All our News
AI Search2026-11-02

Measuring GEO: Five KPIs That Actually Matter

Written by FROMDB Limited
Measuring GEO: Five KPIs That Actually Matter

"Did GEO bring more business?" is a fair question — but "our rankings improved" is not an answer.

GEO results sit in layers: are you cited, are you seen, do people visit, do they enquire. These five KPIs map to those layers.

KPI 1: citation rate

Definition: of a fixed set of 30 prompts, how many AI answers mention your brand.

  • Test monthly, same prompts, same platform versions
  • Log: mentioned or not, position, whether a link appears
  • Target: 20–30% improvement over baseline within three months

KPI 2: AI impressions in Search Console

Impressions from AI Overviews and AI Mode are available in the Generative AI performance report in Search Console.

  • Watch the trend, not a single month
  • See which pages generate AI impressions
  • Rising impressions with falling clicks is normal — summaries absorb clicks

KPI 3: AI referral traffic

In GA4 or server logs, track referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com.

  • Low volume but high intent — the user already saw a recommendation
  • Check landing pages: service pages or articles
  • Compare conversion rate against other channels

KPI 4: branded search volume

Queries containing your brand name in the GSC query report. This is the most underrated GEO metric.

  • After an AI mention, many users search the brand name directly
  • Compare three months before and after the work
  • Also track "brand + service" combinations

KPI 5: share of answer

For the same question, how many companies the AI answer names — and how often you are one of them.

  • Pick ten core questions and record the companies named each time
  • Share of answer = your appearances ÷ total tests
  • Log the competitor set to see who appears consistently

What a monthly report should contain

  • One-page summary: changes, biggest finding, next actions
  • KPI table: current versus baseline for all five metrics
  • Prompt detail: results for all 30 questions
  • Content log: what was published or rewritten
  • Technical log: indexing, schema and speed fixes

When is it working?

  • Months 1–2: your brand starts appearing for core questions, even low down
  • Months 3–4: at least 20% of core questions mention you; AI referral traffic appears
  • Month 6: visible lift in branded search, enquiries attributed to AI platforms

FAQ

How do we measure without tools?

Manually: open a fresh conversation each time (to avoid context carry-over) and log every prompt. Thirty prompts takes 30–45 minutes.

AI answers vary each time — how do we compare?

Fix the prompts, the platform and the cadence (for example the first Monday of each month), and average multiple runs.

Impressions rose but revenue did not — is it useless?

Find which layer broke: mentions without clicks suggests low brand familiarity; traffic without enquiries points to weak service-page persuasion.

Should we buy a third-party AI monitoring tool?

Useful if budget allows, but not required. A spreadsheet and a fixed process are enough at the start.

Want a ready reporting framework?

Our monthly report covers all five KPIs with the 30-prompt detail and next actions. Details: SEO & GEO AI search optimization.

Further reading: The complete GEO guide for Hong Kong, How ChatGPT chooses citations

Sources