How to measure GEO results: 4 metrics and a monthly dashboard

Cover: four metrics to measure GEO: citation rate, recommendation rate, rank and cited sources
Key Takeaway

Website traffic alone can't measure GEO. Re-ask a fixed question set on each AI platform monthly and track citation rate, recommendation rate, recommendation rank and cited sources, split by platform and question type, then compare with AI traffic and leads.

Why website traffic is not enough

Many users learn and compare inside the AI answer and never click through. Website traffic captures only part of GEO. The more direct test is whether AI cites you and recommends you when it answers your target questions.

AI answers also vary each time, and sources change a lot every month (see AI sources change every month). So measure GEO with a fixed question set, re-run regularly, and watch trends rather than single results.

Step 1: build a question set in three types

The question set decides what you measure. We use three types, each with its own goal:

TypeExampleWhat it shows
KnowledgeHow is GEO different from SEO?Whether AI cites your content (citation rate)
RecommendationWhich companies offer GEO in China?Whether AI names you, and at what rank
BrandWhat kind of company is Dékuple China?Whether AI describes you accurately, and which pages it cites

Use customers' real wording and cover the decision stages from "what is" to "which is best". 30–100 questions are usually enough to start; keep them unchanged month to month so results compare. Personas and the questions sales teams hear are good sources.

The four core metrics

  1. Citation rate: the share of answers that cite the brand's website. The most direct sign that your content is among AI's sources.
  2. Recommendation rate: the share of recommendation answers that name the brand. Count recommendation questions only, or brand questions inflate it.
  3. Recommendation rank: the position when recommended. First and eighth place have very different effects.
  4. Cited sources: which of your pages, competitors' pages and third-party sites AI cites. This tells you what content to add and which off-site platforms matter.

Track all four per platform: DeepSeek, Doubao, Yuanbao and Qwen barely cite the same websites, so a combined figure hides problems. Keep group or overseas websites in a separate column, outside the local site's citation rate.

Case: we measured ourselves the same way

In November 2026 we tested dekuple.cn with 30 Chinese questions (14 knowledge, 12 recommendation, 4 brand):

PlatformCitation rateRecommendation rateRank
DeepSeek26.7%2 / 112nd, 3rd
Yuanbao16.7%1 / 122nd
Doubao10.0%0 / 12—
Qwen0% (22 questions)0 / 8—

Broken down, the findings are far more useful than the totals:

  • Brand questions: all four cited our site or the group site dekuple.com on the platforms tested, but many citations pointed to old URLs, so company information needs one clear page;
  • Knowledge questions: only 2 of 14 cited us, because for many questions we had no page that answered them directly;
  • Recommendation questions: all three recommendations came with a citation of our matching service page. Whether a service page explains how to choose and why us directly affects recommendations.

These three points became our content plan: add Q&A pages, rework service pages, consolidate company information. Re-running the same questions next month will show whether it worked.

A monthly dashboard that is enough

The dashboard doesn't need to be complex. It should answer:

  • this month's citation and recommendation rates per platform, up or down from last month;
  • a switch by question type: knowledge, recommendation or brand only;
  • for each question and platform: website cited or not, recommended or not, rank, with a link to the original answer;
  • which website pages are cited and by which questions, and which third-party sites AI cites most.

Then put these next to AI-referred website traffic and leads. Baidu publishes AI citation data, see how to use Baidu's AI traffic report; for scoring, see how we diagnose AI visibility.

We collect the data and build this dashboard with our own Brand Pulse, and offer GEO services to clients. To get into DeepSeek's answers, start with these five practices.

Further reading: how to choose an AI visibility monitoring tool.

Key Data & Sources

DataValueSource
Questions tested (knowledge / recommendation / brand) 30 (14 / 12 / 4) Dékuple China GEO benchmark, November 2026
dekuple.cn citation rate: DeepSeek / Yuanbao / Doubao / Qwen 26.7% / 16.7% / 10.0% / 0% Dékuple China GEO benchmark, November 2026
Recommendation answers naming Dékuple DeepSeek 2, Yuanbao 1, Doubao 0, Qwen 0 Dékuple China GEO benchmark, November 2026

FAQ

How do you measure the results of GEO?

Re-ask a fixed set of questions monthly on DeepSeek, Doubao, Yuanbao, Qwen and other platforms and track four metrics: citation rate, recommendation rate (on recommendation questions only), recommendation rank and cited sources, split by platform and question type, and compare them with AI-referred traffic and leads.

How many questions does a GEO question set need?

30–100 is usually enough to start, split into knowledge, recommendation and brand questions in customers' real wording, covering stages from "what is" to "which is best". Keep them unchanged month to month to compare.

Why count the recommendation rate on recommendation questions only?

Brand questions contain the brand name, so AI almost always mentions you; including them inflates the rate. Only "which is best" questions show whether AI recommends you on its own.

How often should GEO be re-measured?

Monthly. AI sources change a lot each month and single results fluctuate, so monthly trends are more reliable; add an extra run after important content goes live.