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

# Agent Chat: 20 questions worth asking

> A prompt library for Agent Chat, grouped by job: diagnosing drops, watching competitors, planning content, and reporting.

This is a copy-paste question library for Agent Chat, grouped by job: diagnosing drops, watching competitors, planning content, and reporting. Most questions run on data your monitors already collect: tracked prompts and responses, visibility and sentiment, mentions, citations, and competitor detection. A few need extra setup: crawler logs, visitor analytics, or a stored content gap analysis. Copy them as written, or adjust the date ranges and names. Two habits make all of them work better: pin a date range, and ask for the evidence behind the conclusion. See [Asking Agent Chat](/docs/academy/agent-chat) for why.

### Diagnose a drop

When a number moved and you need to know why, make the agent isolate where the change lives before it explains it.

1. "Compare my visibility for the last 7 days against the 7 days before. Which models and prompts drove the change? Show the responses behind the biggest movers."
2. "Find the prompts where my brand appeared last month but is missing this month. Inspect two of those responses and tell me who replaced me."
3. "Did my sentiment change on any model in the last 14 days? Quote the negative responses and identify repeated criticisms."
4. "Did my citation rank, the position of my own domain inside answers, get worse in the last 7 days? Which of my URLs lost citation count versus the 7 days before, and which other domains gained share?"
5. "Check whether AI crawler request volume on my site changed in the last 14 days versus the 14 days before. Which paths fell out of the top crawled pages?"

### Competitor watch

Your competitor set is defined at the project level, so the agent can compare you against the same brands your dashboard uses.

6. "Which competitor gained the most mentions in the last 7 days, and on which prompts? Distinguish prompts where they outrank me from prompts where I'm absent."
7. "Which cited pages mention my strongest competitor but not my brand? Group them by content type."
8. "Build a competitor heatmap summary: per model, who owns my tracked prompts? Where am I strongest and weakest?"
9. "How do AI answers frame me versus my main competitor on comparison prompts? Quote short excerpts from two responses each."
10. "Find listicles cited in AI answers that include my competitors but not me, ranked by how many prompts they influence."

### Content planning

These pull from content gap analysis, query fanouts, and citations, so the output is grounded in what AI actually searched for.

11. "List my 5 weakest content-coverage prompts and the recommendations attached to them. Which are quick wins?"
12. "Collect the query fanouts behind my tracked prompts and cluster them by intent. Which clusters does my site not answer?"
13. "Which of my pages earn AI citations, and what do they have in common? Contrast them with two high-traffic or high-crawl pages that have no citations in this period."
14. "Find Reddit threads and YouTube videos cited in answers to my prompts. Where could a genuinely helpful contribution earn my brand a mention?"
15. "For my prompt topics, which competitor domains are cited most, and which of those topics have the weakest content-coverage scores?"

### Reporting

The agent can aggregate a period, render charts, and hand you the narrative, which turns report prep into a review job.

16. "Build an executive briefing for the last 7 days: visibility, sentiment, self-citation frequency, strongest competitor, and AI-referred visits. One chart, three findings, three actions."
17. "Chart my daily visibility per model for the last 30 days and annotate the days with notable moves."
18. "Summarize what changed on my action board this month: what was completed, what's still open at high severity?"
19. "How many AI-referred visits did I get in the last 30 days, from which assistants, landing on which pages?"
20. "Draft a monthly GEO summary I can paste into Slack: wins, losses, and the one thing to fix next month, each backed by a number."

### What happens next

When a question becomes routine, stop asking it manually: use **Scheduled** in the chat sidebar to set it to recur, with delivery to Slack if your team lives there. The agent runs it on cadence and records each result, notifying you when the answer changed.