webhook. The payment succeeded and the plan was never granted. Reading the code had not surfaced it. Neither had a status code: the request returned success while the account stayed on the free tier.
The general form of that lesson is the one I keep re-learning. A process completing is not the same as the outcome being correct, and "it worked once" is not the same as "it keeps working."
What I am not going to tell you
I am not going to end this with the trick that fixed it, because I do not have one yet. Three things were measured and three things did not work. Adding a confident fourth item - the thing you should do instead - would be the exact move that produced the 100+ articles in the first place.
If you are running the same playbook right now, the useful thing is probably not my advice. It is checking your own referrer list for the channel you have been feeding, and seeing whether it appears at all.
Top comments (0)