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

# Managing prompts

> Choosing, adding and pruning the questions AI Tracker asks on your behalf.

A **prompt** is one question AI Tracker puts to the AI platforms on your
behalf, over and over, so you can watch how the answer changes.

Prompts are the entire measurement. Your visibility score is nothing more than
"how often did you appear in the answers to *these* questions" — so a set of
prompts nobody would really type produces a number that looks precise and means
nothing. This is the page worth spending time on.

## The Prompts page

Prompts are grouped by topic. For each one you get its status, how visible you
are in its answers, where you tend to be positioned, the sentiment of the
mentions, the market it runs in, and when it last ran.

Click any prompt to open its detail view: the full text of every answer, per
platform and per day, with the sources each answer cited and a screenshot of
the search result where one applies. When a score surprises you, this is where
the explanation is.

### Filtering and sorting

The toolbar filters on visibility, position, sentiment, mentions, location,
last-updated and tags, with `equals`, `contains`, `greater than` and `less
than` comparisons. Common uses:

* **Visibility `equals` 0** — the questions you are invisible for. Your work list.
* **Sentiment `less than` neutral** — where you appear but are described badly.
* **Tags `is empty`** — prompts that escaped your labelling scheme.

## Prompt statuses

| Status | Meaning |
| - | - |
| **Active** | Included in the daily runs. Only active prompts cost you anything or produce data. |
| **Suggested** | Generated by AI Tracker and waiting on your decision. Not running yet. |
| **Inactive** | Kept, with its history, but not running. |
| **Archived** | Retired. Read-only, kept for reference. |

Deactivating never deletes history. A prompt turned off in March still shows
its January and February results.

## The active prompt cap

A campaign can only have a limited number of **active** prompts at once. The
Prompts page shows where you stand — active count against the cap — and the
count is enforced when you activate.

<Warning>
  Prompts added past the cap are still saved, but stay **inactive** and will not
  run until you free a slot. Nothing is lost, but nothing new starts collecting
  data either. If you added prompts and they aren't producing results, this is
  almost always why.
</Warning>

Freeing a slot is a matter of deactivating a prompt you care less about. Being
forced to choose is the point: ten questions you genuinely compete on beat
fifty you don't.

## Adding prompts

Use **Add prompt** on the Prompts page. You can paste several at once, one per
line — numbered or bulleted lists are fine, the markers are stripped for you:

```
1. best analytics tools for startups
2. cheapest analytics platform with Xero sync
3. alternatives to Google Analytics for small teams
```

Each batch shares its settings:

<ParamField path="Topic" type="string" required>
  Which topic to file the prompts under. Matched against your existing topics
  without regard to case, and **created if nothing matches** — so check your
  spelling, or you'll end up with `Analytics` and `analytcs` side by side.
</ParamField>

<ParamField path="Location" type="string" default="us">
  The market the prompts are asked from. AI answers vary sharply by country;
  track the markets you sell in rather than defaulting everything to one.
</ParamField>

<ParamField path="Tags" type="string[]">
  Up to 8 tags, 24 characters each, for your own grouping and filtering.
  Lowercased when saved.
</ParamField>

**Limits.** 2,000 characters per prompt, 50 prompts per batch. Over-long
prompts are skipped and listed back to you rather than failing the rest of the
batch.

<Tip>
  Prompts can also be added programmatically — see
  [Create prompts](/api/endpoints#create-prompts) in the API, or just ask an AI
  client connected over [MCP](/api/mcp) to add them for you.
</Tip>

## Writing prompts that measure something

The generated starter set is a reasonable baseline, not a finished job. The
prompts that produce useful data tend to share a few traits.

<AccordionGroup>
  <Accordion title="Write what a customer would type, not what you'd like to rank for">
    `best project management tool for architects` is a real question. `acme
            project management software` is a brand lookup — you will appear in it
    every time, and it will tell you nothing.
  </Accordion>

  <Accordion title="Favour questions where you could plausibly lose">
    A prompt you always win and a prompt you never win are equally
    uninformative. The value sits in the contested middle, where the answer
    changes as your content does.
  </Accordion>

  <Accordion title="Include the qualifiers buyers actually use">
    Price, size, integration, industry, geography. `cheapest`, `for small
            teams`, `that integrates with Xero`, `in the UK`. These are how real
    questions are asked and they surface different competitors.
  </Accordion>

  <Accordion title="Cover the comparison and alternative questions">
    `alternatives to <competitor>` and `<competitor> vs <competitor>` are
    heavily used and are where a challenger brand can realistically break into
    an answer.
  </Accordion>

  <Accordion title="Keep them stable">
    Rewording a prompt restarts its trend. If you want to test a variation, add
    it as a second prompt and keep the original running.
  </Accordion>
</AccordionGroup>

## When a prompt isn't returning results

<AccordionGroup>
  <Accordion title="It has never run">
    Check the status. Suggested and inactive prompts never run. If it's active
    and newly added, results appear after the next daily run.
  </Accordion>

  <Accordion title="It runs on fewer platforms than expected">
    Platform coverage comes from your plan, then any narrowing you applied to
    the campaign or to that specific prompt. Starter tracks ChatGPT only.
  </Accordion>

  <Accordion title="It shows no visibility, but you're sure you're in the answer">
    Open the prompt and read the answer. If your brand appears under a name
    that isn't in your brand variants, add it in **Brand Hub** — detection only
    counts names it has been given. See
    [brand variants](/guides/onboarding#brand-variants).
  </Accordion>

  <Accordion title="One platform has a gap on one day">
    Normal. Each platform is asked once a day and retries on the next run if it
    fails, rather than hammering the same day.
  </Accordion>
</AccordionGroup>


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