Measuring GEO: The Metrics That Track AI Visibility

To measure GEO, score Citation Share on a fixed prompt set across five engines, using five weighted components: Citation Frequency (40%), Cross-Engine Breadth (20%), Query-Type Breadth (20%), Extractability (15%) and Crawl Access (5%). Run each prompt several times, repeat the run monthly, and compare against competitors. A Generative Engine Optimization program without measurement is content marketing with a new label.

Updated October 12, 2026. The weights below are EPR's own diagnostic framework, not an industry standard.

What does GEO measurement need to answer?

GEO measurement needs to answer four questions: whether the brand appears inside AI answers, where it appears, where it is missing, and what changed since the last measurement. The headline metric is Citation Share, a brand's share of the answers across a defined prompt set. The composite is built from five inputs.

What is Citation Frequency?

Citation Frequency is the percentage of prompts in which the brand name appears in the engine's answer. It carries the largest weight because appearance is the threshold condition: a brand that does not appear cannot be evaluated on any other dimension.

A prompt set is a curated list of buyer questions in the brand's category. A consumer-electronics brand might run 50 prompts covering category questions ("best wireless earbuds 2026"), comparison questions ("Sony WH-1000XM5 vs Bose QuietComfort Ultra") and use-case questions ("earbuds for running with sweat resistance"). The 35-Prompt Citation Share Audit provides a ready-made set.

What is Cross-Engine Breadth?

Cross-Engine Breadth is the number of the five major engines that cite the brand. A brand cited in three of five has incomplete coverage and may be invisible to a buyer whose default engine is one of the two that miss it. The metric exposes those gaps and forces engine-specific tuning, which the engine guides cover: ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews.

What is Query-Type Breadth?

Query-Type Breadth is coverage across four question types, the same four the six-step citation audit uses.

• Definitional questions, such as "what is X."
• Comparative questions, such as "X vs Y."
• Procedural questions, such as "how do I X."
• Recommendation questions, such as "best X for Y."

A brand cited only on definitional questions is recognized but not recommended. A brand cited only on comparative questions is contested but does not own the category. The metric shows where the brand stands in the buying process.

What is Extractability?

Extractability is the structural quality of the pages engines cite. Two pages can carry the same content, and the one with clean heading hierarchy, answer-first paragraphs and accurate schema is easier to quote correctly. Measure it by inspecting each cited page against a short rubric: an answer in the first sentence under each heading, sourced numbers, tables for comparisons, and content present in the initial HTML.

Low extractability is a common finding in first-time audits, and it is among the faster problems to fix. Restructuring headings, adding definition blocks and cleaning up markup can be done in weeks. Google says no special schema is required to appear in its AI features, so treat structure as clarity for readers and machines, not as a switch.

What is Crawl Access?

Crawl Access is the floor check: whether engines can reach the pages at all. A robots.txt block on GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot or Google-Extended can make a site invisible to the matching engine, and aggressive bot management at the firewall can do the same. It has the smallest weight because it is usually a yes or no, but when the answer is no the other four components do not matter. The per-engine robot details are in the ChatGPT, Claude and Perplexity guides.

How do you calculate the composite score?

The composite is the weighted sum of the five components, each scored out of 100: Citation Frequency x 0.40, plus Cross-Engine Breadth x 0.20, plus Query-Type Breadth x 0.20, plus Extractability x 0.15, plus Crawl Access x 0.05.

Illustrative scenario: a brand appears in 30% of prompts, in three of five engines (60), in two of four query types (50), scores 50 on extractability and 100 on crawl access. Its composite is 12 + 12 + 10 + 7.5 + 5 = 46.5. The breakdown shows where to work first: frequency and query-type breadth are low, while access is already clean.

EPR reads scores as follows, as a rule of thumb and not as an industry standard: above 60 means a brand actively present in AI answers, above 80 means category-leading, and below 40 means the brand is largely missing from the answer surface and needs foundational work first. This is the scoring formula behind the 5W AI Citation Audit, which runs across the five engines on a defined prompt set and produces a composite plus engine-by-engine and component-by-component breakdowns.

Which other metrics belong on the dashboard?

Four supporting metrics add context to the composite.

How often should you measure GEO?

Measure monthly, using the same prompt set, the same engines and the same method each time. AI citation sets change quickly: EPR's research on the citation half-life of AI visibility wins found that only 10.6% of cited URLs persisted across 28 days. A quarterly cadence misses that movement, and a weekly cadence reads too much into noise.

Run each prompt at least three times per engine, because answers vary run to run, and report a range with a clear method instead of a single score. The first audit sets the baseline, the second confirms direction, and the third shows whether the work is holding. The full method is in The AI Visibility Audit.