The curious users ultimate guide to asking better AI questions

2. Who’s perspectives are/aren’t included?

AI doesn't actually think, but it can sound remarkably human in its responses. Just like people, AI  systems develop preferences and biases based on its training data and how they are created.

Think of it this way: if you only consume news from one political viewpoint your entire life, your opinions would naturally lean that direction. AI models work similarly - they reflect the perspectives embedded in their training materials.

3. Whose work is this?

We're living through a fascinating legal moment for AI copyright concerns. Courts worldwide are wrestling with whether AI training constitutes copyright theft or fair use. The jury's literally still out, but multiple high-profile cases pit AI companies against content creators. Some claim AI training is outright theft. Others argue it's perfectly legal. Nobody knows for certain where the law will land.

This uncertainty could affect you as an AI user. It's unclear whether you might be liable for using AI models trained on copyrighted material. Even if you're using a third-party service, you could potentially face legal challenges if you aren’t careful.

AI models come with different licensing arrangements:

• Open source models: Completely transparent and free to use. You can access the code, training data, and creation methodology.
• Open weight models: The trained model is free to use and modify, but the underlying code and training data may not be available.
• Semi-open licensed models: Typically free for personal use, education and research, but have restrictions on commercial applications.
• Proprietary models: Closed systems with strict licensing terms and limited access.

4. Whose story is this?

Information isn't neutral - it belongs to someone, and the concept of digital sovereignty captures this: it's about who controls stories, information, and personal details. The more sensitive the information, the more control people want, and should have, over it.

Consider Indigenous communities with traditional knowledge or local customs at risk of misrepresentation by sources that lack cultural understanding. When AI confidently explains practices from cultures—whether it has been trained on them or not—the information often comes from people outside those communities. This creates a significant risk of incomplete or harmful explanations.

The fundamental problem is that AI systems cannot distinguish between accurate and flawed information. They generate responses based solely on patterns in their training data, regardless of whether that data authentically represents the stories being discussed.

This principle applies when AI systems make claims about specific communities, cultures, or regions. Who permitted them to tell those stories? Were local voices consulted, or is this information gathered from outsiders looking in and making assumptions?

Being thoughtful about whose voices are represented and who controls shared data helps create a more respectful and accurate understanding of our world. The goal is to become a culturally aware user who recognises the difference between authentic community voices and external assumptions.

5. Is this still true?

AI models are like snapshots frozen in time. They're trained on data from a specific period, then released into a constantly changing world. This creates a fundamental problem: whilst the world keeps evolving, your AI model remains stuck with yesterday's knowledge and assumptions.

Consider an AI model trained to predict travel demand using pre-COVID data. It would expect normal patterns - busy airports, packed hotels, regular business trips. But COVID changed everything. Travel patterns, consumer behaviour, and business practices shifted dramatically. The old model would make spectacularly wrong predictions because it couldn't account for these changes.

This happens in many other ways, too. Fashion trends change, seasons or weather change, political situations are in a state of flux, technology advances, and social attitudes shift. Your AI model might still be giving responses based on outdated assumptions.

6. What am I sharing?

Personally Identifiable Information (PII) is data that could potentially identify who you are, directly or indirectly. What used to require obvious details like your name or email now happens through seemingly anonymous data points. Researchers have shown that just four location data points can identify most people uniquely.

Here's the modern privacy challenge: information that seems harmless individually can become deeply personal when combined. Your phone's operating system, browser type, number of devices, and city location might seem innocent, but together they create a unique fingerprint that identifies you specifically.

Consider the information you may share with AI systems and what puzzle pieces you are offering. Your web searches, browsing habits, location data, health queries, political interests, and shopping patterns individually seem innocuous. Combined, they could reveal your deepest concerns, beliefs, and vulnerabilities in surprisingly intimate detail.

This is why choosing which AI systems to trust becomes crucial. You're not just sharing today's question - you're potentially contributing to a growing profile of yourself that becomes more revealing with every interaction.

The business risk

The same principles apply to organisational data. Seemingly low-level business information can combine to reveal sensitive commercial insights. Unlike personal privacy, business information often has clear sensitivity levels - from public information through sensitive data to confidential. But even public information can become problematic when aggregated and analysed. Your casual questions about company processes, market research, or strategic thinking could inadvertently reveal more than intended.

7. What happens when we stop questioning?

There's something deeply reassuring about computer systems that speak with absolute confidence. When a system tells you something definitively, it's natural to assume it must be correct. But this unquestionable trust becomes particularly problematic with generative AI. These systems excel at sounding authoritative and correct, even when they're completely wrong or are misrepresenting another person's work.

Even if you disagree with an AI's response initially, repeated exposure can gradually shift your perspective to align with the model's outputs—a phenomenon known as system bias. Whether it's believing that socks and jandals are great together. or accepting flawed judgements about groups of people, this subtle influence can be serious. Instead of celebrating our similarities and differences, biased AI responses risk driving us apart by reinforcing harmful stereotypes or misconceptions over time.

8. How can I be a better editor with AI?

Being a good editor requires different skills from being a good writer. Similarly, being a good AI reviewer requires different skills from being a good AI user.  You need to actively look for problems, question assumptions, and spot inconsistencies.

Without the right tools and workflow, even well-intentioned people will fail to catch AI errors effectively.