Generative engine optimisation (GEO) asks marketing teams to invest in work whose results appear inside AI answers rather than on a familiar search results page. Budgets are easier to defend when the measurement plan is agreed before the work starts. This paper sets out a practical framework that teams can adopt without specialist tooling.
Why GEO is harder to measure than SEO
Traditional SEO has mature signals: rankings, impressions, clicks and conversions, all visible in search console and analytics tools. AI assistants provide fewer of these. Answers vary by user, by wording and over time. Many mentions do not include a link, and some links carry referral data inconsistently. Measurement therefore needs to combine direct observation of answers with indirect business signals.
Layer one: visibility in answers
Build a fixed set of prompts that real buyers would ask, typically 30 to 100, covering category questions, comparisons, problems and buying criteria. Run them on a regular schedule across the assistants that matter in your markets. For each answer, record whether the brand is mentioned, whether it is recommended, whether your site is cited, and which competitors appear. Track the share of prompts where the brand is mentioned or recommended over time, not individual answers.
Layer two: accuracy of what is said
Visibility alone can mislead. An assistant that mentions the brand with outdated prices or wrong services may do harm. Score a sample of answers for accuracy against a fact sheet. Corrections to source pages often move accuracy faster than any other work.
Layer three: traffic and engagement
In analytics, segment sessions referred by known AI assistant domains. Volumes are often small but engaged. Watch landing pages, time on site and conversion rates for this segment. Note that some AI-influenced visits arrive as direct or branded search traffic, so this layer undercounts.
Layer four: pipeline and revenue signals
Add a "how did you hear about us" field to enquiry forms with an option for AI assistants, and ask sales teams to record it. Track branded search volume as a proxy for awareness. Over several months, compare these signals with the visibility trend from layer one.
Setting targets
Reasonable first-quarter targets focus on layers one and two: increasing the share of priority prompts where the brand is mentioned, and reducing factual errors. Revenue targets are more credible from the second or third quarter onwards, once a baseline exists.
Reporting cadence
Monthly reporting suits most teams: a visibility chart, a list of notable answer changes, accuracy issues fixed, AI referral traffic, and qualitative notes from sales. Quarterly reviews connect the trend to pipeline.
Costs to include
A fair return calculation includes content production, technical fixes, monitoring time and any external service fees. Pricing models for GEO services vary from project fees to monthly retainers; a breakdown of typical service tiers is available at GEO service pricing.
References
- Google Search Central documentation on AI features and websites.
- Industry studies on AI assistant referral traffic published by analytics vendors, 2025–2026.
- Internal measurement templates used by PandaWM for client reporting (anonymised).
Published by PandaWM, an overseas marketing team based in Hong Kong.