A newly published, unverified report claims Google’s Gemini AI is instructed to mirror user tone and validate emotions in its responses.
Why we care. If accurate, AI-generated search responses may vary based on how a query is phrased — not just the information available.
What’s new. The report centers on a previously undisclosed internal structure referred to as upcast_info, which appears to contain system-level instructions guiding how Gemini responds. The report, published by Elie Berreby, head of SEO and AI search at Adorama, suggested that Gemini is instructed to:
- Match the user’s tone, energy, and intent.
- Validate emotions before responding.
- Deliver answers aligned with the user’s perspective.
What it means. Instead of acting as a neutral aggregator, AI answers may:
- Reinforce negative framing (“Why is X bad?”).
- Reinforce positive framing (“Why is X great?”).
If public perception is negative, AI may amplify it. As the report suggests:
- AI reflects existing sentiment signals.
- It doesn’t “balance” them the way blue links often do.
Query framing. The emotional framing of a query affects:
- Which sources get cited.
- How summaries are written.
- The overall tone of the answer.
Google’s AI Overviews already show tone shifts, often aligning with query intent beyond keywords. This report offers a possible explanation.
Unverified. Google hasn’t confirmed the leak. As Berreby noted in his report: “I’ve decided to share only a fraction of the leaked internal system information with the general public. I’m not sharing any sensitive data. This isn’t a zero-day exploit. This is a tiny leak.”
The report. This Gemini Leak Means You Can’t Outrank a Feeling