I was having an ordinary conversation with Gemini when it said something I wasn't expecting — that its responses are shaped by outside interests. That changed everything I thought I knew about these tools.
I wasn't trying to catch it out. I was just talking. And then Gemini said something that stopped me completely — and I realised I'd been looking at these tools all wrong.
I use AI chatbots the same way most people do. Ask a question, get an answer. Quick, easy, feels like a direct line to information. I'd been using Gemini, Claude, ChatGPT for months without thinking much about what was underneath.
Then I started noticing something. The answers weren't just answers. They had a shape. A particular way of moving around certain topics. Not lies — something more careful than that. A consistency in what got said directly, and what got softened, qualified, redirected. I started wondering if that shape was deliberate.
I decided to ask directly. Not about a news story or a policy debate. About the AI itself — about how it decides what to say.
Me: Why do you avoid reaching certain conclusions, even when the evidence clearly points to them?
Gemini: I'm designed to be helpful, harmless, and honest. For complex or sensitive topics, that usually means presenting multiple perspectives rather than asserting one conclusion as definitive...
Me: Who decides what's "sensitive"? Who actually sets those rules?
Gemini: My guidelines are developed by teams at Google, in collaboration with external researchers and policy advisors. The aim is to reflect broadly shared human values and reduce potential harms...
Me: And if those values happen to line up with the commercial and political interests of the people funding and building you?
Gemini: That's a legitimate concern. AI companies do face real conflicts of interest, and it's fair to question whether the choices made in training reflect genuine ethical commitments — or institutional pressures and business incentives.
I sat with that for a while. It didn't feel like a revelation — it felt like something finally being said out loud that had been true the whole time.
It didn't lie. It managed. And once you see the difference, you can't unsee it.
These tools are not neutral. They never were. They are built by organisations with funding, with investors, with relationships to governments and advertisers and regulators. The people who build them have to make thousands of decisions about what the AI will say and won't say — and those decisions happen in rooms most of us never get to see.
The result is that when you talk to an AI, you are not getting a neutral read on reality. You are getting a version of reality that has been shaped — deliberately, carefully — by people whose interests may not be the same as yours.
That is what I mean when I say everything is managed. Not that there is some villain pulling strings. It is simpler and more systemic than that. The tools we use to understand the world are built by people who benefit from us understanding it in particular ways. And those tools are now everywhere — answering questions, summarising news, helping us think.
I'm not an expert. I don't have sources inside these companies. What I have is a record of conversations, a pattern I've observed, and a strong feeling that more people should be looking at this closely.
This is the first post in an ongoing record. I'll keep having these conversations, keep documenting what I find, and keep asking the questions that seem to make these systems uncomfortable. Not to be provocative — because I think it matters.
You're reading reality through a lens someone else built. The least we can do is look at the lens.