How SillyDino Measures AI Search Visibility

AI search visibility is not one metric

A server log showing that OAI-SearchBot requested a page is not the same as a ChatGPT citation. A citation is not the same as a recommendation. A recommendation is not the same as a website visit, and a visit is not automatically a sale.

Combining those signals into a phrase such as “AI references” can make a result sound stronger than the underlying evidence. SillyDino reports each layer separately.

How SillyDino builds an AI visibility baseline

Measurement begins before pages are rewritten. Otherwise, it becomes difficult to distinguish a genuine change from normal answer variation.

1. Define the business entity. Record the correct brand name, services, products, markets, locations, official profiles, and preferred website URLs.
2. Map real customer questions. Build a fixed set of discovery, comparison, local, service, product, trust, and buying-intent prompts.
3. Check technical access. Review robots.txt, server responses, CDN rules, redirects, canonical URLs, and whether relevant crawlers can retrieve important pages.
4. Capture the baseline. Record current answers, cited sources, missing information, competitors, wording, platform, date, language, and location context.
5. Connect analytics. Confirm that referral sources, forms, calls, purchases, and other useful actions can be measured.
6. Improve the weak signals. Work may involve SEO, clearer service pages, useful original content, internal links, entity consistency, reviews, policies, or stronger public evidence.
7. Retest the same panel. Compare like with like while keeping exploratory prompts separate from the fixed measurement set.

This process complements the broader explanation of what AI search visibility means. It focuses specifically on evidence and reporting, not on a general AEO checklist.

How controlled prompt testing works

Prompt testing is useful, but it is not a traditional rank tracker. AI answers can vary by wording, product mode, account, location, language, available sources, and date.

For every test that becomes evidence, SillyDino records:

• The full prompt without shortening or rewriting it afterward.
• The platform and answer mode used.
• The date, time, language, and relevant location context.
• Whether the session was signed in or otherwise personalized.
• The complete answer or an uncropped evidence capture.
• The visible source URLs and the pages they support.
• The brand’s position and wording, without converting it into a permanent “ranking”.
• Known limitations or conditions that could affect the result.

A fixed prompt panel helps identify directional change. Exploratory prompts help discover new questions and competitors, but they should not be mixed into a before-and-after score only because they produced a favorable answer.

The evidence log behind a SillyDino report

How we describe proof without inflating it

Precise labels make evidence more credible. For example:

• Use “OAI-SearchBot requests recorded in Cloudflare during a 24-hour period,” not “AI recommendations”.
• Use “Google Search impressions,” not “organic reach” unless reach is the platform’s actual metric.
• Use “shown as a source in the captured answer,” not “ranked permanently in ChatGPT”.
• Use “purchases attributed by analytics to chatgpt.com,” not “sales caused entirely by AEO”.

The TopRatedPlaces.ai case study contains Search Console growth and OpenAI-related request activity. Those are two separate signals. The Botanica case study contains search, local discovery, AI-answer captures, and purchases attributed to chatgpt.com. Each source is interpreted within its own limits.

Leading signals and business outcomes

Early work may first improve crawlability, indexation, source coverage, entity clarity, and answer accuracy. These are leading signals. The commercial objective remains qualified action, such as an enquiry, booking, purchase, or call.

SillyDino therefore reports both:

• Visibility evidence: crawler access, indexed pages, citations, answer appearances, AI feature impressions, and referral sessions.
• Business evidence: qualified leads, purchases, bookings, conversion rate, revenue quality, and assisted customer journeys where tracking allows it.

A visibility increase without useful commercial action may still identify progress, but it is not presented as a completed business result.

Important limitations

• No agency controls the final answers produced by ChatGPT, Google AI, Gemini, Perplexity, Claude, or another AI platform.
• OAI-SearchBot activity is not proof that a page was cited in ChatGPT Search.
• ChatGPT-User represents user-triggered retrieval and should not be combined with SearchBot requests under one vague metric.
• Prompt captures are time-specific observations, not permanent rankings.
• Analytics attribution depends on consent, browser behavior, referral handling, platform behavior, and the chosen attribution model.
• Search impressions, citations, visits, and conversions must remain separate in reporting.

This approach follows current guidance from Google, OpenAI, and Bing. Their reporting tools and crawler definitions may evolve, so methodology pages and reports should retain dates and source links.