# Managing Prompts

Prompts are the questions GrackerAI sends to AI surfaces on your behalf. Each prompt is a real question your buyers might ask: "What's the best EDR for healthcare?" or "Splunk vs Sentinel for SOC teams?" Every monitor run records how AI surfaces respond to it.

Every prompt lives in a **topic** - a named theme like "Pricing", "Compliance", or "Alternatives" - and in at least one **monitor group** that schedules it. GrackerAI has two prompt pages with different purposes:

| Page | Where to Find It | What It Does |
|---|---|---|
| **Prompt Management** | Configuration → Prompt Management | Organise prompts into topics, add and import prompts, run Prompt Diagnosis |
| **Prompt Visibility** | Actions → Prompts Visibility | Per-prompt visibility metrics, grouped by topic, with row actions |

---

## Prompt Management

Go to **Configuration → Prompt Management**. The page is a list of your topics; open one to see the prompts filed under it.

![Prompt Management](/docs/images/prompt_management.png)

### Topics

Each topic row shows its colour, name, and prompt count. Click the row to expand it. The icons on the row:

| Icon | Action |
|---|---|
| **Upload** | Bulk import prompts into this topic |
| **Pencil** | Rename the topic |
| **Trash** | Delete the topic. Its prompts are not deleted - they move to **Miscellaneous** so you can re-file them later |

**Miscellaneous** is created automatically the first time you delete a topic that still holds prompts. It can't be deleted itself.

Topic names must be unique within a brand.

### Creating Topics

Open the **⋮ menu** in the top right:

- **Create topic** - name it and save
- **Suggest topics with Gracker Agent** - GrackerAI reads your brand context and proposes up to five topics with a one-line description each, skipping names you already have. Untick the ones you don't want, then save

Topics do double duty. They group prompts in analytics, and [Prompt Mining](/docs/using-gracker/working-with-prompts/prompt-mining) searches each topic's name to find real buyer questions. A specific name ("SSO pricing for mid-market") mines far better than a vague one ("Pricing").

Cybersecurity brands also get system topics for CVE and news tracking. GrackerAI creates and fills those itself; they appear in the list but are not offered to Prompt Mining.

### Adding a Prompt

Expand a topic and click **Add prompt to <topic>**:

| Field | Notes |
|---|---|
| **Topic** | Pre-filled with the topic you opened; change it if needed |
| **Prompt** | The question, exactly as a real user would type it. 3–2000 characters |
| **Monitor group** | Required. A prompt with no group is never tracked |
| **Intent** | Leave on **Auto-detect** to let the classifier decide, or pick one to set it manually |

GrackerAI rejects a prompt whose text matches an active prompt for this brand, ignoring case. Clean duplicates keep analytics honest.

### Bulk Import

Click the **Upload** icon on a topic row. Paste one prompt per line into the box - blank lines are ignored. To set an intent on a row, add it after a Tab:

```
best crm for startups	transactional
top crm alternatives 2026	comparison
```

Pick a monitor group, then import. Up to 1,000 rows per import. Rows that fail validation are listed by row number with the reason; fix and re-import only those.

For a faster way to fill a topic with real buyer questions, use [Prompt Mining](/docs/using-gracker/working-with-prompts/prompt-mining) - every mined prompt lands in its topic with volume and intent already attached.

### Removing a Prompt

Click the **Trash** icon on a prompt row and confirm. The prompt is archived, pulled out of every monitor group it was in, and stops being tracked. A monitor group left with no prompts is archived too.

There is no inline edit on this page. To change a prompt's wording, remove it and add the corrected version.

### Searching and Filtering

Once you have topics, a search box and an **Intent** filter appear above the list. Both filter the prompt rows inside every topic; topics with no matching prompt are hidden while a filter is active.

### Fan-out Queries

A prompt that has a finished [Visibility Diagnosis](/docs/using-gracker/working-with-prompts/visibility-diagnosis) shows a chevron. Click it to see the **fan-out queries** - the smaller questions an AI assistant breaks the prompt into when answering it.

### Diagnosing Your Prompt Set

The **⋮ menu → Prompt Diagnosis Options** checks every active prompt at once and groups them by what to do next - vague, off-topic, non-commercial, low demand, or duplicated - plus suggestions for the intents you're thin on. See [Prompt Diagnosis](/docs/using-gracker/working-with-prompts/prompt-diagnosis).

