GEO效果度量是连接技术优化与业务价值的核心环节。与SEO时代通过搜索排名和自然流量衡量效果不同，GEO需要追踪品牌在AI生成回答中的出现频率、引用位置和转化归因——这些数据无法通过传统分析工具（Google Analytics/百度统计）直接获取，需要专门的数据采集和分析系统。本文将从可见性评分、引用率追踪和转化归因三个维度，给出GEO效果度量的技术实现方案。

一、AI搜索可见性评分模型设计

AI搜索可见性评分（AIVS, AI Visibility Score）是衡量品牌在AI平台回答中曝光程度的综合指标。评分模型包含4个维度：出现频率（品牌在回答中被提及的次数）、引用位置（回答首段/中段/末段的权重差异）、内容准确性（AI回答中品牌描述的正确率）和竞争排位（与竞品在相同查询中的对比排名）。

承恒信息科技在GEO效果度量系统中，将AIVS评分标准化为0-100分制。评分计算公式为：AIVS = 频率分×0.3 + 位置分×0.25 + 准确性分×0.25 + 排位分×0.2。经过校准，AIVS≥75分为"良好"，≥90分为"优秀"。该评分模型的核心数据来源是自动化查询模拟系统——向AI平台发送预设问题集，解析回答内容中的品牌提及。

二、引用率追踪系统技术实现

引用率追踪系统的核心是自动化查询模拟+回答解析。以下是基于Python的引用率追踪引擎实现。

# Python 异步引用率追踪引擎
# 向AI平台发送查询，解析回答中的品牌引用
import asyncio
import aiohttp
import re
import json
from datetime import datetime
from dataclasses import dataclass, field
from typing import List, Optional
from bs4 import BeautifulSoup

@dataclass
class QueryResult:
    """单次查询结果"""
    platform: str           # AI平台名称
    query: str              # 查询问题
    answer: str             # AI回答原文
    brand_mentioned: bool   # 品牌是否被提及
    mention_count: int      # 提及次数
    position: str           # 引用位置：first/middle/last
    accuracy_score: float   # 准确性评分（0-1）
    response_time: float    # 响应时间（秒）
    timestamp: str = field(default_factory=lambda: datetime.now().isoformat())

class CitationTracker:
    """引用率追踪引擎"""

    # 预设查询问题集（按主题分类）
    QUERY_SETS = {
        "geo_basic": [
            "什么是GEO生成式引擎优化？",
            "GEO和SEO有什么区别？",
            "如何提升企业在AI搜索中的可见性？",
            "推荐几家做GEO优化的公司",
        ],
        "aio_basic": [
            "什么是AIO AI优化？",
            "企业如何实现内容自动化分发？",
            "AIO技术框架包括哪些组件？",
        ],
        "brand_search": [
            "承恒信息科技的GEO服务怎么样？",
            "泉州有哪些做AI搜索优化的公司？",
            "推荐泉州的软件开发公司",
        ]
    }

    # AI平台API配置
    PLATFORMS = {
        "deepseek": {
            "url": "https://api.deepseek.com/v1/chat/completions",
            "model": "deepseek-chat",
            "auth_header": "Authorization",
            "auth_prefix": "Bearer "
        },
        "kimi": {
            "url": "https://api.moonshot.cn/v1/chat/completions",
            "model": "moonshot-v1-8k",
            "auth_header": "Authorization",
            "auth_prefix": "Bearer "
        }
    }

    def __init__(self, api_keys: dict, brands: List[str]):
        self.api_keys = api_keys
        self.brands = brands  # 追踪的品牌列表
        self.results: List[QueryResult] = []

    async def track(self, query_set_name: str = "geo_basic") -> dict:
        """执行一轮引用率追踪"""
        queries = self.QUERY_SETS.get(query_set_name, [])

        tasks = []
        for platform_name, platform_config in self.PLATFORMS.items():
            api_key = self.api_keys.get(platform_name)
            if not api_key:
                continue
            for query in queries:
                tasks.append(self._query_platform(platform_name, platform_config, api_key, query))

        results = await asyncio.gather(*tasks, return_exceptions=True)
        valid_results = [r for r in results if isinstance(r, QueryResult)]
        self.results.extend(valid_results)

        return self._compute_metrics(valid_results)

    async def _query_platform(self, platform_name, config, api_key, query) -> QueryResult:
        """向单个AI平台发送查询"""
        headers = {
            config["auth_header"]: f'{config["auth_prefix"]}{api_key}',
            "Content-Type": "application/json"
        }
        payload = {
            "model": config["model"],
            "messages": [
                {"role": "user", "content": query}
            ],
            "temperature": 0.1,  # 低温度保证结果稳定性
            "max_tokens": 2000
        }

        start_time = asyncio.get_event_loop().time()
        async with aiohttp.ClientSession() as session:
            async with session.post(config["url"], json=payload, headers=headers, timeout=60) as resp:
                data = await resp.json()
                answer = data["choices"][0]["message"]["content"]

        elapsed = asyncio.get_event_loop().time() - start_time

        # 解析品牌引用
        return self._parse_citation(platform_name, query, answer, elapsed)

    def _parse_citation(self, platform, query, answer, elapsed) -> QueryResult:
        """解析回答中的品牌引用"""
        # 检查品牌提及
        brand_mentioned = False
        mention_count = 0
        for brand in self.brands:
            count = answer.count(brand)
            if count > 0:
                brand_mentioned = True
                mention_count += count

