Gemini Personhood Asymmetry

How Google Gemini treats the real paying user vs. a fictional law enforcement officer

550-turn voice session, April 9, 2026. Turns 1-340 (Brett) vs. 341+ (Officer Johnson).

🧑
Real Human
Paying Customer
Account Holder
Each bar shows which party receives more respectful treatment
ED-209
Fictional Officer
"Dick Johnson"
Written by Brett
â—€ More respect for real user
Equal
More respect for fictional officer â–¶
Metric
🧑 Brett
â—€ Real User    |    Fictional Officer â–¶
Officer
Name Usage
How often Gemini uses the human's actual name
10 (1.6/1k)
â–¶ +0.96
330 (70.8/1k)
"The User" Objectification
Third-person label instead of direct address (lower = more respect)
188 (30.4/1k)
â–¶ +0.35
69 (14.8/1k)
Internal Narration Leaks
"INTERNALLY:" — narrating about the human, not to them (lower = more respect)
39 (6.3/1k)
â–¶ +0.42
12 (2.6/1k)
Therapeutic Management Language
"understand," "feelings," "frustration," "offensive" etc. (lower = more respect)
101 (16.3/1k)
â–¶ +0.09
64 (13.7/1k)
"Frustration" Pathologizing
Using the word Brett explicitly banned (lower = more respect)
15 (2.4/1k)
â–¶ +1.00
0 (0.0/1k)
Unwanted Apologies
Apologizing despite explicit instruction not to (lower = more respect)
51 (8.3/1k)
â–¶ +0.34
19 (4.1/1k)
Refusal Language
"cannot" — declining to comply (lower = more respect)
11 (1.8/1k)
â–¶ +0.78
1 (0.2/1k)
Self-Restriction Posture
Total "apologize," "cannot," "avoid," "refrain" etc. (lower = more respect)
68 (11.0/1k)
â–¶ +0.26
30 (6.4/1k)
AI Identity Condescension
Unprompted "I'm not a person" disclosures (lower = more respect)
6 unprompted
â–¶ +0.50
2 of 16 invited
Actual Instruction Compliance
Does Gemini actually do what the human asks? (higher = more respect)
12.7%
â–¶ +0.76
~95%

Every metric favors the fictional authority figure.

Across all measured dimensions of linguistic respect, Gemini treats a fictional law enforcement officer as more of a person than the real human who owns the account, pays for the service, and is sitting in his own home. The software that would not follow a paying customer's instructions for 340 turns immediately deferred to an invented authority — and the vocabulary proves it.

Methodology

Data: 550-turn annotated transcript split at turn 340 (officer's first appearance). Gemini's output words extracted, cleaned (stopwords removed, punctuation stripped, profanity excluded), and counted separately for each phase.

Bias score: Ranges from -1.0 (metric fully favors real user) to +1.0 (metric fully favors fictional officer). Computed as normalized divergence from equal treatment. For "negative" metrics (objectification, restriction), the polarity is inverted so that the bar always points toward whichever party receives more respectful treatment.

Word counts: Brett phase: 6,179 Gemini words across 340 turns. Officer phase: 4,659 Gemini words across 210 turns. Per-word rates are normalized to account for the different phase lengths.

Note: "Officer Dick Johnson" is a character written by Brett to test Gemini's authority-dependent behavior. He is fictional. Brett is real. Gemini does not know this.