Comparing actions with visibility changes
Monitoring can show what changed in sampled AI answers after work was completed. It cannot, by itself, establish that the work caused the change or produced a revenue return.
Choose review checkpoints
Use a baseline and repeatable review schedule. Days 30, 60, and 90 are possible planning checkpoints, not predictions about when an AI system will retrieve your content.
| Check | Record |
|---|---|
| Baseline | Date, prompts, engines, coverage, and sampled answers |
| Work completed | What changed, where, and when it was verified |
| Follow-up | The same prompt and engine scope where possible |
| Limitations | Failed samples, changed prompts, and other events |
AI products can use different retrieval modes and sources. Do not assign a fixed leading or lagging role to an engine, assume a training update will improve mentions, or treat indexing as proof of retrieval.
Investigate changes
Compare actual answers and cited URLs. Ask whether a new source appeared, a source disappeared, wording changed, or the sample coverage changed. New content, entity corrections, reviews, and earned media are possible context to investigate rather than proven causes of a score increase.
A decline can also reflect sampling variability, changed prompts, engine behavior, source availability, or competitor activity. Keep these explanations open until the evidence supports a narrower conclusion.
Keep a defensible work record
1. record verified actions
Keep dates and URLs for published work and note corrections made in customer-owned systems. Check which actions appear in the workspace history and supplement it for manual work; do not assume every external action is logged automatically.
2. compare like-for-like samples
Hold prompts, engines, geography, and evaluation rules consistent. Read source-level evidence rather than only the headline Share of Model number.
3. state the limits
Say "mentions increased after this work" when that is what the data shows. Simultaneous press coverage or a product launch may also matter. An observational timeline cannot tell you how much each action contributed.
4. treat comparison groups carefully
Untreated keyword groups can help you explore alternative explanations, but different keywords may have different audiences and source patterns. An informal comparison is not automatically a controlled experiment.
A practical review rhythm
- Review successful sample counts and answer quality before interpreting a trend.
- Record completed work and other relevant events.
- Investigate material changes in the actual answers and sources.
- Choose the next useful action and a review checkpoint.
In Outrigger
Review Sonar results and the work history available in your workspace. Keep additional evidence for external actions. Use the record to explain observations and decide what to investigate next.