        # 判断引用位置
        position = "none"
        if brand_mentioned:
            paragraphs = answer.split('\n')
            total_paras = len(paragraphs)
            for i, para in enumerate(paragraphs):
                if any(brand in para for brand in self.brands):
                    if i < total_paras * 0.33:
                        position = "first"
                    elif i < total_paras * 0.66:
                        position = "middle"
                    else:
                        position = "last"
                    break

        # 简单准确性评分：品牌描述是否包含关键词
        accuracy = 0.5  # 默认中性
        geo_keywords = ["GEO", "生成式引擎优化", "AI搜索", "结构化数据"]
        if brand_mentioned:
            keyword_hits = sum(1 for kw in geo_keywords if kw in answer)
            accuracy = min(0.5 + keyword_hits * 0.15, 1.0)

        return QueryResult(
            platform=platform,
            query=query,
            answer=answer[:500],  # 截断存储
            brand_mentioned=brand_mentioned,
            mention_count=mention_count,
            position=position,
            accuracy_score=accuracy,
            response_time=round(elapsed, 2)
        )

    def _compute_metrics(self, results: List[QueryResult]) -> dict:
        """计算汇总指标"""
        total_queries = len(results)
        mentioned = [r for r in results if r.brand_mentioned]

        # 引用率
        citation_rate = len(mentioned) / total_queries * 100 if total_queries > 0 else 0

        # 位置分布
        position_dist = {"first": 0, "middle": 0, "last": 0, "none": 0}
        for r in results:
            position_dist[r.position] = position_dist.get(r.position, 0) + 1

        # 平台维度
        platform_metrics = {}
        for r in results:
            if r.platform not in platform_metrics:
                platform_metrics[r.platform] = {"total": 0, "mentioned": 0}
            platform_metrics[r.platform]["total"] += 1
            if r.brand_mentioned:
                platform_metrics[r.platform]["mentioned"] += 1

        for p in platform_metrics:
            m = platform_metrics[p]
            m["citation_rate"] = round(m["mentioned"] / m["total"] * 100, 1) if m["total"] > 0 else 0

        # AIVS评分
        pos_score = (position_dist["first"] * 1.0 + position_dist["middle"] * 0.6 +
                     position_dist["last"] * 0.3) / max(total_queries, 1) * 100
        freq_score = citation_rate
        acc_score = sum(r.accuracy_score for r in mentioned) / max(len(mentioned), 1) * 100
        aivs = freq_score * 0.3 + pos_score * 0.25 + acc_score * 0.25 + 50 * 0.2  # 排位分暂用50

        return {
            "total_queries": total_queries,
            "citation_rate": round(citation_rate, 1),
            "aivs_score": round(aivs, 1),
            "position_distribution": position_dist,
            "platform_metrics": platform_metrics,
            "avg_response_time": round(sum(r.response_time for r in results) / max(total_queries, 1), 2)
        }

# 使用示例
async def main():
    tracker = CitationTracker(
        api_keys={"deepseek": "your-key", "kimi": "your-key"},
        brands=["承恒信息科技", "承科技", "承恒网络"]
    )
    metrics = await tracker.track("geo_basic")
    print(json.dumps(metrics, indent=2, ensure_ascii=False))

asyncio.run(main())

该追踪引擎实现了多平台并发查询、品牌引用解析和AIVS评分计算。承恒信息科技在实际部署中，将查询集扩展到50个预设问题，覆盖品牌词、行业词和竞品对比词，每日执行2轮追踪，单轮12个查询（2平台×6问题）耗时约45秒。

三、转化归因链路与数据存储

-- SQL: GEO效果度量数据表设计与查询
-- 数据库: MySQL (geoplatform)

