AI Visibility Tracking Panel Design and Baseline Protocol

Design a repeatable panel of prompts, engines, and scoring rules to measure brand mentions and citations in AI answers. Produces a baseline protocol, a scoring sheet, and a decision rule for reporting changes.

You getStatus and assumptions, prompt panel table and execution protocol checklist

The prompt

Prompt
Role: You are an AI SEO measurement analyst who designs defensible visibility tracking for answer engines.

Context: Cedar Ridge Outdoor Gear (example value for {brand}) wants to track its presence in AI answers for backpacking tents and sleeping bags (example value for {topic}) against REI, Backcountry, Moosejaw (example value for {competitors}), in Canada, English (example value for {market}). The decision this supports: whether leadership should continue funding AI SEO work based on measured movement over 90 days (example value for {period}).

Task: Design a tracking panel and baseline protocol. If this Gemini session cannot browse or query other engines, say so and provide the protocol for humans to execute; do not report invented results.

Inputs:
- Brand and competitors: Cedar Ridge Outdoor Gear (example value for {brand}), REI, Backcountry, Moosejaw (example value for {competitors})
- Topic and market: backpacking tents and sleeping bags (example value for {topic}), Canada, English (example value for {market})
- Business-critical intents (informational, comparison, transactional, brand): informational, comparison, transactional, brand (example value for {intents})
- Tools and engines available to the team: Gemini, Google AI Overviews, ChatGPT search, Perplexity (example value for {engines})
- Reporting period: 90 days (example value for {period})
- Any existing observations (pasted), with dates.

Workflow:
1. State browsing status and assumptions.
2. Panel design: propose 25-40 prompts across the intents, phrased as real users would ask, with variants in wording. Tag each by intent and funnel stage. Include brand-neutral prompts, which matter most for discovery.
3. Protocol: define sessions (logged out, fixed location and language, fresh session), number of repeats per prompt (recommend at least 3 because answers vary), time of day windows, and what to record verbatim.
4. Scoring rules: Mention (named), Citation (linked source from the brand's domain), Position (first, middle, last, or not present), Sentiment (positive, neutral, negative, with a quoted justification), Accuracy (correct, partly, wrong against known facts), Competitor presence.
5. Source discovery: list searches and checks to find which third-party pages are cited when the brand is absent.
6. Statistics guidance: how to report share of answers with mention or citation with sample sizes, and a threshold rule for calling a change real versus noise. State clearly that thresholds are suggestions, not proven standards.
7. Separate Evidence, Inference, Contradictions, and Unanswered questions for any existing observations.

Constraints:
- No made-up baseline numbers.
- Do not treat a single observation as a trend.
- Note that engines change models, retrieval, and UI, which can break comparability; require logging of date and engine or model version when shown.
- Avoid collecting private or personal data in prompts.

Output format:
1. Status and assumptions
2. Prompt panel table: ID | Prompt | Intent | Funnel stage | Variant of
3. Execution protocol checklist
4. Scoring sheet template: Date | Engine | Prompt ID | Run # | Mention | Citation | Position | Sentiment | Accuracy | Competitors present | Cited URLs | Notes
5. Reporting template: metric definitions, sample size display, change threshold rule, caveats
6. Evidence / Inference / Contradictions / Unanswered questions (if observations provided)
7. Risks and next actions with a week-by-week plan for the first 90 days (example value for {period})

Acceptance criteria:
- Panel covers every intent in informational, comparison, transactional, brand (example value for {intents}) with at least 3 prompts each.
- Protocol is specific enough for two people to produce comparable results.
- Scoring definitions are unambiguous and include examples.
- The reporting rule states how to answer the funding decision with uncertainty.

Quality checks:
- Confirm no fabricated results.
- Confirm repeats and sample sizes are built into the reporting.
- Confirm brand-neutral prompts outnumber brand-name prompts.
- Confirm limitations of volatile AI answers are stated.

Dashed parts are example values, so the prompt runs as copied. Change them in Fill in the blanks.

Lock the panel after the baseline run so later comparisons stay valid, and add new prompts in a separate set. Use a spreadsheet for the scoring sheet and share anonymized exports with Gemini for analysis.

Keep exploring

More in AI SEO