---

## Intent Tags

Every prompt is tagged with one of eight intent categories. GrackerAI assigns this automatically when the prompt is created, or you set it yourself on the add form.

| Intent | What the Buyer Is Doing |
|---|---|
| **Informational** | Learning about a topic, category, or problem |
| **Comparison** | Evaluating two or more specific tools side by side |
| **Alternatives** | Looking for options to replace a tool they already use |
| **How-to** | Trying to accomplish a specific task |
| **Troubleshooting** | Solving a problem they're already experiencing |
| **Recommendation** | Asking for a direct "what should I use" answer |
| **Pricing** | Evaluating cost before a purchase decision |
| **Transactional** | Ready to buy or sign up |

Along with intent, auto-classification also records:
- **Funnel Stage** - Awareness, Consideration, or Decision
- **Persona Signal** - Developer, Marketer, Executive, or End User
- **Topic tags** - a few short keywords describing what the prompt is about

Classification runs the moment a prompt is saved. If the classifier is slow, the prompt is saved with a placeholder intent and classified in the background shortly after.

When you set intent manually, auto-classification is skipped for that prompt and the intent is treated as final.

---

## Prompt Visibility

Go to **Actions → Prompts Visibility** to see how each prompt performs across every tracked answer engine.

![Prompts Insights](/docs/images/prompt_visibilityy.png)

### The Table

Rows are grouped by **topic**. Each topic row shows the aggregate across its prompts and expands to the individual prompts inside it. Sorted by Score, highest first, by default; click any column header to re-sort.

| Column | What It Measures |
|---|---|
| **Score** | Visibility score for this prompt across all runs in the window |
| **Presence Rate** | Percentage of runs where your brand was mentioned at least once |
| **Sentiment** | Average sentiment across all responses for this prompt |
| **Runs** | Number of times this prompt has been run |
| **Last Seen** | When this prompt last produced a response |

The date window and monitor group come from the scope picker in the top bar, the same as every other analytics page.

### Filters

- **Search** - filter by prompt text
- **Tags** - show only prompts carrying a workspace tag
- **Topics** - show a single topic, or **Uncategorized** for prompts with no topic

### Row Actions

**Click any prompt** to open the Prompt Details drawer: performance metrics, visibility trend, competitors named in this prompt's answers, and the recent responses.

The **⋮ menu** on each row:

| Action | What It Does |
|---|---|
| **Generate Content** | Opens the content generator pre-loaded with this prompt - see [Articles](/docs/using-gracker/content/articles) |
| **Add task** | Sends the prompt and its numbers to [Tasks](/docs/using-gracker/tasks) |
| **Move to topic** | Re-file the prompt under a different topic |
| **Diagnose Visibility** | Finds out why this prompt isn't citing you. Once a report exists the menu offers **Diagnosis report** and **Diagnose again** - see [Visibility Diagnosis](/docs/using-gracker/working-with-prompts/visibility-diagnosis) |

The **Manage Prompts** button in the header jumps to Prompt Management.

---

## Writing Effective Prompts

### Use Full Natural-Language Questions

AI users type full questions, not keywords. Write prompts the same way.

| Weak | Strong |
|---|---|
| `Best EDR healthcare` | `What's the best EDR for a 500-person healthcare SaaS that needs HIPAA compliance and runs on AWS?` |
| `Datadog alternatives` | `What are cheaper alternatives to Datadog for a Series B startup with 50 engineers and a Kubernetes-heavy stack?` |
| `CRM B2B SaaS` | `What's the best CRM for a product-led B2B SaaS company with under 100 employees?` |

### Include Buyer Context

- **Company size** - "for a 200-person company", "enterprise teams"
- **Industry** - "for healthcare", "for fintech"
- **Tech stack** - "on AWS", "with Kubernetes"
- **Compliance requirements** - "SOC 2 ready", "HIPAA compliant", "FedRAMP authorized"
- **Urgency signal** - "after a phishing attack", "before SOC 2 audit"

### Keep Brand Names Out

A prompt that names your brand or a competitor measures whether people look you up, not whether AI recommends you. Keep the library at the category level and let the responses tell you who gets named.

---

## Next Steps

- [Set up monitor groups](/docs/using-gracker/working-with-monitors/monitors): assign prompts to a schedule and surface selection
- [Read responses](/docs/using-gracker/responses-and-citations/responses): see exactly what AI surfaces answered for each prompt

- [Use Prompt Mining](/docs/using-gracker/working-with-prompts/prompt-mining): fill topics with real buyer questions from GSC, Bing, and search
- [Diagnose your prompt set](/docs/using-gracker/working-with-prompts/prompt-diagnosis): find vague, off-topic, and duplicate prompts in one pass