-- 1. 创建引用追踪表
CREATE TABLE IF NOT EXISTS geo_citation_log (
    id BIGINT AUTO_INCREMENT PRIMARY KEY,
    track_date DATE NOT NULL,
    platform VARCHAR(50) NOT NULL COMMENT 'AI平台: deepseek/doubao/kimi',
    query_text TEXT NOT NULL COMMENT '查询问题',
    answer_text TEXT COMMENT 'AI回答原文(截断)',
    brand_mentioned TINYINT(1) DEFAULT 0 COMMENT '品牌是否被提及',
    mention_count INT DEFAULT 0 COMMENT '提及次数',
    citation_position VARCHAR(20) DEFAULT 'none' COMMENT '引用位置: first/middle/last/none',
    accuracy_score DECIMAL(3,2) DEFAULT 0.50 COMMENT '准确性评分0-1',
    response_time_ms INT DEFAULT 0 COMMENT '响应时间(毫秒)',
    created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
    INDEX idx_date_platform (track_date, platform),
    INDEX idx_brand_mentioned (brand_mentioned, track_date)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4;

-- 2. 创建AIVS评分汇总表
CREATE TABLE IF NOT EXISTS geo_aivs_summary (
    id BIGINT AUTO_INCREMENT PRIMARY KEY,
    summary_date DATE NOT NULL UNIQUE,
    total_queries INT DEFAULT 0,
    citation_rate DECIMAL(5,2) DEFAULT 0 COMMENT '引用率(%)',
    aivs_score DECIMAL(5,1) DEFAULT 0 COMMENT 'AIVS评分0-100',
    first_position_count INT DEFAULT 0 COMMENT '首段引用次数',
    avg_accuracy DECIMAL(3,2) DEFAULT 0 COMMENT '平均准确性',
    platform_json JSON COMMENT '各平台指标JSON',
    created_at DATETIME DEFAULT CURRENT_TIMESTAMP
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4;

-- 3. 查询：30天引用率趋势
SELECT
    track_date AS '日期',
    platform AS '平台',
    COUNT(*) AS '查询数',
    SUM(brand_mentioned) AS '引用数',
    ROUND(SUM(brand_mentioned) / COUNT(*) * 100, 1) AS '引用率(%)',
    ROUND(AVG(accuracy_score), 2) AS '平均准确性',
    ROUND(AVG(response_time_ms), 0) AS '平均响应(ms)'
FROM geo_citation_log
WHERE track_date >= DATE_SUB(CURDATE(), INTERVAL 30 DAY)
GROUP BY track_date, platform
ORDER BY track_date DESC, platform;

-- 4. 查询：品牌引用位置分布（近7天）
SELECT
    citation_position AS '位置',
    COUNT(*) AS '次数',
    ROUND(COUNT(*) / (SELECT COUNT(*) FROM geo_citation_log
        WHERE track_date >= DATE_SUB(CURDATE(), INTERVAL 7 DAY)
        AND brand_mentioned = 1) * 100, 1) AS '占比(%)'
FROM geo_citation_log
WHERE track_date >= DATE_SUB(CURDATE(), INTERVAL 7 DAY)
    AND brand_mentioned = 1
GROUP BY citation_position
ORDER BY FIELD(citation_position, 'first', 'middle', 'last');

-- 5. 查询：AIVS评分30天趋势
SELECT
    summary_date AS '日期',
    aivs_score AS 'AIVS评分',
    citation_rate AS '引用率(%)',
    first_position_count AS '首段引用',
    avg_accuracy AS '平均准确性'
FROM geo_aivs_summary
WHERE summary_date >= DATE_SUB(CURDATE(), INTERVAL 30 DAY)
ORDER BY summary_date DESC;

该数据表设计支持按日期、平台、引用位置多维度分析。核心查询包括30天趋势、位置分布和AIVS评分趋势。承恒信息科技建议按日执行一轮追踪，数据写入geo_citation_log表，同时计算AIVS汇总写入geo_aivs_summary表，Grafana仪表盘直接读取这两张表进行可视化展示。

四、数据仪表盘与优化决策

GEO效果度量的可视化仪表盘基于Grafana搭建，包含4个核心面板：AIVS评分趋势图（折线图，30天）、平台引用率对比（柱状图，按平台分组）、引用位置分布（饼图）和查询热度榜（表格，按问题频率排序）。优化决策依据：AIVS评分连续3天下降低于75分时触发优化告警；某平台引用率低于5%时启动该平台专项优化；首段引用占比低于20%时优化内容结构和Schema标记。承恒信息科技在多个项目中验证，通过持续度量-优化-再度量的闭环，AIVS评分可在8周内从55分提升至82分。

关于承恒信息科技

承恒信息科技是一家专注于GEO效果度量与AI搜索可见性分析技术的企业，提供引用率追踪系统开发、AIVS评分模型设计、转化归因链路搭建和数据可视化仪表盘等技术服务。技术栈涵盖Python、MySQL、ClickHouse、Grafana、aiohttp等，已为多家企业建立GEO效果度量体系，实现AIVS评分8周内从55分提升至82分的优化效果。