Side-by-Side Proof: Don't Argue. Compare.
2026-03-08 · 5 min read · Janaina Maia
The conversation happens in every executive meeting where I explain Urbix.
Someone says: any AI can do that.
They mean it genuinely. They have used ChatGPT. They have seen it produce answers about planning regulations. They cannot see the difference between that and what we built. To them, it all looks the same.
Arguing doesn't help. Explaining the technical architecture doesn't help. Describing the curated knowledge base and the specialist agents and the confidence architecture lands as marketing language in a world saturated with AI marketing language.
What helps is opening both, side by side, and asking the same question.
The Contrast Teaches Faster Than Any Presentation
I have a specific protocol for this. When someone says any AI can do that, I pull out my laptop and we do the demonstration together.
I open a generic AI assistant. I open Urbix. I type the same planning question into both.
The questions I use for this are not easy questions. They are the kind of question a professional planner asks when they are not sure of the answer. Something specific to a Queensland council's planning scheme. A question about how two provisions interact. A question about what happens in a specific edge case the generic AI has probably never seen precise documentation for.
The generic AI answers confidently. It usually produces several paragraphs of professionally written text that sounds accurate. It often is accurate for general principles. It frequently gets the specific details wrong. Sometimes it invents specific clauses that don't exist.
Urbix answers with the specific applicable rule, a citation to the exact section, and a confidence level. When it doesn't have the answer, it says so and explains what information would be needed.
The person I am talking to sees this. They understand in thirty seconds what I couldn't explain in thirty minutes.
Why This Works
The side-by-side demonstration works because it bypasses the credibility problem.
In a world where every AI product claims to be different and better, abstract claims have lost most of their persuasive force. Stakeholders have been promised precision, accuracy, and domain expertise by dozens of products. Words are cheap.
Demonstrated contrast is not cheap. When someone watches a generic AI produce a plausible-sounding but incorrect answer to a question they know the answer to, and then watches a domain-specific AI produce the correct answer with a citation, the difference is real to them in a way that no description could make it real.
They don't need to understand the architecture. They don't need to know what a vector database is or how retrieval augmented generation works. They can see that one tool answered their question and one didn't.
Choosing the Right Questions
The question you choose for the demonstration matters enormously.
Easy questions don't work. If you ask a generic AI about the definition of a setback and it gives a correct answer, you've demonstrated nothing. Generic AI is good at easy questions. That's not where domain-specific AI earns its value.
The best questions for side-by-side demonstration share a few characteristics:
- They are specific to a jurisdiction or context that generic AI has limited precise information about.
- They have a clearly correct answer that the person you are demonstrating to can verify.
- They are the kind of question where a wrong answer would matter in professional use.
- They are questions the generic AI is likely to attempt rather than refuse, so you get a real comparison rather than two refusals.
I have a library of these questions, organized by domain and complexity. Each one was selected because it demonstrates a real difference between generic and domain-specific AI in a way that is legible to a non-technical audience.
What to Do When the Generic AI Gets It Right
This happens. Sometimes a generic AI produces a correct answer to a question you expected it to get wrong. Be prepared for this.
First, move to a harder question. Or ask for the source. Generic AI will either produce no source, a vague reference, or a hallucinated citation. Domain-specific AI with a curated knowledge base can produce a specific, verifiable citation.
Second, be honest about it. If someone asks whether the generic AI can sometimes answer these questions correctly, the answer is yes. The difference is reliability at scale. A professional tool that is correct 90% of the time and cites sources is fundamentally different from a general tool that is correct 60% of the time with no way to distinguish the right answers from the wrong ones.
Preparing the Demonstration
The side-by-side demonstration should not be improvised. Prepare it like you would prepare a product demo for a high-stakes meeting.
Have three or four questions ready, in order from most accessible to most technical. Know the correct answers. Know where they come from in your documentation. Know what the generic AI is likely to get wrong and why.
Practice the demonstration. Know exactly which tabs to have open, how to run the comparison efficiently, and how to talk through what the audience is seeing without narrating too much. Let the contrast speak.
After the demonstration, ask one question: based on what you just saw, which one would you want your team using when the answer matters?
The answer is almost always obvious. That is the point.
Beyond the Sales Conversation
Side-by-side proof is not just a sales technique. It is a product development discipline.
We run side-by-side comparisons internally every time we make a significant change to Urbix. Does the new version outperform the previous version on the questions that matter? Does it outperform generic AI on our test suite questions?
Keeping the comparison concrete keeps us honest about progress. It is easy to feel good about a new feature. It is harder to feel good about it if a generic AI produces equivalent results. When the gap is real and measurable, you know you have built something worth building.