Methodology

Collection uses the same authenticated consumer web interfaces available to subscribed users. Cooper operates active subscriptions for ChatGPT, Gemini, Claude, and Perplexity and submits prompts through their user-facing products rather than provider APIs. This captures interface-level search, model routing, answer formatting, and citation behavior that a subscriber is more likely to experience.

Each query runs in a fresh controlled browser session through geographically distributed proxies. The scoring-version regional mix gives the United States the largest weight, Europe the next-largest weight, and the rest of the world a smaller combined weight while keeping it represented. Sessions exclude prior conversation history so geography and the published prompt remain the primary controlled variables.

Cooper runs 6–12 independent queries for every question and assistant in each measured regional cell. Collection begins with six observations. If product-set overlap and answer ordering show material drift, additional observations are collected until consensus stabilizes or the cell reaches twelve. Stable cells stop at six; variable cells receive more measurement without receiving more scoring weight.

Each observation stores its start time, assistant account, interface surface, region, prompt version, rendered answer, citations, completion state, and extraction status. A snapshot is eligible for publication only after every required cell has reached its collection rule and all extracted entities have passed validation.

Cooper stores each cited URL, normalized page, source class, measurement cell, and the assistant response that exposed it. Evidence is attached to the run that produced it so later source changes do not rewrite the historical record.

Cited pages for a product are pages the assistant cited in that answer and Cooper verified as naming or representing that product. Aggregate source totals also include cited pages that could not be attributed to a specific product, including inaccessible or unsupported pages.

Duplicate normalized URLs are counted once in each applicable view. A page can support zero, one, or multiple products, but only verified product links contribute to product-level counts. A citation does not expose model training data, prove a causal relationship, or mean that Cooper endorses the cited publisher.

Cooper publishes the information required to understand and interpret AIQ without publishing the controls needed to reproduce or manipulate the collection system. Public documentation includes the measurement surface, tracked assistants, buyer-intent families, query range, regional weighting hierarchy, assistant weights, scoring formula, limitations, and material methodology changes.

The exact prompt wording and variants, numerical adaptive consensus thresholds, query timing, proxy vendors and endpoints, account identities, browser and session configuration, retry behavior, anomaly-detection rules, manual-review triggers, internal alias maps, and anti-manipulation controls remain confidential.

Cooper also does not publish the complete raw-response archive. These boundaries protect subscription accounts, collection infrastructure, source licensing, and measurement integrity while reducing the ability of a vendor or third party to target the collection process rather than improve genuine recommendation visibility.

The public archive preserves each verified weekly edition, but it does not publish every individual query, complete browser session, or raw answer. Public history is designed to prove what Cooper reported and which methodology governed that report, not to expose the operational collection system.

Every public edition identifies its measurement window, methodology and scoring version, included assistants and category scope, collection completion status, aggregate scores and ranks, available source summary, and any published correction log. Earlier editions retain the rules and results originally associated with them.

Query-level responses, session metadata, extraction records, validation decisions, and internal quality-control logs are retained under access controls for investigation, quality assurance, and audit. Where additional verification is warranted, Cooper may provide relevant records to qualified independent auditors under confidentiality rather than release the complete underlying archive publicly.