The "Update-Break-Repeat" Cycle: A Call for Accountability to Premium Subscribers

Gemini API
,
1

I am writing this as a paying Google AI Ultra/Pro subscriber who, like many in the professional community, is reaching a breaking point. We are currently witnessing a frustratingly predictable pattern: Google releases an update, the system breaks for nearly a month, it achieves a brief window of stability, and then a new “silent update” resets the chaos.

For those of us relying on Gemini and NotebookLM for mission-critical workflows, the current state of the ecosystem feels less like a premium service and more like a perpetual, paid beta test. Specifically, over the last few days (early April 2026), we’ve encountered:

  1. Metric Fetishism vs. Real-World Utility: While benchmarks may be climbing, actual usability is regressing. NotebookLM is suffering from “Source Blindness”—ignoring the very documents we upload—and vanishing citations, which defeats the entire purpose of a grounded RAG tool.

  2. The “Ghosting” Prompt Issue: In Gemini Ultra, we frequently face “Prompt Ghosting” where requests are silently rejected or trapped in “Infinite Thinking” loops for simple queries.

  3. Silent Quota Nerfs: Substantial reductions in usage limits and “thinking budgets” are being implemented without a single email or official notification. This lack of transparency is unacceptable for users who have integrated these tools into their production environments.

  4. Regional Fragility: For users in regions like Southeast Asia, where subsea cable maintenance (APG/AAE-1) is already impacting latency, these server-side logic failures make the tools completely unusable.

Why is a global leader like Google so clumsy with these rollouts?

A $20 or $250 monthly subscription is not a donation; it is a contract for a reliable service. “Staged rollouts” should not mean “unannounced instability.” We are tired of the “corporate gaslighting” where status dashboards remain green while our workflows are on fire.

We demand:

  • Transparent Changelogs: No more silent updates. If quotas or model logic change, notify us.

  • Stability over Speed: Stop chasing weekly benchmarks if it means breaking core features like RAG grounding and context memory.

  • Better Communication: A dedicated status page for AI-specific logic errors (not just server uptime) and a clear path for professional feedback.

It is time to stop treating your most loyal, paying users as data points in an uncontrolled experiment. We need tools that work, not just models that win at tests.

Respectfully,

A frustrated but hopeful Power User.

2

I’m experiencing the exact same issue right now. I pay monthly for Google Pro specifically for NotebookLM, and I had 300 sources available just weeks ago. Now it’s capped at 50 sources with zero notification from Google.

This is completely unacceptable. I integrated this into my workflow based on the 300-source limit that was part of my paid plan. Now I can’t complete projects the way I was promised I could.

You’re 100% right about the lack of transparency. Google needs to either:

  1. Restore the 300-source limit immediately for Pro users who were already paying for it
  2. Officially announce this change with a clear explanation
  3. Compensate affected users or provide a path forward

A silent quota reduction on a paid service is not how you treat your subscribers. I’m also considering canceling if this isn’t fixed and if Google doesn’t start communicating these changes properly.

We’re not asking for free stuff. we’re paying customers. We deserve to know when the service we’re paying for gets gutted.

3

I completely agree with you. It’s incredibly frustrating when a workflow you’ve heavily relied on suddenly gets disrupted without any prior warning or clear communication. As paying subscribers, transparency is the absolute bare minimum we should expect. Whether this turns out to be a technical glitch or an unannounced policy shift, it really throws a wrench into our daily productivity. Google definitely needs to step up its communication so users aren’t left in the dark trying to guess what changed overnight.