Check your company's AI visibility
Free, no credit card required.
Three environments, three measurements
Gemini is a conversational application, AI Mode is a Google Search experience, and AI Overviews are generated summaries that appear for some searches. The model name does not define the entire product: the interface, available context and information retrieval affect the result. Create three separate series in your report. A Gemini result cannot substitute for an AI Mode measurement.
Why does the brand appear on Google but is not recommended by AI?
Why Google may show different brands
Google Search Central explains that AI Mode and AI Overviews may break a question into related searches. They also use different models and techniques, so answers and links may differ. Gemini may show related sources, but not every answer includes links. Conversation context and attached materials further change the conditions. This explains why differences are possible; it does not establish the cause of a particular brand's decline.
Define what visibility means
A mention is an occurrence of the recognised brand name. A recommendation presents it as an option meeting the user's need. A citation is a link to a particular source. Record these events separately: a cited domain need not be a recommended provider, and a recommended brand may have no link. Open the source and check whether it supports the attributed information. Check that names and links refer to the correct company and its current offering.
AI citation audit: do sources really support information about your company?
Watch the denominator in AI Overviews
Separate the absence of an AI Overview from an overview without the brand. Hypothetical example: an overview appears in 30 of 100 checked queries, and the brand appears five times. Presence across the entire sample is 5/100, or 5%; conditional presence in displayed overviews is 5/30, or approximately 16.7%. These figures answer different questions. Report sample counts and label failed or incomplete measurements separately instead of counting them as absent visibility.
100 queries · 30 AI overviews · 5 with the brand. 5 / 100 = 5%; 5 / 30 ≈ 16.7%. Hypothetical data, not Semly results or a market benchmark.
Build a comparable sample
Use the same intents and reasonably equivalent questions in each environment. Record the date, language, country, model or mode, settings and conversation context you can control. Separate branded questions from category questions. Run successive series under comparable conditions and retain complete answers. One result is an observation; a trend requires repetitions. If a tool measures an API rather than the consumer interface, state that scope in the report.
AI monitoring budget: choosing questions, systems and review frequency
What to improve after finding a difference
First examine specific evidence: inaccurate offering information, a missing page addressing the intent or weak source coverage. To appear as a supporting link in Google Search AI features, a page must be indexed and eligible to display a snippet. Google does not require a special markup or separate AI file. Improve accessibility and existing content according to the diagnosis, then repeat the measurement. Indexing does not guarantee a citation or recommendation.
Reading Semly results and FAQ
In Semly, analyse models and available environments separately alongside prompts, competitors and sources; coverage depends on configuration and plan. Does a better Gemini result imply an improvement in Google Search? You cannot assume that. Is one combined average enough? It can serve as a secondary indicator if you show its composition and weights, but cannot replace three separate results. Does every gap require a new article? First check whether updating an existing page would solve the problem.
Blog and knowledge base without cannibalisation: how to divide topics for SEO and AI
Sources and methodology
Share:
