01Unit of observation
An observation is a single AI answer, obtained for a specific prompt, from a specific provider and model, for a specific market and answer language, at a specific time. Each observation is stored immutably with the prompt text, model version, timestamp, answer text and the sources the provider returned.
02Valid answers and the denominator
Only valid answers enter the denominator. Calls the provider refused, that timed out, came back empty or errored are invalid; they are counted separately on the results screen.
That's why a provider failing doesn't mean your brand is “not visible”.
03Presence rate
Presence rate = number of valid answers mentioning the target brand / number of valid answers. The result is always shown with its numerator and denominator: “in 8 of 12 valid answers (67%)”. Percentages are rounded to the nearest whole number.
A brand mentioned several times in one answer counts once for that answer.
04Keyword searches have no brand score
If a keyword search has no tracked target brand, we don't produce a “your brand's visibility” number. Instead, we show the brands mentioned in the answers and how often they appear. Once you choose a target brand, measurement switches to brand mode.
05Question set (cohort)
Every scan runs on a versioned, frozen question set. In brand searches, none of the questions contain the brand name; the questions represent category-specific discovery, comparison, alternative, trust and purchase intents.
Free scans use a smaller set, paid scans a larger one. The set size is shown on the results screen.
06Brand matching
Brands are matched on word boundaries using their official name, aliases and domain, accounting for Turkish characters and diacritic variants. Ambiguous names (e.g. Apple) aren't counted without a context term and are flagged for review. You can report incorrect matches.
07Sources and citations
Source links are recorded only when the provider itself returns them; we never generate links. We don't claim a link influenced the answer. The “cited presence rate” is calculated only when the link's relationship to the brand can be verified.
08Competitor comparison
Brands are compared only on the same question set, version, market, answer language, provider set and period. If any of these differ, the comparison is blocked or clearly flagged. Share of voice is calculated within the defined competitor set, counting each brand once per answer.
09API answers vs. app answers
Answers are retrieved from the providers' official APIs with web search enabled. Consumer apps (e.g. ChatGPT, the Gemini app) may show different answers due to personalization, location and session history. What GeoDog measures is a sample taken under defined conditions, not a universal “AI ranking”.
10Trends and monitoring
A trend is shown only when there are at least two comparable scans run with the same method. When the method version changes, the chart flags it; results from different methods are never merged into a single line.
11Limitations
AI answers vary; the same question can produce different answers at different times. Sample size is limited and stated with every result. GeoDog doesn't guarantee rankings; it shows what was observed.