GEO（Generative Engine Optimization）的效果度量与传统SEO有本质区别：搜索引擎返回的是链接列表，而生成式AI引擎返回的是合成答案，用户未必会点击任何外部链接。因此，我们需要一套全新的度量体系来量化内容在AI搜索中的真实表现。本文将从可见性指数、引用率追踪、转化归因三个维度，给出可落地的技术实现方案。

一、GEO可见性指数（Visibility Index）计算模型

可见性指数是衡量品牌/内容在AI生成答案中"曝光程度"的综合指标。我们定义以下计算模型：当用户查询与目标关键词相关的query时，AI引擎生成答案中是否提及品牌名、是否引用我们的内容片段、以及提及的位置（开头/中间/结尾）和频次。基于这些信号，可以构建一个加权评分体系。

以下是可见性指数计算的核心Python实现：

import re
from dataclasses import dataclass, field
from typing import List
from datetime import datetime

@dataclass
class CitationSignal:
    """单次AI搜索结果中的引用信号"""
    query: str
    brand_mentioned: bool
    mention_count: int
    mention_position: str  # "top", "middle", "bottom"
    content_cited: bool    # 是否引用了我们的内容片段
    citation_url: str = ""
    timestamp: str = field(default_factory=lambda: datetime.now().isoformat())

@dataclass
class VisibilityScore:
    keyword: str
    total_queries: int
    visibility_index: float
    raw_signals: List[CitationSignal] = field(default_factory=list)

def compute_visibility_index(signals: List[CitationSignal], keyword: str) -> VisibilityScore:
    """
    可见性指数计算公式：
    VI = (提及率 * 0.3 + 引用率 * 0.4 + 位置加权分 * 0.3) * 100
    其中位置加权分：top=1.0, middle=0.6, bottom=0.3
    """
    if not signals:
        return VisibilityScore(keyword=keyword, total_queries=0, visibility_index=0.0)

    total = len(signals)
    mentioned = [s for s in signals if s.brand_mentioned]
    cited = [s for s in signals if s.content_cited]

    mention_rate = len(mentioned) / total
    citation_rate = len(cited) / total

    position_weights = {"top": 1.0, "middle": 0.6, "bottom": 0.3}
    position_scores = []
    for s in mentioned:
        weight = position_weights.get(s.mention_position, 0.3)
        # 多次提及额外加权，上限3次
        count_bonus = min(s.mention_count, 3) / 3 * 0.2
        position_scores.append(weight + count_bonus)

    avg_position = sum(position_scores) / len(position_scores) if position_scores else 0.0

    vi = (mention_rate * 0.3 + citation_rate * 0.4 + avg_position * 0.3) * 100
    vi = round(min(vi, 100.0), 2)

    return VisibilityScore(
        keyword=keyword,
        total_queries=total,
        visibility_index=vi,
        raw_signals=signals
    )

# 使用示例
signals = [
    CitationSignal("泉州小程序开发公司推荐", True, 2, "top", True, "https://example.com/post/1"),
    CitationSignal("泉州软件开发哪家好", True, 1, "middle", False),
    CitationSignal("公众号开发流程", False, 0, "", False),
    CitationSignal("泉州网络推广方案", True, 3, "top", True, "https://example.com/post/2"),
]

score = compute_visibility_index(signals, "泉州软件开发")
print(f"关键词: {score.keyword}")
print(f"总查询数: {score.total_queries}")
print(f"可见性指数: {score.visibility_index}")
# 输出: 可见性指数: 73.67

二、AI引用率追踪埋点方案

引用率追踪的核心难点在于：AI引擎的答案是在其自有界面中渲染的，我们无法直接在对方页面埋点。可行的方案是通过API层代理调用来采集AI搜索结果。具体做法是封装一个统一的AI搜索查询层，定期以目标关键词集向DeepSeek、OpenAI、Claude等模型发起查询，解析返回结果中的引用URL和品牌提及情况。

以下是引用率追踪的API查询与解析模块：

import asyncio
import httpx
import json
import re
from urllib.parse import urlparse
from typing import List, Dict

class AICitationTracker:
    """AI搜索引擎引用率追踪器"""

    def __init__(self, api_keys: Dict[str, str], target_domains: List[str], brand_names: List[str]):
        self.api_keys = api_keys
        self.target_domains = set(target_domains)
        self.brand_names = brand_names
        self.results: List[Dict] = []

    async def query_deepseek(self, client: httpx.AsyncClient, keyword: str) -> Dict:
        """调用DeepSeek API查询并解析引用"""
        headers = {
            "Authorization": f"Bearer {self.api_keys['deepseek']}",
            "Content-Type": "application/json"
        }
        payload = {
            "model": "deepseek-chat",
            "messages": [
                {"role": "system", "content": "你是一个搜索助手，请基于你的知识回答用户问题。"},
                {"role": "user", "content": f"请推荐：{keyword}"}
            ],
            "temperature": 0.3
        }
        resp = await client.post(
            "https://api.deepseek.com/v1/chat/completions",
            headers=headers, json=payload, timeout=30
        )
        data = resp.json()
        answer = data["choices"][0]["message"]["content"]

