We Deleted All Our Demo Data. Here's Why That Made Our Product Better.
Fake scores and simulated audits feel helpful until you realise they're building a product on top of a lie. Here's why we ripped out every line of demo data from SiteNexis — and what we replaced it with.
Every SaaS product that handles complex data eventually faces the same temptation: when the system has no real data to show, fill the screen with something that looks like data. Simulated audit scores. Pre-seeded graph nodes. Placeholder metrics labelled "example." The intent is usually good — help new users understand what the product does before they have run their first real analysis. But the consequence is invisible and corrosive: the product ships a false contract with the user from the very first interaction.
What Demo Data Actually Is
Demo data is a lie told with good intentions. It presents a dashboard as populated when it is not. It shows scores as computed when they were invented. It renders graphs as real when the underlying data does not exist. Most products do this. Almost none talk about it. We called our implementation what it was: a DEMO_USER passthrough in our auth layer, a buildDemoResults function that generated plausible-looking numbers, and a createDemoAudit function that let any unconfigured instance appear fully functional. It shipped as a feature. It was a liability.
The Trust Intelligence Paradox
SiteNexis is a Machine Trust Intelligence platform. It tells you how AI systems form, maintain, and lose trust in web content. Entity consistency. Schema alignment. Contradiction detection. Cross-source validation. We were building a product about honesty in machine-readable content — while running demo data through our own system that directly violated every principle we were asking our users to apply to their websites. The paradox was impossible to ignore once we named it.
▲If your product is about data quality, running fake data in your own system is not a UX shortcut. It is an architectural contradiction that every sophisticated user will eventually notice.
What We Replaced It With
The Zero-Fake Data Architecture (ZFDA) has one rule: if data does not exist, say so. We deleted demo-store.ts and demo-health-data.ts entirely. We rewrote the auth layer so that it never falls back to a pre-seeded demo identity. We replaced every fallback that previously returned fake scores with an explicit empty state: "No data available — run an audit to generate real analysis." We introduced the Graceful Truth Layer to give every API response a machine-readable state field — complete, partial, or empty — so the frontend always knows exactly what it is rendering.
What Happened Next
The product got smaller and more honest. The dashboard, when unconfigured, shows nothing — because there is nothing to show. When an audit is running, it shows partial data with a clear indicator that the crawl is in progress. When an audit completes, it shows real computed results from real crawled pages. There are no simulated nodes in the perception graph. There are no invented scores in the AI Visibility panel. Every number on the screen was computed from a real website. This is what the product was supposed to be from the start.
Run a real audit on your domain. See what AI systems actually compute about your site — no demo data, no simulation.
Run Real Audit