# How questions are picked

URL: /docs/research/how-questions-are-picked

Most feedback tools send every customer the same fixed form. Orvelt does not. You tell it what you want to learn, it keeps a library of questions for the project, and each customer gets a short set of up to three questions chosen for them. Every answer is scored, and the scores decide which questions still need answers, so the next customer is asked those.

```mermaid
flowchart LR
  accTitle: The Orvelt question loop
  accDescr: Goals add related questions to the project's question library. Each customer answers up to three questions picked from it. Every answer is scored, which moves each question's Auto priority, and the priority re-ranks the library for the next customer.
  goals["Goals"] --> library["Questions"]
  library -->|Up to 3| answers["Answers"]
  answers -->|Scored| priority["Auto priority"]:::result
  priority -.->|Re-ranks| library
```

## Goals and the question library [#goals-and-the-question-library]

### Goals say what to learn [#goals-say-what-to-learn]

A goal is one thing your team needs to learn, such as "Learn why first-time lunch customers do not return within 30 days." Set a goal and related questions are added to the project's library. Every project keeps at least one goal, and your plan sets how many it can have. New projects also start with a default set of questions, and you can [add or edit questions yourself](/docs/research/questions/create) in Orvelt Business or [with the API](/docs/developers/questions/create).

### Tags say what an answer reveals [#tags-say-what-an-answer-reveals]

Every question carries one or more of 8 tags: `Trust`, `Reliability`, `Experience`, `Safety`, `Personalization`, `Fairness`, `Identity`, and `Outcome`. A tag names what the answer can tell you about the customer, for example whether they trust the business. Tags drive both the [Auto priority](#how-auto-priority-learns) of a question and the project's Research Radar.

## How each customer's questions are picked [#how-each-customers-questions-are-picked]

Orvelt assembles a question set when a customer starts a response. It never writes new questions while they answer.

```mermaid
flowchart LR
  accTitle: How Orvelt picks a customer's questions
  accDescr: Active questions are filtered to those the customer has not answered, and up to three are picked, favoring questions with higher priority and fewer answers. The set stays fixed for the response.
  active["Active questions"] -->|Not yet answered| candidates["Candidates"]
  candidates -->|Picked| set["Up to 3 questions"]:::result
```

### The selection steps [#the-selection-steps]

1. **Plan cap.** Active questions are ranked by priority, and only as many as your plan allows are eligible.
2. **Already answered.** Questions this customer answered in an earlier response are skipped. Answering any wording version counts as answering the question in that project; publishing another version does not ask it again.
3. **Answer allowance.** Public and private questions share the organization's remaining monthly answers. When fewer answers remain, questions with higher priority and fewer answers are favored. No fresh questions are offered once the allowance is exhausted; saved answers can still be finished.
4. **Pick.** Up to three questions are picked. Questions with a higher priority and fewer answers so far are more likely to be picked, and every eligible question keeps getting some answers.

A customer may get fewer than three, or none, when fewer are eligible. The set stays fixed while they answer: editing an answer does not draw a new set, and changes to your library do not rewrite a response in progress.

## How Auto priority learns [#how-auto-priority-learns]

Priority is a whole number from 1 to 100. With Auto on, Orvelt adjusts it after every scored answer, based on how much the question still has to teach you.

```mermaid
flowchart LR
  accTitle: How an answer moves Auto priority
  accDescr: Orvelt scores each new answer against the question's tags, works out how much the question still needs answers, and moves its priority a small step in that direction.
  answer["New answer"] -->|Scored| need["Still needed?"]
  need -->|Small step| priority["New priority"]:::result
```

### What moves priority [#what-moves-priority]

- **Up:** answers are still revealing something about the question's tags, and customers do not yet agree.
- **Down:** answers rarely say anything about the tags, or customers have answered consistently enough that more answers add little.

Recent answers count more than old ones. Each answer moves the priority only a small, bounded step, so a single unusual answer cannot swing it, and a retried score never moves it twice. When the value changes, Orvelt sends a [`question.updated` webhook event](/docs/developers/feedback/webhooks).

## What you control [#what-you-control]

These settings shape the loop. Change them in Orvelt Business with [Create a question](/docs/research/questions/create) and [Priority and Auto priority](/docs/research/priority), or through the API with [Create questions](/docs/developers/questions/create) and [Auto priority](/docs/developers/questions/priority).

### Settings [#settings]

| Setting | Effect |
| --- | --- |
| Goals | Decide which related questions join the library. |
| Tags | Decide what each answer is scored against, and so what moves Auto priority. |
| Status | Only `active` questions can be picked. Archived questions stop being picked and stop moving. |
| Auto priority | On: Orvelt moves the priority. Off: the priority stays where you set it. Switching keeps the current value. |
| Priority | Your starting or fixed value. A higher value raises the chance of being picked; it never guarantees it, and never overrides eligibility, privacy, or plan limits. |
| Answer visibility | Controls who can read an answer. Public and private answers each count against the same monthly allowance. |