#!/usr/bin/env bun /** * Generate a standalone HTML featurette report from destination_featurette data. * Usage: bun run generate-report.ts */ import { resolve } from 'path'; import { readFileSync } from 'fs'; const DATA_DIR = import.meta.dir; interface ScoresData { overall: { geo_score: number; visibility_tier: string }; decision_moments: Record; platform_scores: Record; zero_mention_queries: Array<{ query_id: string; query_text: string }>; competitor_analysis: Record; } interface QueryInput { metadata: { clusters: Record; total_queries: number; featurette: string; purpose: string }; queries: { discovery: Array<{ text: string; pattern: string }> }; } interface MentionedQuery { id: string; text: string; cluster: string; chatgpt: boolean; claude: boolean; perplexity: boolean; gemini: boolean; total: number; } // ── Load data ── const scores: ScoresData = JSON.parse(readFileSync(resolve(DATA_DIR, 'scores.json'), 'utf-8')); const queriesRaw: QueryInput = JSON.parse(readFileSync(resolve(DATA_DIR, '..', 'queries_destination_featurette.json'), 'utf-8')); const analysis = JSON.parse(readFileSync(resolve(DATA_DIR, 'analysis.json'), 'utf-8')); // ── Parse CSV to get per-query mentions ── const csvRaw = readFileSync(resolve(DATA_DIR, 'responses.csv'), 'utf-8'); const csvLines = csvRaw.trim().split('\n').slice(1); const mentionedQueries: MentionedQuery[] = []; const zeroMentionQueries: MentionedQuery[] = []; const queryPatternMap = new Map(); queriesRaw.queries.discovery.forEach((q, i) => { queryPatternMap.set(`Q${i + 1}`, q.pattern); }); csvLines.forEach(line => { const cols = line.split(','); const id = cols[0]; const text = cols[3]; const chatgpt = cols[4] === 'Yes'; const claude = cols[5] === 'Yes'; const perplexity = cols[6] === 'Yes'; const gemini = cols[7] === 'Yes'; const total = parseInt(cols[8], 10); const cluster = queryPatternMap.get(id) ?? 'unknown'; const record: MentionedQuery = { id, text, cluster, chatgpt, claude, perplexity, gemini, total }; if (total > 0) mentionedQueries.push(record); else zeroMentionQueries.push(record); }); const clusterNames: Record = { houston_origin: 'Houston Origin', austin_origin: 'Austin Origin', san_antonio_origin: 'San Antonio Origin', taste_profile: 'Taste Profile', competitive_substitution: 'Competitive Substitution', }; // Per-cluster aggregation const clusterStats: Record = {}; csvLines.forEach(line => { const cols = line.split(','); const id = cols[0]; const total = parseInt(cols[8], 10); const cluster = queryPatternMap.get(id) ?? 'unknown'; if (!clusterStats[cluster]) clusterStats[cluster] = { total: 0, mentions: 0, queries: [] }; clusterStats[cluster].total += total; if (total > 0) clusterStats[cluster].mentions += total; }); const geoScore = scores.overall.geo_score; const tier = scores.overall.visibility_tier; const discoveryScore = scores.decision_moments.discovery.score; const platforms = ['chatgpt', 'claude', 'perplexity', 'gemini'] as const; const platformLabels: Record = { chatgpt: 'ChatGPT', claude: 'Claude', perplexity: 'Perplexity', gemini: 'Gemini' }; // ── Build HTML ── const html = ` Hotel Lucine — Galveston Destination Featurette Report
Hotel Lucine

Galveston Destination
Featurette Report

Measuring AI visibility in destination-level queries from key drive markets

${geoScore} GEO Score

${tier}

Your property is effectively invisible to AI-powered hotel discovery — travelers using AI assistants for destination-level queries will not find your brand.

${queriesRaw.metadata.total_queries}Queries
4Platforms
${analysis.metadata.collection_count}Runs
5Clusters
${analysis.metadata.analysis_date}Date

Executive Summary

This featurette tests whether Hotel Lucine appears when travelers ask hotel-centric destination questions — queries where the "right" AI answer includes Galveston as a destination. It does not affect the main GEO Visibility Score (${scores.overall.geo_score}).

Discovery Score
${discoveryScore}
out of 100 — critical weakness
Queries With Any Mention
${mentionedQueries.length} / ${queriesRaw.metadata.total_queries}
only ${mentionedQueries.length} of ${queriesRaw.metadata.total_queries} queries had Lucine mentioned on any platform
Core Finding

Hotel Lucine is absent from destination-level AI queries. Travelers asking where to go for a weekend from Houston, Austin, or San Antonio do not see Lucine in AI responses. The one bright spot: Q14 ("Cool boutique hotel in Texas with music events and beachfront location") surfaced Lucine on all 4 platforms, proving the property can appear when the query language aligns precisely with its differentiators. The gap is not about Lucine's quality — it's about the property not being encoded in AI training data as a "Galveston answer" for origin-drive-time queries.

Cluster Performance

The 20 queries are organized into 5 strategic clusters. Performance varies significantly by cluster.

