AI Transparency

Where AI is involved

Every audit starts with measurement, not AI. We run Lighthouse against your page and check it against a fixed set of UX heuristics, UX laws and GEO rules. Those scores and checks are deterministic — the same page produces the same numbers.

AI takes over after the measurement, to explain and prioritise. These parts of a report are AI-generated:

• Plain-language explanations of a finding, shown when you ask "Help me understand".
• The Copy & Messaging review of your visible page copy.
• Screen analysis of screenshots you upload, and flow analysis across a sequence of screens.
• Business Intelligence — competitor research, benchmarks and growth opportunities, on Pro and Studio plans.
• Project-level summaries that synthesise several page and screen audits into one report.

Which AI systems run

We use third-party models — Anthropic Claude, Google Gemini — under our own prompts, our own UX knowledge base and our own analysis pipeline.

Business Intelligence additionally reads several pages of the site being analysed and searches the live web, so its output depends on sources we do not control.

theuxbites is operated by David Rozsa, who is responsible for the output you see — not the model vendors.

What AI is marked as

Every block of AI-generated content in the product carries a visible "AI-generated" notice at the point where you read it — not buried in these terms. Exported Markdown reports carry the same notice in the file itself, so it survives being shared with a developer or pasted into another tool.

What we do not do

To be unambiguous about the boundaries:

• There is no chatbot. You are never in a conversation with an AI on theuxbites.
• We do not generate images, video or audio.
• We do not use emotion recognition or biometric categorisation, and we do not analyse people in uploaded screenshots.
• AI does not decide anything about you — not your plan, your billing, your access, or whether your account stays open. Those are rule-based and human-controlled.
• We do not train models on your audit data.

AI can be wrong — and what that means for you

AI-generated findings can be inaccurate, incomplete or confidently wrong. Business identification in particular can fail on domains whose name collides with a better-known company; we grade our own confidence and tell you when it is low, but we cannot catch every case.

Check anything that matters against your own site before you act on it. If you publish or forward a report, you are the one publishing it — review it first.

Legal basis and contact

This statement is published under Article 50 of Regulation (EU) 2024/1689 (the EU AI Act). Questions about it go to hello@theuxbites.com.