Proof interview prompt
A proof interview prompt turns the AI into an interviewer. It asks one question at a time until it has a dated, sourced fact list: clients, results, credentials and reviews. It's the fastest way to get proof out of your head and onto the page.
- Stage
- Plan, step 4 of 40
- Use it
- Before you write any copy, and once a year after that. The fact list feeds every other writing prompt.
- You need
- 30 quiet minutes
- Access to your invoices, reviews and project list
- You get
- A numbered fact list with proof, a date and a publish decision for each fact. Output schema included.
The prompt
Interview me to build a fact list for my website.
Ask one question at a time and wait for my answer.
Cover years in business, number of clients or projects,
results with numbers, certifications, awards, notable clients,
reviews, guarantees and why buyers choose me.
For every answer, ask how I could prove it and since when it's true.
Never suggest an answer. If I don't know, write UNKNOWN.
After 15 questions, return the fact list as a numbered table:
fact, proof, true as of, publishable (yes, no or ask client).JSON version, for teams and automations
Fill each value under "inputs", keep the rest. It follows the prompt schema. When to use JSON.
{
"$schema": "https://websitesdesigner.com/insights/prompts/prompt.schema.json",
"id": "proof-interview",
"name": "Proof interview prompt",
"stage": "plan",
"step": 4,
"of": 40,
"version": "2026-11-19",
"url": "https://websitesdesigner.com/insights/prompts/proof-interview/",
"inputs": {},
"instructions": [
"Interview me to build a fact list for my website.",
"Ask one question at a time and wait for my answer.",
"Cover years in business, number of clients or projects, results with numbers, certifications, awards, notable clients, reviews, guarantees and why buyers choose me.",
"For every answer, ask how I could prove it and since when it's true.",
"Never suggest an answer.",
"If I don't know, write UNKNOWN.",
"After 15 questions, return the fact list as a numbered table: fact, proof, true as of, publishable (yes, no or ask client)."
],
"output": "A numbered fact list with proof, a date and a publish decision for each fact.",
"checks": [
"Every fact has a number, a name or a date.",
"Every fact says how it could be proved.",
"Client names and results are marked yes, no or ask client."
],
"output_schema": {
"type": "object",
"properties": {
"facts": {
"type": "array",
"items": {
"type": "object",
"properties": {
"number": {
"type": "integer"
},
"fact": {
"type": "string"
},
"proof": {
"type": "string"
},
"true_as_of": {
"type": [
"string",
"null"
]
},
"publishable": {
"type": "string",
"enum": [
"yes",
"no",
"ask client"
]
}
},
"required": [
"number",
"fact",
"proof",
"true_as_of",
"publishable"
],
"additionalProperties": false
}
}
},
"required": [
"facts"
],
"additionalProperties": false
}
}Output schema (JSON Schema)
Ready for strict structured-output modes: every field required, unknown values nullable. In a chat, paste it after the prompt with "Return only JSON that matches this schema." How output schemas work.
{
"type": "object",
"properties": {
"facts": {
"type": "array",
"items": {
"type": "object",
"properties": {
"number": {
"type": "integer"
},
"fact": {
"type": "string"
},
"proof": {
"type": "string"
},
"true_as_of": {
"type": [
"string",
"null"
]
},
"publishable": {
"type": "string",
"enum": [
"yes",
"no",
"ask client"
]
}
},
"required": [
"number",
"fact",
"proof",
"true_as_of",
"publishable"
],
"additionalProperties": false
}
}
},
"required": [
"facts"
],
"additionalProperties": false
}What good output looks like
- Every fact has a number, a name or a date.
- Every fact says how it could be proved.
- Client names and results are marked yes, no or ask client.
What experts miss
- Ask the publishing question every time. Client names and results often need permission. Marking it now saves an awkward takedown later.
- Let the interviewer push for numbers. You say "lots of clients". A good interviewer asks how many, since when, and how you know.
- Record the date each fact was true. Proof without a date can't be updated, so it quietly goes wrong.
Why it works
Most small businesses have more proof than their site shows. It sits in invoices, inboxes and memory. An interviewer that asks for numbers, dates and sources turns vague claims into facts you can publish and defend.
How to adapt it
- Run it with a colleague answering, then compare the two lists.
- Add your industry's credentials, such as licences or insurance.
Mistakes to avoid
- Letting the AI fill in an answer you hesitated over.
- Publishing a client name marked "ask client".
Go deeper
The list feeds the fact list prompt and the case study prompt. Trust signals explains which facts matter most to buyers. This prompt is part of the plan stage. It follows the AI website design prompts guide and the six steps in how to use AI to design a website.