Series intro. I'm a non-CS solo dev who built and shipped a production stock screener almost entirely by "vibe coding" with an AI agent. This series is about the costs of every shortcut. Part 1 was a reliability incident that tanked my Google rankings. This one is the slow, humbling sequel: getting Google back was supposed to be a checklist. It wasn't.
The thing nobody warns you about
I built StockDigging, a free screener that ranks Korean and US stocks by fundamentals. I did "all the SEO things" — and my AI agent did most of them: clean titles, meta descriptions, semantic markup, a tidy sitemap, fast pages. Technically, the site is in good shape.
Months later, Google was sending me roughly two visits a day.
When I opened Search Console to find the broken thing, there wasn't one. The most common status on my best pages was:
Crawled — currently not indexed.
Google visits the page. Confirms it's crawlable, canonical, fast, allowed. And then quietly declines to index it. No error to fix. The page is technically perfect and Google just... passes. That's not a bug. That's Google saying "this site hasn't earned it yet."
That sentence took me an embarrassingly long time to actually hear.
What the AI and I kept doing (and why it barely moved)
My instinct — and the AI's default — was to optimize. Rewrite titles for click-through. Tune keyword phrasing. Resubmit the sitemap. Add more schema. The AI was genuinely great at this: fast, thorough, technically correct, never bored by the 40th meta description.
And it changed almost nothing. Because none of it touches the actual gate.
The AI did exactly what I asked: clean, correct, on-page SEO. What it — and I — missed is that for a new, low-authority site, on-page tactics are table stakes, not the lever. We were polishing the left column and ignoring the right.
What Google actually rewards (for a new site)
- Authority. Other credible sites linking to you. This is the slow, social, off-page part that no checklist and no agent can shortcut. It's most of the game, and it's the part everyone's "SEO tips" article skips.
- Time and stability. Google distrusts new, and it distrusts flaky. A reliability wobble (the Part 1 story) cost me months of trust, not days.
- Pages that genuinely deserve to exist. Useful, distinct, well-linked from the rest of your site — not thin pages tuned to a keyword.
The uncomfortable summary: the SEO checklist everyone obsesses over is the price of entry, not the thing that wins. You can do all of it perfectly and still sit at "crawled, not indexed" because the site hasn't earned trust yet.
The twist: Google was never the whole story
Here's what actually reframed it for me. I'd quietly equated "SEO" with "Google." So I optimized for Google, and panicked when Google was slow.
Then I looked at where my traffic actually comes from:
Naver — Korea's dominant search engine, and my site covers Korean stocks — was quietly driving the real traffic the whole time. Google was the ~2-visits-a-day line at the bottom.
And the part I did not see coming: ChatGPT was sending me more visits than Google or Bing. AI answer engines are already a live discovery channel, not a someday-thing.
The lesson isn't "Google doesn't matter." It's that "SEO" isn't one engine. Different engines reward different things on different timelines, and by tunnel-visioning on the slowest, least-forgiving one, I missed that people were already finding the site — just through a door I wasn't watching.
What I'd tell myself (and my AI) at the start
- Ask the AI for tactics; own the strategy yourself. It will ship flawless meta tags. It will not spontaneously tell you "the real lever here is authority, and you can't tactic your way there." You have to know to ask.
- For a new site, budget months, not weeks. "Crawled — currently not indexed" is a waiting game and a trust game, not a bug hunt. Treating it like a bug just generates busywork.
- Don't conflate SEO with Google. Look at where visitors actually arrive from before you decide you're failing. You might already be winning somewhere you forgot to check.
- Treat AI answer engines as a channel. Being citable (clear, factual, well-structured) is starting to matter alongside being crawlable.
The single most useful thing I learned: I'd been treating Google SEO as a checklist my AI could finish for me. It isn't. It's a reputation you earn slowly — and while you earn it, your audience may already be finding you through a door you forgot was open.
StockDigging is a free stock screener for Korean (KOSPI/KOSDAQ) and US markets — no login, browse the rankings directly.
Next in the series: the watchdog cron job I added as a safety net, which then started hijacking my main pipeline and firing false alarms for a week.
Top comments (0)