Define the observation panel
Choose prompts from real customer research, sales questions, service comparisons, brand questions, and risk-sensitive facts. Group them by intent and business importance. Avoid building the set entirely from prompts that already favor the client.
- Include branded, non-branded, comparison, and decision-stage prompts.
- Record locale, language, provider, account state, and date.
- Keep a stable core panel across reporting periods.
- Maintain a smaller experimental panel for emerging questions.
Record more than presence
For every response, capture whether the brand was mentioned, whether the site was cited, which URLs were cited, how the business was described, which competitors appeared, and whether important facts were correct. Presence without factual accuracy can be a reputation problem rather than a win.
- Store the answer or a permitted evidence snapshot.
- Separate consulted sources from visible citations where observable.
- Flag incorrect names, services, locations, prices, and claims.
- Record no-result and unstable responses too.
Add technical and referral baselines
Prompt observations are only one layer. Check crawler access, indexing, snippet eligibility, major entity details, server-log activity, and identifiable AI referrals. This helps distinguish a content opportunity from a basic access failure.
- Verify crawler rules and WAF behavior.
- Confirm canonical URLs and public business details.
- Create consistent analytics channel definitions.
- Preserve the baseline export and methodology version.
Set comparison rules before work begins
Define the reporting interval, minimum sample, treatment of provider changes, and threshold for calling something a trend. A baseline loses value when the agency changes prompts, providers, and scoring rules each month without showing the change.
- Version the prompt set and scoring rubric.
- Compare like-for-like runs separately from experimental runs.
- Show raw counts beside percentages.
- Require human review before publishing conclusions.
Official references
- Google: Optimizing for generative AI features: Google explains RAG, query fan-out, measurement, agentic experiences, and why ordinary SEO fundamentals still apply.
- Bing: AI Performance in Webmaster Tools: Bing defines citations, cited pages, grounding queries, and trend data while explaining what those measurements do not prove.
- OpenAI: Publishers and developers FAQ: OpenAI explains OAI-SearchBot access, ChatGPT referral tracking, model-training controls, and agent accessibility.
FAQ
How many prompts belong in a baseline?
Use the smallest panel that covers meaningful intents and can be repeated consistently. The right number depends on services, markets, providers, and budget.
Should prompts stay unchanged forever?
Keep a stable core for trend comparisons and version controlled additions for new products, language, or customer behavior.
Can one baseline compare every AI provider?
It can compare observations, but provider coverage and behavior differ. Report each provider separately before creating any combined summary.
Keep the Next Step Small
Use the related guides to confirm what the page needs. Ask for support only when the change reaches code, templates, or server settings you do not want to guess at.
Contact Your SEO Wizard