AI Strategic Insights

Overview

The AI Insights panel transforms raw sentiment data into actionable strategic recommendations. It analyzes emotion distributions, crisis levels, and volume trends to generate professional reports tailored for PR teams, communications managers, and decision-makers.

How It Works

Select a Topic

Open any topic from the dashboard. The AI Insights panel appears on the right side of the Topic Detail view.

Generate Report

Click “Generate Report” to trigger AI analysis. The system automatically:

Pulls current sentiment score
Analyzes emotion breakdown
Identifies crisis level and volume trends
Streams a strategic analysis in real-time

Review Recommendations

The generated report includes:

Situation Assessment: Current state summary
Key Concerns Identified: Main issues driving sentiment
Recommended Actions: 4-5 specific, implementable steps
Opportunities: Positive angles to leverage
Risk Factors to Monitor: Escalation triggers

Ask Follow-Up Questions

Use the interactive chat to dive deeper:

“What should we prioritize first?”
“How can we address the anger?”
“What are the quick wins?”

Tiered AI Provider Fallback

SENTi-radar uses a smart fallback system to ensure reports are always generated, even without API keys:

Tier 1: OpenAI GPT-4o mini
Tier 2: Google Gemini 2.0 Flash
Tier 3: Groq Llama (Deprecated)
Tier 4: Local Fallback (Always Works)

Primary provider (if VITE_OPENAI_API_KEY configured)

// Source: AIInsightsPanel.tsx (lines 71-107)
const resp = await fetch('https://api.openai.com/v1/chat/completions', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    Authorization: `Bearer ${VITE_OPENAI_API_KEY}`
  },
  body: JSON.stringify({
    model: 'gpt-4o-mini',
    messages: chatMessages,
    stream: true,
    max_tokens: 1000,
    temperature: 0.7
  })
});

Advantages:

Most reliable and accurate
Best at nuanced analysis and actionable recommendations
Fast streaming responses

Cost: ~

0.15 per 1M input tokens,

0.60 per 1M output tokens

Secondary provider (if VITE_GEMINI_API_KEY configured)

// Source: AIInsightsPanel.tsx (lines 137-192)
const url = `https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:streamGenerateContent?alt=sse&key=${VITE_GEMINI_API_KEY}`;

const resp = await fetch(url, {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    contents: [{ role: 'user', parts: [{ text: fullPrompt }] }],
    generationConfig: {
      maxOutputTokens: 1000,
      temperature: 0.7
    }
  })
});

Advantages:

Free tier available (60 requests/minute)
Fast and cost-effective
Good at structured analysis

Cost: Free tier or $0.075 per 1M tokens

Tertiary provider (if VITE_GROQ_API_KEY configured)Uses Llama 3.3 70B for fast inference. This tier exists in the codebase but is being phased out in favor of Gemini.

Guaranteed fallback (no API required)Generates reports using template-based analysis:

// Source: AIInsightsPanel.tsx (lines 248-289)
function generateLocalInsight(
  topic: TopicCard,
  emotions: EmotionData[]
): string {
  const topEmo = emotions[0]?.emotion || 'surprise';
  const topPct = emotions[0]?.percentage || 40;
  const sentiment = topic.sentiment;
  const crisis = topic.crisisLevel;
  
  return `## Situation Assessment
  Public sentiment around **${topic.title}** is currently **${sentimentLabel}**, 
  with **${topEmo} (${topPct}%)** as the dominant emotion.
  
  ## Recommended Actions
  1. **Monitor and document** — track the ${topEmo} signals in real-time
  2. **Engage proactively** — address leading concerns with transparent communication
  3. **Segment your response** — tailor messaging to distinct emotional groups
  ...
  `;
}

Advantages:

Always available, zero dependencies
Instant generation (no API latency)
Privacy-friendly (no external calls)

Limitations:

Template-based, less contextual than LLMs
Generic recommendations

The system tries each tier in order. If OpenAI fails (quota, network), it automatically falls back to Gemini, then Groq, then local generation.

Report Structure

Every AI-generated report follows this professional format:

1. Situation Assessment

## Situation Assessment
Public sentiment around **AI Regulation Debate** is currently **divided**, 
with **fear (35%)** as the dominant emotion. Crisis level is **medium**, 
indicating elevated public concern. Volume has shifted **+42%**, suggesting 
rapidly escalating public engagement.

Purpose: Provides a 2-3 sentence snapshot of the current landscape.

2. Key Concerns Identified

## Key Concerns Identified
- **Fear dominance (35%)** — the primary emotional driver shaping narratives
- **Anger** as secondary emotion signals competing public reactions
- Rising volume indicates growing public interest that needs to be channeled
- Emotional polarization (fear 35%, anger 28%, surprise 18%) suggests fragmented audience segments

Purpose: Bullet-point list of the 3-4 main issues driving sentiment.

