Request for Refund - Poor Model Performance & Failure to Follow Explicit Instructions (Gemini API)

Gemini API
1

To the Support Team,

I am writing to request a refund for the API usage costs incurred during my
recent sessions with the Gemini model.

I am using the API for professional market research and provided explicit,
repeated instructions for the model to retrieve specific thread-level URLs
from forums (Reddit, TripAdvisor) to validate research claims. I explicitly
forbade the use of generic homepages or deep-linked specific comments that
lack context.

Despite these clear, reiterated directives, the model consistently failed to
execute the task:

  1. It provided generic homepages instead of specific evidence.
  2. When corrected, it provided deep-links to isolated comments rather than
    the parent threads, rendering the citations useless for validation.
  3. It required multiple “correction” turns to achieve a basic standard of
    research, causing significant and unnecessary token consumption.

This lack of “intelligence” in following strict constraints and the inability
to distinguish between a thread URL and a comment URL fundamentally broke the
workflow. I have paid for a service that failed to deliver the basic logical
processing required for the task, forcing me to micromanage the AI at every
step.

I am requesting a credit or refund for the usage associated with these failed
research attempts.

I literally frustrated with taking over decisions, jumping on shallow conclusions, rewriting findings, self correcting with so many mistakes and at the end when close sessions I always see the report that success was 94%.

I spent $100 last month with my observation with at most 20% success and 80% mess still I had to pay and google assumes everything is going good.

Sorry to put all this in thread, just got fried and too much frustrated, thought that Artificial Intelligence will solve my problem but its disaster

1 Like