If AI is going to tell us things we might use to make decisions, we need the right to access the source.
Information Isn't the Same as Truth
One of the most remarkable capabilities of modern AI is its ability to gather, organize, and explain information. It can take a question that might once have required hours of research and produce a useful answer in seconds.
That's an extraordinary achievement.
But the ease with which AI produces an answer can obscure something important: an answer isn't necessarily a fact, and a convincing explanation isn't necessarily a correct one.
AI can misunderstand a source, combine unrelated information, overlook important context, or confidently present something that simply isn't true. Even when the information is accurate, the conclusions drawn from it may be questionable.
This creates an unusual challenge for product designers. We want technology to make complicated things easier to understand. We want to provide information to our users more readily. But making information easier to consume cannot come at the cost of make it harder to question.
Making information easier to consume cannot come at the cost of make it harder to question.
AI Is the Middleman
When an AI system tells me something about a business, a person, a product, or the world, I want to know where that information originated. The more impactful the information, the more the source becomes critical.
It's not because I distrust everything AI produces, but because I understand the difference between a source of information and a system that interprets it.
AI is not the information. It is the middleman. It finds information, makes connections, summarizes evidence, and presents conclusions. Those are useful capabilities, but they don't make AI the core authority on the underlying facts.
That distinction matters. AI products can't allow the presentation of information to equate with its truth. That is product equivalent of 'take my word for it', and even in the most altruistic use cases, it robs the end user of the clarity, confidence, and context that is likely unique to them.
If an AI-powered product tells a business owner that a competitor has changed its pricing, that customer sentiment is declining, or that a market opportunity is emerging, those statements could affect what the owner does next. And it assumes that the AI understands the situation that business owner faces without the context only that business owner has.
Verification Is a User's Right
I came to believe that access to verification shouldn't be treated as a premium feature, a courtesy, or something reserved for technically sophisticated users. I came to believe that a user has the right to be invited to access the source of information presented to them, to the fullest extent possible.
A user has the right to be invited to access the source of information presented to them.
Further, that right comes with a corresponding responsibility for anyone who creates products that present AI-generated information.
That doesn't mean every piece of information can be made public. Privacy, security, intellectual property, and legitimate restrictions on access still matter. Nor does it mean every conclusion can be verified with a single link. Some conclusions depend on multiple sources, incomplete evidence, or interpretation.
But those limitations are reasons to be more transparent, not less.
When evidence can be shown, show it. When information is incomplete, say so. When something is an inference rather than an established fact, make the distinction clear.
And when a claim can't reasonably be verified, don't present it with the certainty of something that can.
Confidence Is Not Accuracy
It's tempting to address this problem by attaching a confidence score to an AI-generated answer. Confidence assessments can be useful. They give people another signal to consider when deciding how much weight to place on information. But confidence and accuracy aren't the same thing.
An AI system can be highly confident in an incorrect conclusion. A correct conclusion can be expressed with uncertainty. And a numerical score can create an impression of precision that the underlying evidence doesn't support.
So confidence should be treated as an aid to judgment, not a substitute for it. In many cases, users are better served by understanding why information is considered reliable, what evidence supports it, what contradicts it, and what remains unknown. That requires more thoughtful product design than simply displaying a percentage beside an answer.
Designing for Informed Judgment
I encountered these questions firsthand while building SpotTake, an AI-powered business intelligence product that gathers publicly available information and turns it into structured reports.
As the reports became more sophisticated, I realized that providing information wasn't enough. I needed to give customers a reasonable way to evaluate it.
That led to several deliberate product decisions:
- Source references: Provide direct links to the material behind the research, allowing customers to investigate further.
- Confidence assessments: Communicate how well-supported information appears to be, without presenting that assessment as proof of accuracy.
- Information gaps: Identify important things the research couldn't establish, rather than silently treating missing information as settled.
None of these measures guarantees that every report is correct. Nor do they eliminate the responsibility to evaluate the quality of sources and the conclusions drawn from them.
But they give users something more valuable than a polished answer alone: a basis for exercising their own judgment.
And that's the point.
Trust Should Be Earned, Not Requested
There's a tendency in digital product development to think of trust as something we can create through presentation. A professional interface, reassuring language, familiar patterns, and confident explanations can all make a product feel trustworthy.
Those things matter, but they're not enough.
A product can look credible while being wrong. It can feel transparent while concealing uncertainty. And it can make decisions feel effortless by hiding the very information people need to make those decisions responsibly.
Trust isn't simply an experience we design. It's a relationship we earn through the behavior of the product.
That means being willing to expose uncertainty, acknowledge limitations, and give people ways to question what they're being told—even when doing so makes the experience a little less seamless.
The Larger Responsibility
My concern extends beyond any one product.
As AI becomes a more common intermediary between people and information, we risk becoming accustomed to receiving answers without knowing how those answers were established.
Over time, that could change not only how we use technology, but how we understand knowledge itself. When people can no longer distinguish between evidence, interpretation, and assertion, they become increasingly dependent on the systems that present information to them. That's a dangerous trade to make in exchange for convenience.
I'm enthusiastic about what AI can help us accomplish. But I don't believe its value should come at the expense of our ability to question, verify, and understand the information we rely on.
Those of us who build AI-powered products have an opportunity—and I believe an obligation—to preserve that ability.
Trustworthy AI shouldn't ask us to stop questioning. It should make questioning easier.