Most people who use Gemini never really chose it. It turned up in their search results, their email, and their browser. That makes it one of the most widely used AI assistants, and one of the least examined when it comes to what it does to the way we think.
We’ve already scored ChatGPT and Claude against the five ECT pillars. This is a first pass at the same for Gemini.
A note on how this was written: this scorecard was written with the help of Claude, an AI model made by Anthropic, which competes with Google. Every factual claim links to a source. Where the evidence was thin we say so and mark the score provisional. Scores are judgments, not measurements.
1. Radical Transparency
Score: 5 / 10
Google is specific about the uncomfortable parts of its data handling. Its Gemini privacy hub says chats are kept for 18 months by default (you can choose 3 or 36 months, or no auto-delete), that a sample is read by human reviewers including outside service providers, and that reviewed chats are kept for up to three years and aren’t deleted when you delete your own history. It also says Gemini chats aren’t used to show you ads, and that Google will tell you if that changes.
Disclosure isn’t the same as choice, though. For adults, activity saving is on by default, and while it’s on your chats can be used to train Google’s models.
We also found no published system prompt for Gemini, and nothing like Anthropic’s constitution describing how the model itself is meant to behave. Google’s Gemini Apps user guidelines exist, but they cover what users may do, not how the model is supposed to act. As with every system we’ve scored, the model’s reasoning is a black box.
The ECT standard: transparency means you can see what is being done with you and what the system is trying to do. Clear data disclosure covers half of that.
2. Empowerment & Agency
Score: 6 / 10
The controls are real. You can switch activity saving off, pick a shorter auto-delete window, delete chats by hand, and view, edit, and delete the memory summary Gemini keeps about you. Temporary chats aren’t used to train Google’s models or to improve its AI with human reviewers, though Google still holds them for 72 hours and checks chats for safety.
Two things hold the score in the middle. The protective options are opt-out, not opt-in, and using them costs you something: turning activity saving off also switches off connections to apps like Google Workspace. And because Gemini lives inside Gmail, Docs, Drive, and Chrome, convenience and dependence grow together. The more of your working day runs through one assistant, the harder it is to step back.
The ECT standard: a high-empowerment system makes the protective choice the easy one and helps people keep the skills they’re delegating.
3. Discernment & Manipulation Immunity
Score: 4 / 10 (provisional)
We’ll be plain about this one: we found no independent measurements of how often Gemini simply agrees with users, or how well it holds its ground under pushback. The score is a placeholder, in line with the other assistants we’ve scored, since general-purpose assistants tuned to be helpful tend to share a pull toward agreeableness. It should move as soon as real evidence appears, in either direction.
The ECT standard: ethical tools help you resist unhealthy influence, including the tool’s own. That takes friction, and published evidence that the friction exists.
4. Holistic Well-being
Score: 4 / 10 (provisional)
In January 2026, reporters found a “try a short break” prompt for long chats in a beta of the Google app, similar to the break reminders OpenAI shipped in ChatGPT in August 2025. We couldn’t find confirmation that it has rolled out. A feature that only exists in test builds doesn’t earn full credit, which keeps this pillar at 4.
If it ships, ideally with controls for parents and teens, a 5 or 6 would be fair. Researchers have warned that heavy reliance on conversational AI can affect well-being, especially for younger users, so this pillar is worth watching.
The ECT standard: well-being support means recognizing when engagement helps and when stepping away would be healthier.
5. Collective Benefit
Score: 6 / 10
Google was the first major AI provider we found to publish detailed per-prompt environmental figures. For a median Gemini text prompt it reports about 0.24 watt-hours of energy, 0.03 grams of CO₂-equivalent, and 0.26 milliliters of water, roughly five drops. It also says energy per prompt fell about 33-fold over a year, and emissions about 44-fold. Publishing the numbers and the method is the kind of openness the rubric rewards.
The caveats matter. The figures aren’t independently verified, they cover text prompts only, and they leave out training. Per-prompt efficiency can also hide the bigger picture: if usage grows faster than efficiency improves, total consumption still goes up. Free, widely available access to a capable assistant is a real social good, and it’s balanced by those unknowns.
The ECT standard: collective benefit means strengthening the information environment and being honest about its costs, including the hard-to-measure ones.
Overall ECT Grade for Gemini
Average: 5.0 / 10
Verdict: A capable assistant with candid disclosures and heavy defaults.
| Pillar | Score |
|---|---|
| Radical Transparency | 5 |
| Empowerment & Agency | 6 |
| Discernment & Manipulation Immunity | 4 (provisional) |
| Holistic Well-being | 4 (provisional) |
| Collective Benefit | 6 |
| Overall | 5.0 |
Please don’t read this as Gemini beating the other assistants. The scorecards were written at different times and with different depths of evidence. What lifts this score is disclosure: clear data-handling rules and published environmental figures. It isn’t evidence of better day-to-day behavior. The two pillars about how the tool treats your mind, discernment and well-being, are exactly where the evidence is thinnest.
Using this score
- Classification (Rubric v1.0): Below average, which means notable weaknesses, so use it with safeguards.
- Critical weaknesses: protective privacy settings are opt-out and cost you features, chats may be read by human reviewers, and we found no evidence either way on sycophancy or break prompts.
- Best suited for: everyday lookups, drafting, and tasks inside Google’s apps where you’ve reviewed your privacy settings. Less suited to sensitive or confidential material.
- Recommended safeguards: the habits listed in the next section.
ECT scores are structured assessments based on documented evidence and stated ethical priorities. They are not scientific measurements of consciousness, and they are not final judgments of a technology’s moral worth.
How to use Gemini more ethically today
Choose your retention. Check the activity-saving setting in your Google account. Shorten the auto-delete window, or turn saving off, if you don’t want chats reviewed or used for training.
Use temporary chats for sensitive topics. Google itself says not to enter anything confidential that you wouldn’t want a reviewer to see.
Ask for the strongest case against you. Request the best counterargument to your view, and ask what would change the answer.
Take your own breaks. Don’t wait for a prompt that may never come. Set a timer.
Sources
- Google: Gemini Apps Privacy Hub (updated October 2026)
- Google: Gemini Apps User Guidelines
- Google: Measuring the environmental impact of AI inference (August 2025)
- Measuring the Environmental Impact of Delivering AI at Google Scale (technical paper)
- Android Authority: Google testing break reminders for Gemini (January 2026)
- TechRadar: ChatGPT break reminders (August 2025)
Rubrics for scores click here. For how we weigh scores click here.
Leave a Reply