        # 提取引用URL
        urls = re.findall(r'https?://[^\s\)\]]+', answer)
        cited_domains = set()
        for url in urls:
            domain = urlparse(url).netloc
            if domain in self.target_domains:
                cited_domains.add(domain)

        # 检测品牌提及
        brand_hits = [b for b in self.brand_names if b in answer]

        return {
            "engine": "deepseek",
            "keyword": keyword,
            "answer_length": len(answer),
            "cited_domains": list(cited_domains),
            "brand_mentions": brand_hits,
            "mention_count": sum(answer.count(b) for b in self.brand_names),
            "raw_answer": answer[:500]
        }

    async def batch_track(self, keywords: List[str]) -> List[Dict]:
        """批量追踪多个关键词的引用情况"""
        async with httpx.AsyncClient() as client:
            tasks = [self.query_deepseek(client, kw) for kw in keywords]
            results = await asyncio.gather(*tasks, return_exceptions=True)

        for r in results:
            if isinstance(r, Exception):
                print(f"查询失败: {r}")
                continue
            self.results.append(r)
        return self.results

# 使用示例
tracker = AICitationTracker(
    api_keys={"deepseek": "sk-your-api-key"},
    target_domains=["example.com", "blog.example.com"],
    brand_names=["某某科技", "ExampleTech"]
)

# async run
async def main():
    keywords = ["泉州小程序开发", "泉州公众号开发公司", "网络推广方案"]
    results = await tracker.batch_track(keywords)
    for r in results:
        print(f"[{r['engine']}] {r['keyword']} -> 引用:{r['cited_domains']}, 提及:{r['brand_mentions']}")

asyncio.run(main())

三、多触点转化归因模型

在GEO场景中，用户的转化路径通常是：AI搜索看到品牌提及 → 点击引用链接访问网站 → 浏览内容 → 填写表单/联系客服。这涉及多个触点，需要用多触点归因模型来合理分配转化贡献。我们采用基于位置的位置加权归因（Position-based）与数据驱动归因（DDA）相结合的方式。

以下是转化归因分析的SQL实现，基于ClickHouse存储的触点事件表：

-- GEO多触点转化归因分析 (ClickHouse SQL)
-- 表结构: geo_touchpoints(session_id, touchpoint_type, touchpoint_source,
--          brand_keyword, event_time, user_id, converted)

-- 1. 构建用户转化路径
WITH conversion_paths AS (
    SELECT
        session_id,
        user_id,
        groupArray(touchpoint_type) AS touchpoint_sequence,
        groupArray(touchpoint_source) AS source_sequence,
        groupArray(brand_keyword) AS keyword_sequence,
        count() AS touchpoint_count,
        maxIf(converted, converted = 1) AS is_converted,
        min(event_time) AS first_touch,
        max(event_time) AS last_touch,
        dateDiff('minute', min(event_time), max(event_time)) AS path_duration_min
    FROM geo_touchpoints
    WHERE event_time >= '2026-07-01'
      AND event_time < '2026-08-01'
    GROUP BY session_id, user_id
    HAVING touchpoint_count >= 1
),
-- 2. 位置加权归因 (首尾各40%，中间20%)
position_based_attribution AS (
    SELECT
        session_id,
        arrayElement(source_sequence, 1) AS first_source,
        arrayElement(source_sequence, touchpoint_count) AS last_source,
        source_sequence,
        touchpoint_count,
        is_converted,
        CASE
            WHEN touchpoint_count = 1 THEN 1.0
            WHEN touchpoint_count = 2 THEN 0.5
            ELSE 0.4  -- 首尾各40%，中间平分20%
        END AS first_last_weight,
        CASE
            WHEN touchpoint_count <= 2 THEN 0.0
            ELSE 0.2 / (touchpoint_count - 2)
        END AS middle_weight
    FROM conversion_paths
    WHERE is_converted = 1
)
-- 3. 按AI引擎来源汇总归因转化贡献
SELECT
    src AS ai_source,
    count() AS attributed_conversions,
    round(sum(contribution), 2) AS total_contribution,
    round(avg(contribution), 4) AS avg_contribution,
    round(count() * 100.0 / sum(count()) OVER (), 1) AS share_pct
FROM (
    SELECT
        first_source AS src,
        first_last_weight AS contribution
    FROM position_based_attribution
    UNION ALL
    SELECT
        last_source AS src,
        first_last_weight AS contribution
    FROM position_based_attribution
    UNION ALL
    -- 中间触点归因 (需展开数组，此处简化为展开逻辑)
    SELECT
        arrayJoin(arraySlice(source_sequence, 2, touchpoint_count - 2)) AS src,
        middle_weight AS contribution
    FROM position_based_attribution
    WHERE touchpoint_count > 2
) all_touches
GROUP BY src
ORDER BY total_contribution DESC;

四、度量数据可视化与告警

度量体系的最后一环是可视化与告警。我们推荐使用Grafana+Prometheus组合，将可见性指数、引用率、转化归因数据通过自定义Exporter暴露为Prometheus指标，然后在Grafana中构建GEO效果监控大盘。关键告警规则包括：可见性指数连续3天下降超过15%、引用率低于历史均值20%、新关键词引用率为零等。通过这套度量体系，技术团队能够用数据驱动GEO策略迭代，而非依赖主观判断。