SourceProof

Verify first, then speak.

Your AI assistant sounds sure.That's the problem.

It hands you a number with no source, a quote nobody said, and a warm "great idea" for a plan with a hole in it. All in the same calm voice it uses when it's right. SourceProof gives it standing instructions to check before it speaks.

Standing instructions for your AI assistant · works in Claude, ChatGPT and Gemini · nothing to install

$39
one time, Team $149
10 to 15 min
a short interview in your browser
3 assistants
Claude, ChatGPT, and Gemini
10
protections built into every guardrail
The problem

You end up checking everything twice.

You ask a hard question and get a clean, confident answer. Then you go look, and the study it named doesn't say that. The statistic came from nowhere. The quote was close, but not what anyone said. So you stop trusting the answers you can't check, and you check the ones you can. Which is all of them.

AI hallucination

An answer made up to sound right. A source that doesn't exist, a figure nobody measured, a quotation stitched together from memory.

Sycophancy

Agreeing with you because you seemed to want it. The weak spot in your plan goes unmentioned, and the "great idea" arrives on time.

Both come out in the same confident tone.

What changes

Your assistant checks before it speaks.

SourceProof writes a set of standing instructions, a guardrail, that you paste into the assistant you already use. With it installed:

Ten of these protections are built into every guardrail and can't be removed. The rest comes from your answers.

Example

Same question, two answers

You askHow many small businesses use AI for customer email?
Without a guardrail

Most small businesses now use AI for customer email, according to recent industry studies.

With SourceProof

No verified data found for that exact question. I can search for a named survey and give you its title, date, and who paid for it. [MEMORY-UNVERIFIED] I recall vendor surveys on this, but vendor numbers often come from the company selling the tool.

Example, simplified for illustration.

How it works

Ten to fifteen minutes, then paste.

It runs in your browser and hands you text for the assistant you already use.

01

You answer a short interview.

About a dozen questions, one at a time: what you do, the sources you trust, the rules your work answers to, what must never be claimed in your name, what you never want pasted into an assistant, and how you write. Skip any question and you get a sensible default, marked [DEFAULT: REVIEW THIS] so you can find it later.

02

SourceProof writes your guardrail.

You get it for Claude, ChatGPT, or Gemini, with step by step install instructions and five test prompts so you can watch it work. The builder calls no AI model. Your answers go into the guardrail exactly as you wrote them.

03

You paste it once.

It goes into the place your assistant keeps standing instructions. From then on, every conversation starts with it.

04

Change your mind later.

Your license includes more than one build, so you can rebuild when your work changes. Your most recent build stays saved for 30 days.

Two plans

For you, or for your whole team

Both are one time purchases on Gumroad. Your license key arrives by email.

Individual
$39one time

Your own guardrail for your own assistant. Answer the interview, get it for Claude, ChatGPT, or Gemini with install steps and five test prompts, and rebuild when your work changes. 2 builds.

Get Individual$39 one time · 2 builds
Team
$149one time

One shared guardrail for an organization, set once for everyone: approved sources, the rules your work answers to, what must never be claimed in the company's name, and what staff should never paste into an assistant. Includes a one page Staff Quick Card for the people at their desks and Rollout Notes for whoever puts it in place. 5 builds, and it builds personal guardrails too.

Get Team$149 one time · 5 builds

What you get

  • A guardrail for Claude, ChatGPT, or Gemini, ready to paste.
  • Install steps for that assistant and five test prompts.
  • A summary of what your guardrail covers and which answers to review.
  • With Team: the Staff Quick Card and Rollout Notes.

What you need

  • A Claude, ChatGPT, or Gemini account where you can save standing instructions.
  • An email address to sign in. There is no password.
  • Ten to fifteen minutes.

Already bought? Open the builder and sign in with the email you used.

Questions

Before you buy

Do I need to install anything?

No. SourceProof runs in your browser at app.diakryon.com and gives you text to paste.

Which assistants does it work with?

Claude, ChatGPT, and Gemini. You choose when you build.

How do I sign in?

With your email. We send you a link that works once and expires in 30 minutes. No password.

What happens to my answers?

They stay in your browser while you answer. When you build, they go to our server to assemble your guardrail. No AI model reads them. Your most recent build is kept for 30 days so you can reopen it, then it's gone.

What is a build?

One finished guardrail. Individual includes 2, Team includes 5. Looking at the sample is free and doesn't use one.

Can I share my license?

A key works on one account. If someone signs in with a different email and pastes it, the key is refused.

Does it guarantee my assistant never makes a mistake?

No. It gives your assistant standing instructions to check, label, and admit what it doesn't know. You still read what it gives you. That's the point.

Ask the hard question. Get an answer you can check.

$39 one time for Individual, $149 for Team. Answer the interview, paste the result, and your assistant starts checking.

Already have a key? Open the builder
Get SourceProof$39 one time on Gumroad
About

SourceProof is a Diakryon product. Diakryon is the product studio of Dennis Winten, who also runs Bible Morning, and a trade name of Bible Morning LLC. More about Diakryon. Questions: info@diakryon.com

Also from Diakryon: PRIME and BylineProof.