Question sets

What the panel is asked is what gets calculated: every rate, split and issue in a report comes from these questions. Start from a built-in set, change it, and export it to share or keep. Sets are saved in this browser.

Import a set

Draft a set from a description

What the site sells or says, and what you want to learn For

Ask a model

Checking for a connected model…

Use any AI chat

Copy the prompt into ChatGPT, Claude or any other chat, then paste its answer here. Without a description, the prompt leaves a gap for you to fill in.

The chat's answer

The prompt, exactly as the model gets it

You write configuration files for Jevsona, a tool in which simulated readers judge a web page. You answer with one JSON object and nothing else.

Write the questions a simulated reader panel answers about a commercial page (a product, service or landing page) of this site:

[Describe your site here: what it offers, and who visits it]

Each question is asked to every reader after they read the page. Every figure in the report comes from these questions:
the good answers of a single-choice question make a rate, and the answers listed under "issue" become findings.
Return a question set in exactly this JSON shape (this example is for a different site; do not copy its content):

{
  "format": "jevsona/question-set",
  "version": 1,
  "name": "Knife shop product page",
  "pageType": "commercial",
  "questions": [
    {
      "id": "offer_comprehension",
      "kind": "single",
      "metric": "offer_comprehension_rate",
      "prompt": "After reading this page, how clearly do you understand what is being sold and for whom?",
      "options": [
        "CLEAR",
        "MOSTLY_CLEAR",
        "PARTIALLY_CLEAR",
        "CONFUSING"
      ],
      "positiveOptions": [
        "CLEAR",
        "MOSTLY_CLEAR"
      ],
      "issue": {
        "type": "COMPREHENSION_FAILURE",
        "options": [
          "CONFUSING"
        ]
      },
      "enabled": true
    },
    {
      "id": "trust_gaps",
      "kind": "multi",
      "metric": null,
      "prompt": "Which of these stop you from trusting this shop?",
      "options": [
        "NO_REVIEWS",
        "UNCLEAR_RETURNS",
        "NO_CONTACT_DETAILS",
        "NO_MAJOR_TRUST_GAP"
      ],
      "positiveOptions": [],
      "maxSelections": 3,
      "noneOption": "NO_MAJOR_TRUST_GAP",
      "issue": {
        "type": "TRUST_GAP",
        "options": [
          "NO_REVIEWS",
          "UNCLEAR_RETURNS",
          "NO_CONTACT_DETAILS"
        ]
      },
      "enabled": true
    }
  ]
}

Rules:
- "format" is "jevsona/question-set", "version" is 1, "pageType" is "commercial". Every question has "enabled": true.
- 10 to 18 questions. Ids are lower_case_with_underscores and unique.
- "prompt" speaks to the reader, one plain sentence of at least 8 characters.
- Options are CAPITALS_WITH_UNDERSCORES. Single-choice questions have 3 to 6 options, ordered best to worst.
- A single-choice question that should be a figure in the report has "metric": "<id>_rate" and at least one "positiveOptions"; otherwise "metric": null.
- A multi-select question ("kind": "multi") has "metric": null, "positiveOptions": [], "maxSelections" of 2 or 3, and a "none of these" answer
  that is both the last option and "noneOption".
- "issue" is optional: a type in CAPITALS_WITH_UNDERSCORES and the options that count as that problem. Only name bad answers.
- Cover: understanding the offer, relevance to the reader, trust and what undermines it, what information is missing,
  objections, whether the tone and language fit, effort to read, and readiness to take the next step. Add what is specific to this site.
- Each reader's cost is one per single-choice question plus one per option of a multi-select (its none answer excluded); keep the total under 240.
- Answer with the JSON object only: no code fence, no comments, no text before or after.

Commercial pages

Commercial page built in21 questions · 61 judgments per reader · questions.v1

Editorial and blog

Editorial / blog built in15 questions · 25 judgments per reader · questions.v1

A run records the exact questions it was asked and their version. Two reports compare directly only when that version matches.