Methodology · version 1 · 15 October 2026
Measure what was actually observed.
The pilot uses a fixed set of buyer questions and an evidence log. It does not treat one AI answer as a stable ranking, a universal visibility score, or proof of business impact.
1. Agree the question set
At kickoff, agree 20 questions reflecting product discovery, use cases, comparisons and purchase criteria for the client’s target buyers. Record the exact wording and a prompt-set version before collecting the baseline. Use the same wording for later comparisons, and record any revisions separately.
2. Record the run conditions
Agree which AI assistant surfaces can be accessed legitimately. For each attempt, record:
- Prompt ID, exact prompt, date and time, and run number
- Assistant and surface, visible model/version, language, market and available search settings
- Whether it was a fresh conversation; any known personalisation or account context
- The answer as observed, cited URLs, and evidence reference or capture
- Whether the brand was mentioned, whether its website was cited, and any factual errors
- A completed, blocked, unavailable or unmeasured status, with the reason
Unknown settings are marked unknown. An inaccessible surface is not recorded as zero visibility. No simulated assistant answer is substituted for a real observation.
3. Keep the metrics simple
- Mention rate: completed eligible responses mentioning the brand, divided by all completed eligible responses in the stated comparison set.
- Owned-domain citation rate: completed eligible responses citing the client’s domain, divided by all completed eligible responses in the stated comparison set.
- Coverage: completed observations compared with planned observations. Report blocked and missing runs separately.
Keep nonbranded discovery questions, offer-validation probes and branded accuracy checks in separate groups. Offer-specific and branded questions do not count toward the primary discovery rate. For aeoaeo’s own launch study, the 20-question set contains 13 primary nonbranded questions, three offer-validation probes and four branded accuracy checks. These are hypothesised buyer questions, not verified search-demand data. Show numerator and denominator, prompt-set version, dates and assistant breakdown alongside any rate. Report competitors with the same definitions. Raw observations remain available for review.
4. Improve and repeat
Agree four pages to improve and up to eight hours of technical fixes. Keep dated changes, including what was shipped and what remains a recommendation. Repeat the prompt set at agreed checkpoints under comparable conditions; record any changes in models, search settings or access.
5. Review the limits
Responses vary with wording, date, location, personalisation, model and retrieval behaviour. A before-and-after difference does not by itself establish that a site change caused it. This protocol does not promise statistical significance or a confidence interval from a small set of dependent responses.
Site readiness, assistant observations and business outcomes are separate measures. Traffic, enquiries and revenue require first-party analytics or CRM evidence; unavailable metrics stay unmeasured. No ranking, AI mention or lead outcome is guaranteed.
What you receive
The 30-day pilot includes the 20-question baseline, four agreed page improvements, up to eight hours of agreed technical fixes, and a results review with next steps. The price is ₹75,000 plus applicable taxes.