VisibilityOS Citation Labeling Reference
Use this reference when two people need to score the same AI answer the same way. Record whether the answer names the correct product and whether it links to an official site. Save the evidence for each decision.
Agree on the eligible websites
Write down the product name, approved aliases, and the official domains you will count before the first run. When a source redirects, record the citation as shown and the final destination. Keep the domain list and rules consistent across the answers you compare.
Label two fields on each complete answer
- Brand mentioned. Mark yes when the answer names the product and refers to the correct company. Record the exact sentence. A same-name company in a different category does not qualify.
- Official link present. Mark yes when a link in the answer or source panel resolves to an approved official destination. Record the link and its final address.
- Unresolved evidence. If the source panel cannot be opened or the company identity remains ambiguous, preserve the original answer and record what could not be verified. Exclude incomplete pairs from reporting and show that count separately.
Worked examples
- Correct name and a directory article. Brand mentioned: yes. Official link: no. The directory article describes the correct company but the answer only links to that third-party page.
- Official link in the source panel without a product name in the answer. Brand mentioned: no. Official link: yes. Save the source panel URL so another person can reproduce the decision.
- Same name, different business. Brand mentioned: no. Official link: no. Save the answer sentence and the source that identifies the other business.
- Answer produced, source panel failed to load. Save the answer and record incomplete link evidence. Check again under the same conditions before assigning the final label.
Keep the evidence with the labels
For each row, keep the exact question, assistant and model, search setting, date, answer excerpt, citation and final URL. When you compare two runs, use the same question and settings for both. The original VisibilityOS study separately tracked brand presence and link changes across 817 matched comparisons.
Read the original VisibilityOS study · Open the public audit template · Review the same-name company worked example