A LINE bot research toggle transforms vague summaries into cited reports using Gemini API grounding

The quick summary button originally served as a TL;DR option, but users frequently requested "the actual report with references." Integrating Google Search Grounding via the Gemini API upgraded these responses from three-line overviews into properly cited, structured briefs that enhance credibility. Below is the exact prompt used, along with implementation notes for connecting it to your own LINE webhook.

The prompt (ready to copy)

You are a research analyst. The user provides a URL or topic.
1. Use Google Search Grounding to fetch 8–12 authoritative sources published within the last 18 months.
2. Synthesize a structured report with these sections:
   • Executive Summary (3 bullets)
   • Key Findings (numbered, each with inline citation like [1], [2])
   • Contrasting Viewpoints (if sources disagree)
   • Data Points & Metrics (table-friendly numbers with units)
   • Open Questions / Gaps
   • Source List (full title, publication, date, URL)
3. Tone: professional, neutral, no fluff.
4. Output language: match the user's input language.
5. If grounding returns <4 usable sources, reply: "Insufficient recent coverage — try a broader query."

Why this approach succeeds

The prompt enforces clear constraints to prevent inaccuracies:

• Source requirements ("8–12 sources, last 18 months") block outdated or low-quality references.
• Structured output with numbered sections ensures consistent formatting, letting the LINE flex message render each as a collapsible card.
• Citation mapping using [n] tags aligns directly with the Source List, allowing users to tap citations for full references.
• Coverage check ("insufficient coverage" response) prevents the model from generating unreliable results for niche topics.

Connecting to LINE’s webhook (Node.js example)

app.post('/webhook', line.middleware(config), async (req, res) => {
  await Promise.all(req.body.events.map(handleEvent));
  res.sendStatus(200);
});

async function handleEvent(event) {
  if (event.type !== 'message' || event.message.type !== 'text') return;
  const text = event.message.text.trim();

  if (text.startsWith('research:')) {
    const query = text.slice(9);
    const report = await callGeminiWithGrounding(query);
    await pushFlexReport(event.source.userId, report);
  }
}

The callGeminiWithGrounding function calls generativelanguage.googleapis.com/v1beta/models/gemini-1.5-pro:generateContent with the prompt in systemInstruction and tools: [{ googleSearchRetrieval: {} }]. The response’s groundingMetadata.groundingChunks provides citation URLs, which map to the [n] markers before constructing the flex message JSON.

Cost and performance considerations

• Token usage: ~2,200 input / 1,800 output tokens per report (Gemini 1.5 Pro).
• Latency: 6–9 seconds end-to-end, with grounding adding ~3 seconds.
• Cost: ~$0.018 per report at current pricing, making it viable for a freemium bot tier.

Common implementation note

Some grounding results include paywalled PDFs. These are filtered out unless the domain is arxiv.org or pubmed.ncbi.nlm.nih.gov, ensuring the "tap to open" feature remains functional.

Using this setup converts user queries into verifiable reports, replacing guesswork with structured, cited analysis.