Plain English. No legal jargon. About 15 minutes. Questions about who they are, what they own, and who should receive it.
Built-in warnings for situations that need a solicitor — business interests, large estates, complex trusts. We say so rather than letting an inadequate will through.
Their answers go to Claude alongside a detailed brief we've written specifying exactly how each clause should be drafted.
Four documents come out in a single pass: the Will, Signing Instructions, a Cover Sheet, and an optional personal letter.
Multiple independent checks before the document is delivered. Four dedicated slides cover each one — what it is, who does what, and what it catches.
No will reaches a user without passing this process.
They print the will and the Signing Instructions. The instructions walk them through exactly how to sign with two witnesses — the most common failure point for DIY wills.
They store the signed original. Done.
If you die, your wife Margaret receives your entire estate. Charitable gifts of £5,000 each go to Marie Curie and Crisis UK, and £2,500 to your brother Peter, before the rest passes to Margaret. If Margaret predeceases you, Sarah and Michael share equally. Sarah is your primary executor, with Michael as substitute.
Sarah Jane Smith (primary) · Michael David Smith (substitute)
Margaret Anne Smith — entire residue · Sarah & Michael equally if Margaret predeceases
Marie Curie (£5,000) · Crisis UK (£5,000) · Peter James Smith (£2,500) · Rolex Oyster Perpetual to Michael
The findings from the four reviewers are synthesised by AI, which determines which issues are material, decides what to fix, and makes the changes to the question flow and drafting brief. This is largely AI iterating on its own output. The human role is high-level direction — setting priorities, not reviewing every individual finding. All changes are tracked and reversible.
In the last 48 hours: 6 new questions added, 12 legal terms replaced with plain English, a survivorship clause error fixed, two contradicting clauses resolved.
After every change, we open the live product in a browser we control and inspect it as a real user would — screenshots of every screen, document labels, the sample output. We also check what's happening underneath: which version of each document is loading from the database, whether the data flowing through the product is correct. This catches problems that look fine on the surface but aren't. Ran four times in the last 48 hours.
Approximately 50 generated wills are sent to an independent TEP solicitor — not affiliated with legalhelp — for structured review. The solicitor provides clause-level feedback: what is wrong, why, and what professional standard was not met. This is not a pass/fail assessment; it is a detailed critique designed to surface the issues AI reviewers cannot find in their own output.
From that feedback, the team selects the issues to address, updates the reviewer briefs and drafting prompt, and re-runs the internal quality loop. This cycle repeats until the output is stable across the sample set. The entire process — which wills were reviewed, the solicitor's findings, what changes were made, the system version before and after — is documented for audit.