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

# Summaries & Opinions

> How the AI Analyst handles summary and recommendation requests using prior thread context

When you ask the AI Analyst to summarize findings or give a recommendation after running an analysis, it uses **Summary or Opinion** mode — responding directly from thread context without running new SQL.

***

## When This Workflow Is Used

This workflow handles **follow-up requests** — questions that refer to results already in the thread rather than asking for new data.

### Summary triggers

```
Summarize the findings
Summarize this for leadership
Summarize that in three bullets
Rewrite that in plain English
Explain the key takeaways
Write that as a haiku
```

### Opinion triggers

```
What should we do next?
What is your recommendation?
What would you focus on?
What's the highest-leverage action here?
What does this mean for our strategy?
```

These requests share key characteristics:

* They refer to **prior results in the same thread**, not new data
* They ask what the data **means** or what to **do**, not for fresh computation
* They do not require running a new SQL query

<Info>
  Summary and Opinion requests require prior analysis in the same thread. If no analysis has been run yet, the AI Analyst will ask you to run one first.
</Info>

***

## How It Works

<Steps>
  <Step title="Classify Intent">
    The AI detects summary or opinion language and checks for prior analysis context in the thread.
  </Step>

  <Step title="Load Thread Context">
    The AI retrieves the retained analysis context from the current thread — results, findings, and prior SQL outputs.
  </Step>

  <Step title="Generate Response">
    The AI synthesizes a summary or recommendation directly from thread context. No SQL is executed.
  </Step>
</Steps>

<Note>
  If the classifier is not confident enough that this is a summary or opinion request, it falls back to Standard Analysis and runs fresh SQL instead.
</Note>

***

## What You'll See

Summary and Opinion responses are fast — no SQL runs, so there's no table identification or query execution step.

| Component             | Description                                                     |
| --------------------- | --------------------------------------------------------------- |
| **Response**          | Plain text summary or recommendation grounded in thread context |
| **Suggested prompts** | Three follow-up questions to continue the analysis              |

Summary and Opinion responses do not include SQL, CSV, or charts.

***

## Summary vs. Opinion

|                  | Summary                                        | Opinion                                             |
| ---------------- | ---------------------------------------------- | --------------------------------------------------- |
| **What it does** | Reports what was found                         | Recommends what to do                               |
| **Trigger**      | "Summarize", "rewrite", "explain the takeaway" | "What should we do?", "What's your recommendation?" |
| **Output style** | Factual recap of findings                      | Judgment-based guidance                             |

<Tip>
  You can get both in one request: "Summarize the findings and tell me what to do next." The AI will summarize and add recommendations.
</Tip>

***

## Tips for Best Results

<AccordionGroup>
  <Accordion title="Run analysis first">
    Summary and Opinion need prior results to work from. Ask your data question first, then ask for a summary or recommendation in the same thread.
  </Accordion>

  <Accordion title="Stay in the same thread">
    Thread context does not carry across threads. Ask follow-ups in the same thread where the analysis ran.
  </Accordion>

  <Accordion title="Mention the bot in channels">
    In Slack channels, include `@SourceMedium` in your follow-up: `@SourceMedium summarize that for leadership.` In DMs, you can reply naturally without a mention.
  </Accordion>

  <Accordion title="Be specific about format">
    "Summarize in three bullets" or "give me a one-sentence takeaway" shapes the output. Generic requests like "summarize" also work but may return a longer response.
  </Accordion>

  <Accordion title="Ask for a fresh analysis if results are stale">
    Summary and Opinion use retained context, not live data. If the thread is old or you need updated numbers, run a new analysis first.
  </Accordion>
</AccordionGroup>

***

## Limitations

* **Requires prior context** — If no analysis has run in the thread, the AI will ask you to run one first
* **No fresh SQL** — These responses use retained thread context only; they do not re-query BigQuery
* **No charts or CSV** — Output is plain text only
* **Context depth** — Very long threads may not retain every detail; the AI works from a condensed summary of prior results
* **Confidence threshold** — Low-confidence summary or opinion classifications are rerouted to Standard Analysis, which runs fresh SQL instead

***

## Related

<CardGroup cols={2}>
  <Card title="Standard Analysis" icon="bolt" href="/ai-analyst/workflows/standard">
    How specific metric questions are handled with fresh SQL.
  </Card>

  <Card title="Deep Analysis" icon="brain" href="/ai-analyst/workflows/deep-analysis">
    How open-ended strategic questions trigger multi-perspective analysis.
  </Card>
</CardGroup>

<Card title="Thread continuity and follow-ups" icon="message-lines" href="/ai-analyst/thread-continuity-follow-ups">
  Learn how thread context works in DMs vs channels, and when to start a new thread.
</Card>