如何构建自动化 LLM 可见性扫描仪

import os, requests, json
from datetime import datetime
from collections import defaultdict

API_KEY = os.environ['SCAVIO_API_KEY']
SH = {'x-api-key': API_KEY, 'Content-Type': 'application/json'}

BRAND_DOMAIN = 'scavio.dev'
KEYWORDS = [
    'best search api for ai agents',
    'mcp search tool setup',
    'web search api comparison 2026',
    'serp api free tier options',
    'search api for rag pipeline',
]

def scan_llm_visibility(keyword, domain):
    data = requests.post('https://api.scavio.dev/api/v1/search',
        headers=SH, json={'query': keyword, 'country_code': 'us'}, timeout=10).json()
    domain_l = domain.lower()
    # AI Overview / Answer Box (feeds LLM responses)
    ai = data.get('ai_overview', data.get('answer_box', {}))
    ai_cited = domain_l in json.dumps(ai).lower() if ai else False
    # Featured Snippet (high-priority LLM source)
    featured = data.get('featured_snippet', {})
    in_featured = domain_l in json.dumps(featured).lower() if featured else False
    # People Also Ask (LLM follow-up context)
    paa = data.get('people_also_ask', [])
    in_paa = any(domain_l in json.dumps(q).lower() for q in paa)
    # Organic position (base visibility)
    organic = data.get('organic_results', [])
    org_pos = next((i+1 for i, r in enumerate(organic) if domain_l in r.get('link', '').lower()), None)
    # Top 3 domains (who LLMs are citing)
    top3 = [r.get('displayed_link', '').split('/')[0] for r in organic[:3]]
    return {
        'keyword': keyword,
        'ai_cited': ai_cited,
        'has_ai_overview': bool(ai),
        'in_featured': in_featured,
        'in_paa': in_paa,
        'organic_pos': org_pos,
        'top3_domains': top3,
    }

scans = []
for kw in KEYWORDS:
    scan = scan_llm_visibility(kw, BRAND_DOMAIN)
    scans.append(scan)
    signals = []
    if scan['ai_cited']: signals.append('AI')
    if scan['in_featured']: signals.append('FS')
    if scan['in_paa']: signals.append('PAA')
    pos = f'#{scan["organic_pos"]}' if scan['organic_pos'] else '-'
    print(f'  {kw[:40]:40} | {pos:5} | {", ".join(signals) or "none"}')
print(f'\nCost: ${len(KEYWORDS) * 0.005:.3f}')

def calculate_llm_score(scans, domain):
    total = len(scans)
    # Component scores
    ai_cited = sum(1 for s in scans if s['ai_cited'])
    has_ai = sum(1 for s in scans if s['has_ai_overview'])
    featured = sum(1 for s in scans if s['in_featured'])
    paa = sum(1 for s in scans if s['in_paa'])
    top3 = sum(1 for s in scans if s['organic_pos'] and s['organic_pos'] <= 3)
    top10 = sum(1 for s in scans if s['organic_pos'] and s['organic_pos'] <= 10)
    # Weighted score (100 max)
    ai_score = (ai_cited / max(has_ai, 1)) * 35  # AI citations worth most
    featured_score = (featured / total) * 25  # Featured snippets
    paa_score = (paa / total) * 15  # PAA presence
    organic_score = (top3 / total) * 15 + (top10 / total) * 10  # Organic ranking
    total_score = ai_score + featured_score + paa_score + organic_score
    print(f'\n=== LLM Visibility Score: {domain} ===')
    print(f'  Overall: {total_score:.0f}/100')
    print(f'\n  Component Breakdown:')
    print(f'    AI Citations:    {ai_score:.0f}/35 ({ai_cited}/{has_ai} keywords with AI)')
    print(f'    Featured Snippet: {featured_score:.0f}/25 ({featured}/{total} keywords)')
    print(f'    People Also Ask: {paa_score:.0f}/15 ({paa}/{total} keywords)')
    print(f'    Organic Top 3:   {top3}/{total} keywords')
    print(f'    Organic Top 10:  {top10}/{total} keywords')
    return {
        'total_score': total_score,
        'ai_score': ai_score,
        'featured_score': featured_score,
        'paa_score': paa_score,
        'organic_score': organic_score,
    }

visibility = calculate_llm_score(scans, BRAND_DOMAIN)

def generate_recommendations(scans, visibility, domain):
    print(f'\n{"=" * 60}')
    print(f'  LLM VISIBILITY SCANNER REPORT')
    print(f'  Brand: {domain} | Date: {datetime.now().strftime("%Y-%m-%d")}')
    print(f'  Score: {visibility["total_score"]:.0f}/100')
    print(f'{"=" * 60}')
    recs = []
    # AI citation gaps
    ai_gaps = [s['keyword'] for s in scans if s['has_ai_overview'] and not s['ai_cited']]
    if ai_gaps:
        recs.append({
            'priority': 'HIGH',
            'action': 'Optimize for AI citations',
            'keywords': ai_gaps,
        })
    # Featured snippet opportunities
    fs_gaps = [s['keyword'] for s in scans if not s['in_featured'] and s['organic_pos'] and s['organic_pos'] <= 5]
    if fs_gaps:
        recs.append({
            'priority': 'MEDIUM',
            'action': 'Target featured snippets (already ranking top 5)',
            'keywords': fs_gaps,
        })
    # Not ranking at all
    absent = [s['keyword'] for s in scans if not s['organic_pos']]
    if absent:
        recs.append({
            'priority': 'HIGH',
            'action': 'Create content (not ranking)',
            'keywords': absent,
        })
    print(f'\n  Recommendations ({len(recs)}):')
    for r in recs:
        print(f'\n  [{r["priority"]:6}] {r["action"]}')
        for kw in r['keywords'][:3]:
            print(f'    - {kw}')
    # Competitor visibility
    print(f'\n  Top Competitors (by AI/organic presence):')
    all_domains = defaultdict(int)
    for s in scans:
        for d in s.get('top3_domains', []):
            if d and domain not in d:
                all_domains[d] += 1
    for d, count in sorted(all_domains.items(), key=lambda x: -x[1])[:5]:
        print(f'    {d:30} | Top 3 in {count}/{len(scans)} keywords')
    print(f'\n  Scan cost: ${len(scans) * 0.005:.3f}')
    print(f'  Monthly (daily): ${len(scans) * 0.005 * 30:.2f}')

generate_recommendations(scans, visibility, BRAND_DOMAIN)