Google deindexed my site from 6,500 pages to 19. Here's how I'm digging out.
A few months ago my project RealWorth.io (historical inflation + purchasing power, 6 countries back to 1750) was sitting at ~6,500 indexed pages. Then I made a classic mistake: I changed my sitemap twice in one week and hit "Validate Fix" in Search Console before I'd actually fixed anything. Google's indexed count collapsed to 19. Not a penalty — just a total loss of trust in a low-authority programmatic site.
Here's what I've learned climbing back out, in case it saves someone the same hole:
The crash was self-inflicted, and the fix was mostly deleting things. I had ~5,500 programmatic pages, and a big chunk were junk: "$1 in 2021 → $1," pages with "0 weeks of wages" showing up in the actual search snippet. I cut the sitemap from 5,500 to ~4,600 real pages and hard-404'd about 2,600 "ghost" URLs I'd accidentally left crawlable. Fewer, better pages beat more pages every time.
Technical recovery ≠ ranking recovery. My site scores 100 on PageSpeed, has clean schema, 168ms response. Google still won't re-index it. The bottleneck was never technical — it's that a DR 0 domain has no reason for Google to trust it. Fixing the tech just stops you from actively losing; it doesn't win anything back.
The surprise winner: AI search. While Google sat frozen, Bing indexed everything, and Microsoft Copilot citations went from near-zero to 8,000 in last three months. ChatGPT now sends real visitors. If you're starting fresh with no authority, Bing/AI visibility comes way faster than Google.
Still climbing. The next phase is the boring truth everyone avoids: I need actual backlinks, not more tech tweaks. Publishing original data research, one study a month, and doing the outreach I kept putting off.
Happy to answer anything about the deindexation, the recovery, or the programmatic SEO setup. What would you dig into first with a domain that has 5 hours a week and no budget?
Since Copilot citations moved first, write the ten buyer questions someone would actually type and run them weekly in ChatGPT and Perplexity. Log date, question, engine, and whether the answer named you, a competitor, or nobody, because that tells you which pages the models already trust. If you have never checked, a free five prompt snapshot is enough for the first row.
To answer your actual question about 5 hours a week and no budget: I'd skip general link building entirely and go narrow. One piece of data-led content per month is right, but the outreach target matters more than the content quality at this stage. Who specifically has linked to similar historical economic data sites, and have you run a competitor backlink audit to find the exact journalists, teachers, and bloggers who cover this topic and already link to sources like yours?
Deindexing at that scale (6,500 to 19 pages) usually signals either a algorithmic quality/thin content penalty or a severe technical crawl directive issue (like an accidental
noindexrollout, canonical loops, or robots.txt misconfiguration).The first step out of this is separating technical indexability from topical quality. Are you seeing manual action warnings in Search Console, or did coverage/indexing drop silently after a sitemap or site architecture change?
Happy to take a quick look at your GSC patterns or site architecture if you want a fresh set of eyes on it!
No manual actions warnings in Search Console, then or now. I made a bunch of changes at once around that time , some that probably shouldn't have happened and others I should have just waited on. So looking back, I genuinely can't point to one clean trigger.
The technical side seems fine now. From what I understand, the one thing still holding me back is trust. The site is new and has basically no authority, so Google doesn't spend much crawl budget on it and won't commit to indexing most of the pages, even though it can reach them fine. It's not "can Google see this page," it's "does Google think this domain is worth indexing yet." And the answer is still mostly no.
The most interesting part is the distinction between technical recovery and ranking recovery. The founder clearly learned that fixing the system doesn't automatically restore the demand channel.
Exactly right, and I learned that the hard way. This is my first project I've taken public, so I keep figuring things out only after I've already made the mistake.
That kind of mistake is probably giving you some of the most useful lessons this early. What have you learned from it so far?
This comment was deleted 13 days ago