Case studies: tying 5xx in logs to lost rankings

Case studies: tying 5xx in logs to lost rankings

Server errors the bot saw, and the cost.

→ E-commerce post-mortem (Lumar) — logs showed Googlebot hitting 503s on 12% of crawls during a 3-day load spike; affected category pages dropped an average 4 positions and took 5 weeks to recover.
Takeaway: a temporary 5xx has a long ranking tail.

⭐ Pick of the week — Onely error-rate threshold study — found Googlebot started reducing crawl rate once logged 5xx exceeded ~5% of its requests, a self-protective backoff. Quantified the slowdown.
Takeaway: errors don't just lose pages, they shrink your whole crawl budget.

→ Google's 'reduce Googlebot crawl rate' doc — the mechanism behind the backoff.
Takeaway: understand the backoff before you trigger it.