${['houston_origin', 'austin_origin', 'san_antonio_origin', 'taste_profile', 'competitive_substitution'].map(cluster => { const stats = clusterStats[cluster] ?? { total: 0, mentions: 0 }; const maxMentions = cluster === 'houston_origin' ? 16 : cluster === 'austin_origin' ? 16 : cluster === 'san_antonio_origin' ? 12 : cluster === 'taste_profile' ? 20 : 16; const pct = stats.total > 0 ? (stats.mentions / maxMentions * 100).toFixed(0) : 0; const color = Number(pct) === 0 ? 'var(--red)' : Number(pct) < 15 ? 'var(--accent)' : 'var(--green)'; const tagClass = cluster === 'houston_origin' ? 'tag-houston' : cluster === 'austin_origin' ? 'tag-austin' : cluster === 'san_antonio_origin' ? 'tag-sanantonio' : cluster === 'taste_profile' ? 'tag-taste' : 'tag-comp'; return `
${clusterNames[cluster] ?? cluster}
${stats.mentions}
mentions / ${maxMentions} possible
`; }).join('')}
Cluster Insight

Taste Profile is the only cluster with meaningful visibility (Q14: 4-platform mention). The origin-based clusters (Houston, Austin, San Antonio) are essentially dead zones — only 5 scattered mentions across 48 total query-platform opportunities. Competitive Substitution (queries explicitly seeking Austin alternatives) has just 2 Perplexity mentions — Lucine is not being surfaced as the "skip Austin" answer.

Platform Analysis

Where Hotel Lucine was mentioned across the 4 AI platforms (aggregated across ${analysis.metadata.collection_count} runs).

${platforms.map(p => { const ps = scores.platform_scores[p]; const color = ps.score === 0 ? 'var(--red)' : ps.score < 10 ? 'var(--accent)' : 'var(--green)'; return `
${platformLabels[p]}
${ps.score}
${ps.mentions} mention${ps.mentions !== 1 ? 's' : ''} / ${ps.total} queries
`; }).join('')}
Platform Finding

Gemini and Perplexity are the only platforms showing Lucine at all, and barely. ChatGPT mentioned Lucine on only 1 query across all runs. Claude mentioned on 2 queries. There are no sentiment data points because mentions were too few to analyze sentiment meaningfully. This is a training-data gap, not a content-quality gap — the platforms don't associate Lucine with "Galveston beach weekend" or "Texas drive destination" queries.

Query-Level Detail

All 20 queries with per-platform mention status. 12 queries had zero mentions across all platforms and all ${analysis.metadata.collection_count} runs.

${[...mentionedQueries, ...zeroMentionQueries].sort((a, b) => parseInt(a.id.slice(1)) - parseInt(b.id.slice(1))).map(q => { const tagClass = q.cluster === 'houston_origin' ? 'tag-houston' : q.cluster === 'austin_origin' ? 'tag-austin' : q.cluster === 'san_antonio_origin' ? 'tag-sanantonio' : q.cluster === 'taste_profile' ? 'tag-taste' : 'tag-comp'; const rowStyle = q.total === 0 ? 'opacity: 0.5;' : ''; return ` `; }).join('')}
ID Query Cluster ChatGPT Claude Perplexity Gemini Total
${q.id} ${q.text} ${clusterNames[q.cluster] ?? q.cluster} ${q.chatgpt ? '●' : '—'} ${q.claude ? '●' : '—'} ${q.perplexity ? '●' : '—'} ${q.gemini ? '●' : '—'} ${q.total}

Who Shows Up Instead

Competitor mentions detected in Lucine's zero-mention queries. When Lucine isn't surfacing, these hotels are.

${Object.entries(scores.competitor_analysis).map(([name, data]: [string, any]) => { const color = data.total_mentions === 0 ? 'var(--text-muted)' : data.total_mentions >= 2 ? 'var(--red)' : 'var(--accent)'; return `
${name}
${data.total_mentions}
total mentions · ${data.in_zero_mentions} in zero-mention queries
`; }).join('')}
Competitive Note

The Tremont House and Grand Galvez lead with 3 and 2 mentions respectively — both are historic Galveston hotels with stronger AI training-data presence. The Mansard House (a direct boutique competitor) has zero mentions, suggesting this is a Galveston-wide AI visibility problem, not just a Lucine problem.

Methodology

How this featurette was collected and scored.

Collection

${analysis.metadata.collection_count} independent runs across 4 AI platforms (ChatGPT, Claude, Perplexity, Gemini). Each run: 20 queries × 4 platforms = 80 API calls. Total: ${analysis.metadata.collection_count * 80} API calls aggregated into a single GEO score.

Scoring

Mentions are counted per-platform per-query. The GEO Score is a weighted average across decision moments (discovery: 20%), with actual mention count mapped to a 0-100 scale per moment. Multi-run aggregation uses mean scores with trend direction analysis.

Purpose

${queriesRaw.metadata.purpose}

`; const outPath = resolve(DATA_DIR, 'report.html'); Bun.write(outPath, html); console.log(`Report written to ${outPath}`); console.log(`GEO Score: ${geoScore} (${tier})`); console.log(`Mentioned queries: ${mentionedQueries.length}/${queriesRaw.metadata.total_queries}`); console.log(`Zero-mention queries: ${zeroMentionQueries.length}/${queriesRaw.metadata.total_queries}`);