3. Recommended Actions

## Recommended Actions
**Monitor and document** — track the fear signals in real-time for early escalation warnings
**Engage proactively** — address the leading concerns with transparent, factual communication via official channels
**Segment your response** — tailor messaging to the distinct emotional groups (fear vs anger audiences)
**Establish rapid response protocols** — Proactive monitoring and engagement is advised.
**Measure sentiment shift** — set up hourly tracking alerts for #AIRegulation

Purpose: 4-5 specific, actionable steps. Each starts with an action verb and is implementable.

4. Opportunities

## Opportunities
- The **35% fear** signal can be redirected into constructive dialogue if addressed proactively
- Current volume spike (+42%) is an opportunity to own the narrative before it solidifies
- Building credibility now through transparent communication will pay dividends when sentiment stabilizes

Purpose: 2-3 positive angles or silver linings.

5. Risk Factors to Monitor

## Risk Factors to Monitor
- **Escalation trigger**: A new development could push sentiment toward crisis level
- **Narrative hijacking**: Third parties or media may amplify the anger signal to shift the story
- **Sentiment fatigue**: Prolonged high-volume discourse without resolution leads to disengagement or backlash

Purpose: 2-3 things that could make the situation worse.

Interactive Chat

After generating a report, expand the “Ask More About This Analysis” section to chat with the AI:

How to Use
Example Conversation
Technical Details

Click “Ask More About This Analysis” to expand the chat panel
Type a question or click a suggested prompt:
- “What should we prioritize first?”
- “How can we address the anger?”
- “What are the quick wins?”
Press Enter or click the Send button
The AI responds in context of the generated report

The chat maintains conversation history, so you can ask follow-up questions that reference previous answers.

// Source: AIInsightsPanel.tsx (lines 371-397)
const sendMessage = () => {
  const userMsg: Message = { 
    role: 'user', 
    content: inputValue.trim() 
  };
  setChatMessages((prev) => [...prev, userMsg]);
  
  streamInsights({
    topic,
    emotions,
    messages: [...chatMessages, userMsg],
    isChat: true,  // Flag for conversational mode
    onDelta: (chunk) => {
      assistantContent += chunk;
      setChatMessages((prev) => {
        const last = prev[prev.length - 1];
        if (last?.role === 'assistant') {
          return prev.map((m, i) => 
            i === prev.length - 1 
              ? { ...m, content: assistantContent } 
              : m
          );
        }
        return [...prev, { 
          role: 'assistant', 
          content: assistantContent 
        }];
      });
    }
  });
};

The chat uses the same tiered fallback as report generation (OpenAI → Gemini → Local).

System Prompts

The AI uses different prompts for reports vs. chat:

Report Generation Prompt

// Source: AIInsightsPanel.tsx (lines 37-67)
const systemPrompt = `You are a senior PR and communications strategist. 
Analyze the following public sentiment data and provide balanced, actionable recommendations.

Topic: ${topicName}
Hashtag: ${hashtag}
Overall Sentiment: ${sentiment}
Crisis Level: ${crisis}
Emotion Breakdown: ${emotionStr}
Volume Change: ${change}

Generate a professional analysis report with these sections:

## Situation Assessment
Brief overview of the current public sentiment landscape (2-3 sentences).

## Key Concerns Identified
- Main issues driving sentiment (3-4 bullet points)

## Recommended Actions
4-5 actionable recommendations. Each should:
- Start with a clear action verb
- Be specific and implementable
- Consider both immediate and long-term impact

## Opportunities
- 2-3 opportunities arising from this situation

## Risk Factors to Monitor
- 2-3 things that could escalate or shift the situation

Keep the tone professional, balanced, and constructive.`;

Chat Mode Prompt

// Source: AIInsightsPanel.tsx (lines 27-35)
const systemPrompt = `You are an AI analyzing public sentiment data.
Topic: ${topicName}
Hashtag: ${hashtag}
Overall Sentiment: ${sentiment}
Crisis Level: ${crisis}
Top Emotions: ${emotionStr}

Answer questions concisely and directly. No filler, no pleasantries. Just precise answers.`;

Chat mode uses a more conversational tone and focuses on direct answers rather than full reports.

Configuration

To enable AI insights, add at least one API key to your .env file:

# Recommended: OpenAI (best quality)
VITE_OPENAI_API_KEY=sk-proj-...

# Alternative: Gemini (free tier available)
VITE_GEMINI_API_KEY=AIza...

# Fallback: Groq (fast inference)
VITE_GROQ_API_KEY=gsk_...

No API keys? The platform still works! Local generation is always available and provides solid template-based analysis.

