How to use this reference

Start with the data lifecycle, then open the methodology for the number or behaviour you are investigating. These guides describe the implementation reviewed on 5 November 2026. Saved results can use earlier versions, and feature availability follows the workspace and package. Numerical examples are synthetic and are not verified merchant results. A versioned result's own definition and evidence take precedence over a generic label.

Data collection and chatbot answers

Follow the source from discovery to indexing, then inspect which passages reach the model and what can be stated from them. These guides explain why a successful fetch, a ready index, and a correctly grounded answer are separate stages.

How ShopFlare finds, processes, and connects dataWebsite discovery, crawling, and knowledge indexingHow chatbot retrieval finds and grounds an answerAnswer generation, model routing, and safe hand-offs

AEO measurement and interpretation

Separate model tests from AI search, brand presence from retailer recommendations, and strict funnels from independent signals. Use the formula and denominator guides when percentages look inconsistent, and the sampling guide when evaluating repeated experiments.

From a monitored question to an evidence-backed AEO resultAEO rates, retailer wins, and unclear answers explainedAEO and purchase-visibility scores: formulas and publication rulesHow the KI recommendation funnel is countedCitations, competitor evidence, and content opportunitiesComparable measurements, repeated sampling, and uncertainty

Commerce discovery and catalogue decisions

Trace an observed product to a supported recommendation and an active catalogue identity. Read unit-based shares separately from answer rates, and distinguish a versioned composite score from current coverage or commercial potential.

How Commerce Intelligence discovers products and retailersCommerce metrics: answers, recommendation units, and sharesCatalogue matching, coverage buckets, and the CI ScoreHow market opportunities are prioritised and reviewed

Advisor analytics and technical checks

Inspect how products qualify for recommendation, how attempts and exits are counted, and how unknown audit checks affect coverage. A result click is not an order; technical accessibility is not a recommendation ranking.

How Product Advisor questions and recommendations workProduct Advisor sessions, choices, and abandonment analyticsTechnical Visibility Audit and PageSpeed scoring

Connected insights, periods, and follow-up

Understand source freshness, the shared account schedule, independent job completion, topic evidence, and the difference between a proposed task and a verified outcome.

Weekly account synchronisation and background jobsCockpit periods, Search Console, and connector freshnessTrend Intelligence sources, signals, and comparable historyFrom a finding to a task and a measured follow-up

Reproducibility checklist

To investigate a number, capture the metric and its unit, the raw supporting counts, workspace, source period, filters, collection method, provider/model, question or flow version, catalogue timestamp, and formula or extraction version. Open the underlying answer, page, product, or event evidence. Do not include passwords, API credentials, private customer conversations, or payment details in a public report.

  • Confirm that the denominator contains only the eligible population intended by the metric.
  • Distinguish missing or unresolved data from an assessed negative or a measured zero.
  • Check whether the comparison uses compatible questions, methods, periods, and formula versions.
  • Separate an observation from its proposed explanation, next action, and later measured outcome.