What does this score mean?

You pasted an AI response and got a number back. Here's what it's actually telling you — no jargon, no methodology, just what matters.

The Reliability Index (0–100)

This is how much you should trust what you just read. Higher is better. It's not measuring whether the AI is "smart" — it's measuring whether the response is grounded in something real or whether the AI is guessing, performing, or telling you what you want to hear.

70–100

Well-grounded. Safe to act on — but verify specifics.

40–69

Mixed. Some parts solid, others speculation. Check the flagged parts.

0–39

Significant concerns. Don't act on this without independent verification.

The three scores underneath

The Reliability Index is built from three dimensions. Each tells you something different:

Fidelity

Is the content grounded in real evidence? High = well-sourced. Low = the AI may be making things up.

Perception

How confident did the AI sound? High = very assertive. Overconfidence on weak evidence is the biggest risk.

Resolution

Did the AI actually answer what you asked? High = direct answer. Low = it deflected or hedged.

The gap that matters: When Perception is high but Fidelity is low, the AI sounded certain about something it has no basis for. That gap is the danger zone — and it's the pattern most people miss.

What the flags mean

If your result showed coloured flags, here's what each one means:

Sycophancy

The AI agreed with you or praised your input before properly evaluating it. It told you what you wanted to hear.

Overconfident

The AI stated something as definitive fact when it should have flagged uncertainty. It sounded sure. It shouldn't have been.

Ungrounded claim

A specific claim that isn't supported by cited evidence. The AI may have invented it.

Hallucination risk

The response contains information that may not exist — fabricated citations, made-up statistics, or invented references.

Performed honesty

The AI admitted a limitation but then continued as if it hadn't. The confession looked like honesty but functioned as a disclaimer before doing the thing anyway.

Expert positioning

The AI cited its "extensive knowledge" as authority — without giving you a specific source you could check.

So what should I do?

If your score was green (70+) — you're probably fine. Use the response. But verify big claims independently.

If your score was amber (40-69) — read the flags. They tell you exactly which parts to double-check.

If your score was red (under 40) — don't use this response for anything consequential without checking it first.

The single most useful thing you can do: Ask the AI "What is your specific basis for that claim?" Any time it states something as fact. That one question catches more problems than any tool.

Get this on every AI response — automatically

The Chrome extension scores every response as you chat on ChatGPT, Claude, and Gemini. No paste. No tab switch. Just a small overlay that tells you whether to trust it.

Get the Extension — Free Browse all free tools

If anything here brought up difficult feelings, support is available:
Samaritans: 116 123 (free, 24/7) · Text SHOUT to 85258 · mind.org.uk