Geolocation API accuracy problem

Hey everyone, I'm running into a persistent, frustrating issue with the core functionality of my web tool, 'What is My Country? - Find Your Current Country & IP Location'. We're facing significant accuracy problems with the IP geolocation data, and it's really impacting user trust and the tool's reliability.

Our current setup relies on server-side IP lookups, combining data from several commercial and open-source IP geolocation APIs like MaxMind GeoIP2 and IP-API. This all runs on a distributed cloud infrastructure. The problem is, we're seeing roughly 10-15% of requests return incorrect country data. This isn't just a minor city-level offset; we're talking about completely wrong countries, especially noticeable with mobile users, traffic from corporate networks, and international visitors. It's causing genuine confusion for users expecting precise IP geolocation.

We've already tried a bunch of things to troubleshoot this. We implemented multiple API providers and developed logic to reconcile their data, hoping to average out inaccuracies. We explored client-side navigator.geolocation, but that proved too intrusive with browser permissions, had CORS issues, and frankly, wasn't reliable enough for our use case. We've thoroughly checked all server-side proxy headers like X-Forwarded-For and X-Real-IP to ensure the correct client IP is consistently being passed to our lookup services. We've even analyzed our network routing and CDN configurations for any potential IP masking or misdirection that might be happening upstream. We also attempted to integrate known VPN/proxy detection services, but this added unacceptable latency and complexity without a significant accuracy boost in the core IP geolocation.

My current hypotheses lean towards the inherent limitations of IP-based geolocation itself, especially with the complexities of modern ISP routing, the surge in VPN usage, and mobile carrier NAT. Dynamic IPs certainly don't make things easier either.

So, I'm looking for some deep technical insights here. Are there specific, perhaps lesser-known, IP geolocation API providers that genuinely excel in accuracy for these challenging scenarios like mobile users or VPNs? What are the most effective server-side strategies or perhaps even network configurations that you've found to consistently ensure the most accurate client IP is always available for lookup, even behind complex corporate firewalls or mobile networks? I'm also wondering if anyone has developed advanced reconciliation algorithms or perhaps even machine learning approaches for intelligently combining data from multiple IP geolocation APIs to achieve significantly higher accuracy than simple averaging. And finally, are there any robust, low-overhead methods for pre-filtering or identifying VPN/proxy traffic *before* it hits our main IP geolocation logic, without introducing significant performance bottlenecks?

Seeking practical, technical advice from those who've truly tackled similar deep-seated IP geolocation accuracy challenges. Help a brother out please.