Which features use AI, what gets sent, what it is genuinely good at, and where it fails. Written so you can check it rather than take our word for it.
Four features use AI. Everything else does not. AI is used to explain and rephrase — never to decide whether you qualify for something, and never to submit anything on your behalf.
We would rather you understood the limits than trusted us blindly. That is the whole point of publishing this page.
| Feature | Uses AI | What is sent | What it is good at | Where it falls short |
|---|---|---|---|---|
| Ask Sorted | Yes | What you type | Explaining what a situation means and what usually happens next | Cannot know your full circumstances or read your documents unless you paste them |
| Decode a letter | Yes | The letter or document you add | Identifying the sender, the type of letter, the deadline and the safest next step | Can misread poor-quality photos; can miss unusual local variations |
| Plain-English translator | Yes | The text you paste | Turning official or legal wording into normal English | May simplify away a caveat that matters in your specific case |
| Full situation scan | Yes | What you describe | Spotting several connected issues in one description | Breadth over depth — always verify each route it suggests |
| Benefits check | No | — | Rules we have written from official guidance | Only as current as our last review date, which is shown |
| Scam check | No | Nothing leaves the page | Pattern-matching known scam signals in the browser | Cannot confirm a message is genuine, only flag known patterns |
| Deadlines | No | — | A curated calendar with sources and check dates | Curated national dates; your letter may state a different one |
| Guides | No | — | Written and checked by a person against official sources | Carry a last-checked date; rules change between reviews |
Being specific is more useful than being reassuring. The failure modes we see:
That is a reasonable choice and the site still works. The guides, benefits check, deadline calendar, scam pattern-matching and local help all run without sending anything to an AI provider. Start from the guides or the A–Z.
Four: Ask Sorted, Decode a letter, the plain-English translator and the full situation scan. Everything else — the benefits questionnaire, scam pattern-matching, the deadline calendar, the guides — runs without AI.
No. Eligibility questions are answered by rules we have written from official sources, not by an AI model. AI explains and rephrases; it does not decide.
Yes. It can miss something specific to your case, be out of date on a rule that changed recently, or state something confidently that is incorrect. That is why every answer links to the official source and why we publish a corrections log.
The text or document content you provide, plus the instruction asking for an explanation. Not your name, contact details or saved cases, unless you have typed them into the box yourself.
Yes. The guides, benefits check, deadline calendar, scam pattern-matching and local help all work without sending anything to an AI provider.
Tell us and it goes in the public corrections log with the date. That is how this stays checkable.