SAG / ARCHITECTURE NOTE
Evidence and Provenance Tracking: A Structure for Turning AI Answers into Verifiable Results
Learn how to create reproducible results by linking questions, observation conditions, sources, reviews, and approval history instead of merely saving the text of AI answers.
Record observation conditions before the answer
AI answers can vary depending on the service, model, time, language, and account status. Therefore, saving only a sentence shown on screen makes it difficult to explain the result later. To make the data comparable, record the exact question, observation channel, execution time, target country and language, and input conditions used.
SAG separates observed results from interpretation. The exact text and links actually confirmed are recorded as observations, while inadequate explanations or priorities for improvement are identified as analytical opinions. Values that could not be confirmed are preserved as unobserved rather than changed to 0.
Provenance makes answers trustworthy
Provenance is information that makes it possible to trace where a result came from. The following links must be maintained.
| Stage | Information to record |
|---|---|
| Question | Original wording, language, customer intent |
| Observation | Channel, time, execution conditions |
| Evidence | Source URL, page title, section reviewed |
| Assessment | Criteria for sufficient or insufficient, analyst's opinion |
| Action | Target of the change, owner, approval status |
| Recheck | Before-and-after differences under the same conditions |
With these links in place, you can answer the question “Why was this improvement proposed?” by returning to the original evidence.
The roles of automation and expert review
Automation is useful for collecting and comparing many questions and pages in a consistent format. Experts review the service context, regulatory wording, and aspects that customers may actually find difficult to understand. SAG is designed to route automated collection results through review and approval statuses before delivery, rather than treating them immediately as a final report.
Reports that explain before-and-after changes
A good report explains changes rather than presenting a single score. It shows which text was revised, what evidence was strengthened, and how the answers and sources changed for the same question. Evidence and provenance tracking is not a technology that makes AI answers absolute facts; it is a structure for verifying observed results and improving them responsibly.
Technical references by topic
How to continue reading about this technology
Read about the problems that SEO, AEO, GEO, entities, and JSON-LD each address.
SAG / KNOWLEDGE LINKS