> ## Documentation Index
> Fetch the complete documentation index at: https://tryprefer.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# How does Prefer run prompts and collect AI answers?

> Prefer runs your prompts in tracking runs, on a schedule or on demand, and stores each engine's full answer, brands named, cited URLs and sentiment.

Prefer runs your tracked prompts in tracking runs: one full pass that sends every enabled prompt to every enabled engine, on a schedule or when you click Run tracking now. For each answer Prefer stores the full text, every brand named in order, every cited URL and the sentiment of each mention.

*Last updated 3 October 2026*

## The unit is the tracking run

Prefer collects AI answers in tracking runs. A run is one full pass: every enabled prompt goes to every enabled engine, in each country you track it in. Runs happen on a schedule on every paid plan. You can also start one yourself with Run tracking now in the top bar. A first run typically finishes in a few minutes.

**A run is the unit of time in Prefer: every delta compares the latest run with the previous one, not today with yesterday.** Engines can shift within hours. So if you run tracking twice in one day, you see two deltas, not one blended daily figure.

## What happens to each answer

For each prompt and engine, Prefer waits for the full answer and then works through four stages:

1. **Store.** The complete answer text is saved, so you can read it later in View all chats.
2. **Extract.** Every brand named is recorded in order of appearance, and every cited URL is captured. Citations are de-duplicated per run, prompt, engine and URL.
3. **Score sentiment.** Each brand mention gets a polarity: positive, neutral or negative.
4. **Roll up.** The signals become Visibility Score, Share of Voice, Citation Share, Avg Position and Sentiment, per engine and blended. Gaps against the leader feed the Action Center.

Prefer also records the fan-outs. Engines often split one question into several searches before they answer. The prompt detail view shows those prompt fan-outs with citation badges, so you can see which sub-searches pulled in which sources.

## Where and how it asks

**In 20 major markets Prefer runs prompts with precise Google geolocation; in every other country it states the market in the question itself.** Prefer Enterprise tracks every other AI model you need, such as Claude, Microsoft Copilot, Grok, Meta AI, DeepSeek and Qwen, the same way. The docs do not describe the collection method for the five core engines beyond what is stored, so this page does not either.

| | Free audit | Ongoing tracking |
| - | - | - |
| How often | Once | On a schedule, plus manual runs |
| Prompts | About 30 | Your plan's allowance |
| Engines | 3 | All 5, or the ones you enable |
| Output | A report and starter action plan | Trends, deltas, alerts and the Action Center |

## Why runs matter for reading results

AI answers vary between asks. In [Prefer's September 2026 study](https://tryprefer.com/data/ai-answer-study/), two runs of the same question repeated 96% of cited sites on Perplexity but 35% on Gemini. That is why Prefer stores every run rather than one snapshot, and why a trend across runs tells you more than any single answer.

To see one pass on your own brand, [run the free AI visibility audit](https://app.tryprefer.com/audit).

## Frequently asked questions

<AccordionGroup>
  <Accordion title="How often does Prefer run my prompts?">
    Prefer runs your prompts on a schedule on every paid plan, and you can start a run yourself at any time with Run tracking now in the top bar. Prefer does not publish a fixed run frequency, so read your numbers by run: every delta compares the latest run with the one before.
  </Accordion>

  <Accordion title="Does Prefer use the engines' APIs or a browser?">
    Prefer does not publish its collection method. For the five engines on every plan and the extra models on Enterprise, the docs describe what Prefer stores from each answer rather than the collection method, so we do not claim one here.
  </Accordion>

  <Accordion title="Can I read the exact answer Prefer collected?">
    Yes. In Prefer, open any prompt in Monitor, then Prompts, and click View all chats. You see the exact answer each engine returned, with the brands named, the sources cited and the sentiment.
  </Accordion>
</AccordionGroup>

## Also asked as

* How does Prefer collect AI answers?
* What is a tracking run in Prefer?
* Can I run Prefer tracking manually?

## Sources

* [Prefer docs: How Prefer works (captured 2026-10-03)](/docs/get-started/how-prefer-works)
* [Prefer docs: AI answer engines Prefer measures across plans (captured 2026-10-03)](/docs/concepts/engines)
* [Prefer docs: Quickstart, How Prefer works, Sources and citations (captured 2026-10-03)](/docs/get-started/quickstart)
* [Prefer: Answer Engine Insights; Prefer docs: Prompts console](https://tryprefer.com/platform/answer-engine-insights/)
* [Prefer docs: How Prefer works; Prefer pricing FAQ](https://tryprefer.com/pricing/)
* [Prefer AI answer study, September 2026](https://tryprefer.com/data/ai-answer-study/)

## Related

* [What counts as a tracked prompt in Prefer?](/docs/help/what-is-a-tracked-prompt-in-prefer)
* [Can I read the full AI answer in Prefer?](/docs/help/does-prefer-show-the-full-ai-answer)
* [What metrics does Prefer report?](/docs/help/what-metrics-does-prefer-report)
* [Do AI answers change if you ask the same question twice?](https://tryprefer.com/learn/answers/do-ai-answers-change-when-asked-twice/)
* [Answer Engine Insights, the monitoring module](https://tryprefer.com/platform/answer-engine-insights/)
* [Query fan-out, defined](https://tryprefer.com/learn/glossary/query-fan-out/)
* [Do AI answers change when asked twice?](https://tryprefer.com/learn/answers/do-ai-answers-change-when-asked-twice/)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.