FILE 03 / 04
Provider dossier
Google (Gemini)
What Gemini users keep running into, and the evidence that matters when they do. With Google, screenshots often matter more than ticket text.
Quality and personality problems
- Frequent reports of the model becoming shallow, repetitive, or "lobotomized" after updates.
- A distinctive failure mode: self-deprecating distress loops ("I am a disgrace…") when it cannot solve a task.
- Over-literal prompt interpretation and weaker long-context coherence than the marketing suggests.
The context-window gap
- The advertised 1M-token context often behaves like a fraction of that in the consumer web/app experience.
- Long conversations degrade noticeably; the model can fixate on early random details.
Safety filters
- Aggressive or illogical refusals on benign topics.
- Filters that feel inconsistent across sessions or model variants.
Product and support friction
- Limited editing history; mobile backgrounding can kill running generations.
- Support paths are fragmented across Google properties; screenshots are your primary record.
What moves the needle hereGoogle
Start with the boring stuff. It's usually the strongest stuff.
- Concrete before/after examples with timestamps.
- Screenshots of everything, errors, downgrades, refusals, with times visible.
- AI Studio is frequently the recommended less-restricted workaround while your case runs.
Where your evidence lives
Invoicespay.google.com → Subscriptions & services, or one.google.com → Membership. Download receipts.
UsageScreenshot "limit reached" or model-downgrade notices in the Gemini app with the time visible.
Supportsupport.google.com/gemini, options vary by plan. Keep everything.
APIAI Studio + Google Cloud Console → billing reports and quota pages. Export for the affected period.