Scope, actual collection conditions, definitions, source capture and the boundaries of interpretation.
What was collected
Ten approved US-employer situations; 40 frozen English questions; one primary completed answer per question in ChatGPT, Gemini, Claude and Perplexity. Collection took place on 3-5 November 2026. There were 160 primary texts and 180 recorded attempts, including technical failures and retries. The first completed primary result was retained. Q27 Gemini's completed generic non-answer was not replaced.
The questions were unbranded. No current-name SSSI recommendation was added from a programme name, practitioner, legacy SMS mention or delivery partner. Named providers count once per answer. Explicit aliases were reviewed across questions; corporate parents, university units and unresolved relationships remain distinct.
Actual product and routing conditions
These were consumer interfaces using recorded default modes and dedicated testing logins where available, not a uniform logged-out or API benchmark. Chrome Incognito and product-private modes were used as recorded. The browser route was checked through a US-New York VPN connection. This is a routing condition, not proof that every product treated the account as a US resident.
Q01 Gemini used a guest Flash-Lite/NordVPN condition before later signed-in Temporary Auto sessions. Product labels, errors and changes were logged. Q21's replacement-window attachment was unverified; Q24 Claude mentioned India time despite the New York route. These conditions limit claims about a controlled US-user experiment.
Sources and completeness
All 160 answer texts are saved with hashes. There are 3,170 textual source events. Supporting citations, provider links, retrieved-only entries and entries with unresolved roles remain separate; events can carry more than one role. No answer screenshots are included.
154 source records are complete. Q27, Q28 and Q32 ChatGPT each retain one unresolved original citation URL. Q36 Perplexity retains seven unresolved bundle members; Q37 Gemini retains eight marker mappings with all six sidebar rows saved; Q38 Perplexity retains fourteen member mappings with all 54 panel URLs saved. Source counts may therefore be lower bounds. These exceptions do not change the complete-text recommendation count.
Scope of interpretation
Four formulations are different buyer intents, not repeated identical prompts. Results describe the recorded sample and conditions. They do not establish market share, hidden model reasoning, service quality or guaranteed future visibility.
The SSSI content review uses archived supplied extracts. A fresh browser request was denied by a saved permission policy; reader failures did not establish site-side blocking. Use owner-side inspection for exact-page crawl/index diagnosis. Google guidance supports ordinary technical eligibility and useful content, not a special AI markup or word-count requirement. [12,13]
Research follow-up checks
Primary sources and their limits.
These checks are separate from the sources the products displayed. Publication dates, research periods and inspection notes retain their original meaning.
X01 CCL: influence delivery choices
Current official page; earlier selected primary review
Buyer routes, not independent delivery effectiveness
Repeated annual survey; coverage expanded from about 10,000/7 countries to 16,000/11; publisher states consistent country/eligibility rules for comparisons; associations not causation