Agree the question panel before sampling
We begin with the buyer decisions relevant to your offer: comparisons, suitability, service scope, specifications or delivery. The panel is agreed with your business and tied to one target country and English, unless your package explicitly covers more markets.
Starter, Growth and Scale track 15, 30 and 60 questions respectively. The single-market Audit and Monitor use ten questions. The International Audit uses ten questions and three competitors per country across three supported countries. Question counts are service limits, not search-volume or market-demand estimates.
Record a reproducible observation
- Record the exact question, run identifier, date/time and search experience.
- Record the language, requested target country and actual geography or simulation method. State limitations and do not present unsupported country simulation as a local result.
- Save the available answer and source links, noting unavailable or incomplete evidence.
- Record mentions, citations, recommendations, factual accuracy and competitor presence separately.
- Keep the panel and method comparable between dates; document changes instead of silently changing the denominator.
The standard surfaces are ChatGPT Search, Google AI Overviews and Perplexity. Scale adds Microsoft Copilot. Samples refer to those named experiences, not every model or conversation on each platform.
Keep the metrics and denominators explicit
| Metric | What it means | Limit |
|---|---|---|
| Brand mention | The answer explicitly names the brand. | A mention can be neutral or negative; it is not automatically a recommendation. |
| Own-domain citation | An available answer links to the client’s own website. | Third-party mentions are tracked separately; a citation is not a click. |
| Recommendation | The answer recommends the offer in the recorded context. | The context and supporting text must be retained. |
| Factual accuracy | The answer matches approved client facts. | Requires a client-approved source of truth. |
| Sampled mention rate | Valid sampled answers mentioning the brand ÷ valid sampled answers × 100. | Report the numerator, denominator and exclusions for each platform. |
For Google, first record whether an AI Overview appeared. “No Overview” is different from an Overview with no brand mention. Report the Overview appearance count as well as the brand result; do not silently treat unavailable answers as zero visibility or combine unequal platform samples into an unexplained score.
What a report row contains
The following is an unfilled report template, not a live AI result or a client case study. Download the observation CSV template to inspect the fields.
| Question | Platform | Country method | Answer / evidence | Mention / citation |
|---|---|---|---|---|
| Which AEO service package fits a small business? | To be recorded | To be recorded | Not sampled | Not measured |
Reporting cadence and business attribution
Managed plans and the Monitor use weekly samples and monthly reporting. The one-time Audit and International Audit use two sample dates. Managed reports also show delivered work, access or approval blockers and next priorities.
Referral visits and conversions add context when suitable analytics access exists. They are different from mentions and citations, and do not establish that a content edit caused a sale. Search Console’s Web performance includes Google AI search traffic without a standalone AI Overview revenue measure. Bing’s AI Performance reporting supplies citation observations for supported experiences when data is available.
AI answers can vary with wording, time, location and platform changes. A sampled panel is not a census of every answer or the total market. Historical results do not guarantee future mentions or sales.
Primary platform references
- Google: AI features and your website — eligibility, content and reporting guidance.
- OpenAI: crawler documentation — search and training controls are separate.
- Perplexity: crawler documentation — search crawler and user-requested fetches.
- Bing: AI Performance reporting — citation activity and its interpretation.
References reviewed on 16 October 2026. Platform documentation can change. Our method records the actual experience and limitations used in each report.