We ask real buyer questions. We save the answers. Then we examine where Eric Strate appears, which pages are cited, and what deserves work.
A prompt is simply the question you type into an AI assistant.
OBSERVETRACE THE EVIDENCEDECIDE WHAT TO TESTAn ongoing study. An open record.
?NEW TO AI SEARCH?The 60-second guide to this pageOpen the guide +
Follow the question. Then follow the evidence.
This is a study of saved AI answers about hiring and comparing services. It helps us decide what to investigate. It does not tell us what every buyer sees.
What is an LLM?
LLM means large language model: the kind of AI system behind assistants such as ChatGPT. An assistant can generate an answer and, in some experiences, use web search. We keep the saved ChatGPT and ChatGPT Search results separate.
Mention or citation?
A mention means the answer names Eric Strate. A citation means the reported source list includes an EricStrate.com page. An answer can do one without the other. Neither is a guaranteed recommendation.
What does “unknown” mean?
Some saved records have missing or inconsistent source information. We cannot call them cited or uncited, so we show them separately and leave them out of the citation percentage.
01 / THE OBSERVATIONS
Less guesswork. More evidence to work with.
Choose the AI experience and questions you want to inspect. Every figure below comes from saved answers in this study. Results can differ for other users, times, and tools.
Source data completenessThis measures usable data, not performance.
02 / THE RECORD OVER TIME
Does the pattern repeat?
Follow the trend, then open a date to see its actual answers.
Eric named in the answerEricStrate.com in the sources
FOLLOW THE SITE’S CONTENT
Pages the answers cite
↗
Open an EricStrate.com page, or select a count to inspect the answers that listed it.
CONTEXT, WITHOUT ENDORSEMENT
Outside sources recorded
≡
These are source addresses reported by the AI. Select a count to inspect the saved evidence.
External addresses are non-clickable references. Inclusion is not a recommendation or endorsement.
03 / FROM EVIDENCE TO ACTION
What deserves attention?
READ THE RECORD FOR YOURSELF
The question. The answer. The supporting sources.
A prompt is the buyer question. The saved answer is what the AI returned. The source list is what it reported citing. Open a record to read it and compare another capture of the same question, AI experience, and market.
These buttons focus this list only. The big-picture totals continue to use the filters above. “Not cited” excludes records with unknown source data.
THE WORK BEHIND THE FINDINGS
What to check. Why it matters. What comes next.
These priorities come from the selected saved answers. They identify data issues and content questions to investigate. A missing mention or citation alone does not prove a page is deficient. Website changes require a page review and a supported reason.
↳
Investigate → review → test → compare. A recommendation is a proposed next step. This page does not automatically edit service pages or treat one changed answer as proof of improvement.
A JOURNAL, NOT JUST A SCOREBOARD
What changed. What we can learn from it.
Weekly field notes connect the answers with possible explanations, source-page investigations, and the next piece of work. Each entry keeps its dates and supporting answer records. Explanations remain hypotheses unless the evidence supports more.
↻
Collect daily. Reflect weekly. A weekly review covers the previous completed Monday–Sunday period. The current short baseline is a starting point, not an improvement claim.
2026-10-15T22:03:44.643Z
The starting point: answers we can inspect.
What happened
We fixed a cohort of 58 commercial prompts and imported 226 recent stored answers. ChatGPT Search: Eric Strate was referenced in 34 of 113 readable answers; ericstrate.com appeared in 22 of 71 answers with usable source capture. 42 records have unknown source capture. ChatGPT: Eric Strate was referenced in 23 of 113 readable answers; ericstrate.com appeared in 10 of 113 answers with usable source capture. 0 records have unknown source capture.
What it might mean
These records establish an inspectable starting point. Prompts have different ages, and responses vary between AI experiences. The results give us specific answers and source pages to investigate; they do not yet demonstrate an improvement over a controlled baseline.
What remains uncertain
This is a recent-window import, not an exhaustive historical export or actual buyer query log. Missing source parsing is kept separate from confirmed uncited answers. Source-list order is not recommendation rank. A citation to a WARP page alone does not establish that the protocol caused it.
The next investigation
Keep the cohort stable during at least two weeks of tuning. Compare repeated captures of the same prompt, AI experience and market; verify how cited pages support the answer; then document the strongest supported findings and unresolved questions.
W
THE IDEA WE’RE INVESTIGATING
Where WARP fits
WARP, the Web Answer Retrieval Protocol, is a structured way to publish a clear answer alongside visible supporting evidence. This lab examines whether saved AI answers align with cited pages. A citation alone does not establish that WARP caused the result.
Where the data comes from: ZeroRank stored responses to a fixed set of questions. These are samples, not real buyer conversation logs or a promise of the same answer for every user. Tracking parameters are removed from displayed source URLs and URLs inside captured text.
What we count: a brand mention is an Eric Strate or EricStrate reference in captured text. It is not automatically a recommendation. A site citation is an EricStrate.com URL in the provider-reported source list. Source-list order is not brand rank. A source list does not identify which sentence or WARP block produced an answer.
How percentages work: matching unique answers divided by unique answers with usable data for that measure. Unknown or inconsistent source capture is excluded from the citation percentage and counted separately. Source bars count answers reporting that URL or domain, including partially captured lists; those bars are not restricted to the citation-rate denominator.
Fair comparisons: AI experiences stay separate. Topics describe the question, not the answering engine. Dates use America/Los_Angeles. Missing dates stay gaps. Historical imports and different run frequencies can affect the sample. The provider’s recent window is capped at 50 records per question.
External references: outside source addresses are plain text with no outbound hyperlink. They are records of reported citations, not endorsements. Links to EricStrate.com remain available for evidence review.
What the study cannot prove: changes can coincide with page edits, different retrieval sources, model changes, or normal answer variation. This observational study does not isolate a cause. Recommendations propose investigations; they do not automatically edit service pages.