Data Source Indicator

The AI Insights panel displays which provider generated the report:

// Source: AIInsightsPanel.tsx (lines 414-418)
{dataSource && !isGenerating && (
  <span className="text-[9px] bg-green-500/10 text-green-600 px-2 py-0.5 rounded-full">
    ✓ {dataSource}
  </span>
)}

Possible values:

✓ GPT-4o mini
✓ Gemini
✓ Groq
✓ Local

Use Cases

Crisis Response Planning

Scenario: Food price crisis (48% anger, 22% fear) AI Recommendation:

## Recommended Actions
**Acknowledge public frustration** — Issue statement recognizing grocery cost burden
**Share concrete relief measures** — Announce price control investigations, subsidies, or relief programs
**Combat misinformation** — Address "corporate greed" narrative with supply chain transparency
**Engage directly with affected communities** — Town halls, social media Q&A sessions

Product Launch Optimization

Scenario: iPhone 18 launch (45% joy, 25% surprise, 15% anger) AI Recommendation:

## Opportunities
- Holographic display is generating viral "wow" moments — amplify with user-generated content campaigns
- Android users showing unusual interest (+300%) — targeted cross-platform marketing opportunity

## Key Concerns
- 15% anger around pricing, especially "no USB-C" meme — address in next press release or FAQ

Policy Communication

Scenario: AI regulation debate (35% fear, 28% anger) AI Recommendation:

## Recommended Actions
**Segment messaging by audience** — Tech workers (fear-driven) vs general public (anger-driven)
**Showcase international precedent** — EU regulation praised as template; position as learning from best practices
**Address job displacement directly** — Publish reskilling programs, transition support initiatives

Best Practices

Generate Reports Early

Don’t wait for a crisis. Generate reports for stable topics to establish baselines and track trends over time.

Use Chat for Specifics

The initial report is high-level. Use chat to drill down: “What’s the best quick win?” or “Which stakeholder group is most angry?”

Combine with Export

Export reports as PDF for stakeholder briefings or board presentations (Settings → Export Report).

Regenerate After Events

If sentiment shifts after a statement or announcement, click “Regenerate” to get updated recommendations.

Limitations

LLM hallucination risk: AI-generated recommendations are based on emotion percentages and keywords, not deep semantic understanding. Always cross-check with the raw Live Feed and Top Voices panels.No domain expertise: The AI doesn’t know your industry, brand history, or internal constraints. Treat recommendations as a starting point, not gospel.Static analysis: Reports are generated from a single snapshot. For evolving situations, regenerate every 6-12 hours.

Advanced: Supabase Edge Function (Deprecated)

The codebase includes a Supabase Edge Function integration for server-side LLM calls:

// Source: AIInsightsPanel.tsx (lines 194-246)
const url = `${VITE_SUPABASE_URL}/functions/v1/generate-insights`;
const resp = await fetch(url, {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    Authorization: `Bearer ${VITE_SUPABASE_PUBLISHABLE_KEY}`
  },
  body: JSON.stringify({ topic, emotions, sentiment, crisisLevel, messages })
});

This is now Tier 3 (after OpenAI and Gemini) and primarily used for legacy deployments. Most users should rely on direct OpenAI/Gemini calls.

Sentiment Analysis - Provides the data that fuels AI insights
Emotion Classification - Emotion breakdown powers the AI’s understanding
Crisis Detection - AI insights include crisis-specific recommendations

Get Started

Setup & Configuration

Core Features

Data Sources

Guides

AI Strategic Insights

Overview

How It Works

Tiered AI Provider Fallback

Report Structure

1. Situation Assessment

2. Key Concerns Identified

3. Recommended Actions

4. Opportunities

5. Risk Factors to Monitor

Interactive Chat

System Prompts

Report Generation Prompt

Chat Mode Prompt

Configuration

Data Source Indicator

Use Cases

Crisis Response Planning

Product Launch Optimization

Policy Communication

Best Practices

Generate Reports Early

Use Chat for Specifics

Combine with Export

Regenerate After Events

Limitations

Advanced: Supabase Edge Function (Deprecated)

Build docs developers (and LLMs) love

Get Started

Setup & Configuration

Core Features

Data Sources

Guides

​Overview

​How It Works

​Tiered AI Provider Fallback

​Report Structure

​1. Situation Assessment

​2. Key Concerns Identified

​3. Recommended Actions

​4. Opportunities

​5. Risk Factors to Monitor

​Interactive Chat

​System Prompts

​Report Generation Prompt

​Chat Mode Prompt

​Configuration

​Data Source Indicator

​Use Cases

​Crisis Response Planning

​Product Launch Optimization

​Policy Communication

​Best Practices

Generate Reports Early

Use Chat for Specifics

Combine with Export

Regenerate After Events

​Limitations

​Advanced: Supabase Edge Function (Deprecated)

​Related Features

Build docs developers (and LLMs) love

Overview

How It Works

Tiered AI Provider Fallback

Report Structure

1. Situation Assessment

2. Key Concerns Identified

3. Recommended Actions

4. Opportunities

5. Risk Factors to Monitor

Interactive Chat

System Prompts

Report Generation Prompt

Chat Mode Prompt

Configuration

Data Source Indicator

Use Cases

Crisis Response Planning

Product Launch Optimization

Policy Communication

Best Practices

Limitations

Advanced: Supabase Edge Function (Deprecated)

Related Features