// how it works
What Reality Filter does
1 — The problem
A new AI tool launches. Everyone says it changes everything.
“10× faster. Beats every model on every benchmark. This changes everything.”
LAUNCH POST · 40,000 REPOSTSYou can't tell if that's true without losing half an hour to benchmarks, threads and reviews.
2 — What you do
Type the name. That's the whole input.
No sign-up, no setup, nothing to install. Just the name of a tool, model or app.
3 — What we check
We gather the evidence, and name every source.
Real benchmark numbers and named reviewers — never “reports suggest”.
4 — What you get back
One clear answer, and how much is still hype.
Four possible answers: worth trying now · watch later · too early · mostly fog machine. You can decide from this alone — the full evidence is one click away.
Every week another AI tool launches claiming to change everything. Reality Filter tells you whether it actually does — one clear answer, backed by evidence we name and sources we quote, so you can decide in about a minute instead of an afternoon.
The problem it solves
Every AI launch sounds identical. “10× faster.” “Beats every benchmark.” “This changes everything.” Some of it is true. A lot of it isn't. And the only way to find out has been to spend twenty or thirty minutes yourself, digging through benchmark tables, developer threads and reviews, before you can decide whether to switch to it, pay for it, or ignore it.
Reality Filter does that digging and hands you the conclusion.
What you actually do
You type the name of an AI tool, model or app into the search box — ChatGPT, Gemini, Claude, Cursor, or whatever you saw people posting about this morning. That's the entire input. There's no sign-up, no account, and nothing to install.
What you get back
One answer, out of four. Not a score, not a wall of research — a decision:
- WORTH TRYING NOW
It does what it says. Go ahead and use it.
- WATCH LATER
It's real, but there are catches worth knowing before you commit.
- TOO EARLY
It might turn out fine. The evidence just isn't in yet.
- MOSTLY FOG MACHINE
The claims run well ahead of anything anyone can show.
Alongside it you get a hype risk reading — a plain sentence on how much of the story is still unproven — and a short note on what you should do next, written separately for people building with it and people buying it.
Where the answer comes from
Every verdict is researched by hand from current sources, and it names them. Actual benchmark names and scores. Named reviewers and independent testers. What real users and companies report after living with the thing. You'll never read “reports suggest” — if we can't point at who said it, it doesn't go in.
We also publish the evidence against our own call when it exists. A verdict you can only agree with isn't worth much.
Why you can trust it
We keep a public, dated track record — a small, hand-picked set of calls, a few added each month, each one published before anyone knew who'd turn out to be right. Not every verdict goes on it, deliberately: it's a record of the calls we choose to stand behind, not a log of everything we've ever looked up. The ones that do go on it stay there, and we don't quietly edit them — when the facts change we re-check the tool, update the verdict, and leave the original call standing next to the correction.
That has already happened: GPT-5.6 was filed as “too early”, then moved to “worth trying” when it became generally available — and the original call is still quoted inside the packet rather than deleted. We've also called a product doing $1.2B a year “mostly fog machine”. If this only ever agreed with press releases, it would be worthless to you.
What it costs
Nothing. Reading any verdict is free and needs no account. If we haven't checked something yet, you can leave an email and a real person researches it by hand and sends you the answer, usually within a day.
Check a tool now