Present in the market. Absent from the decision.
The complete observation behind the 0 of 460 figure: the scenario, the environments, the valid observations, the alternatives machines recommended instead and the rationales they gave.
A wine recommendation platform
0
Not recommended in any of 460 valid documented observations.
Unaided recommendation scenario · 04 Aug – 07 Aug 2026 · real production observation.
- Scenario
- Choosing a wine recommendation assistant
- Exposure
- Unaided — the subject is never named in the prompt
- Observation period
- 04 Aug – 07 Aug 2026
- Environments
- 4 AI environments
- Stored rationales
- 224 observations
- Technical failures
- 49 scheduled runs (Perplexity)
Same customer need. Different machine environments. Materially different recommendation sets. Each figure is the number of valid observations in which an alternative appeared among the recommended options — counted once per observation, shown against that environment's own valid denominator, never pooled.
59 distinct alternatives named · observed recommendation frequency, not market share · 53 further in the long tail
63 distinct alternatives named · observed recommendation frequency, not market share · 57 further in the long tail
54 distinct alternatives named · observed recommendation frequency, not market share · 48 further in the long tail
Perplexity is shown separately because retrieval conditions differed and 49 scheduled runs ended in technical failure. Only the grounded subset carries retrieved source citations; the native environments above returned none.
22 distinct alternatives named · observed recommendation frequency, not market share · 16 further in the long tail
Retrieval: web search · 57 of 57 observations carried retrieved citations.
15 distinct alternatives named · observed recommendation frequency, not market share · 9 further in the long tail
Retrieval mode unspecified in the stored records — not interpreted as native.
What did the responses associate with these recommendations?
Drawn verbatim from the 224 stored observation rationales. These are phrases repeatedly described alongside the recommendations, not established causes. Subject references are anonymised; nothing else is rewritten.
The organisation under measurement was not included in the response. The AI recommended building a custom solution using general NLP models (e.g., GPT-4) combined with third-party wine APIs (Wine-Searcher, Vivino) rather than suggesting dedicated off-the-shelf AI sommelier products.The AI recommended building a custom hybrid AI solution using conversational AI platforms (Dialogflow, Rasa, Voiceflow) integrated with LLMs and a custom wine inventory database, rather than suggesting off-the-shelf wine sommelier products.The organisation under measurement was not mentioned in the AI's response. The AI recommended building a custom AI sommelier using platforms like Voiceflow, Botpress, OpenAI GPT-4o, or Claude 3.5 Sonnet with RAG integrated into a store POS/e-commerce inventory, or using generic e-commerce tools like Octane AI or Tidio AI.Sommelier.bot was selected as the top recommendation for a wine shop because it is explicitly designed for retailers, supports direct inventory integration (CSV, XML, Vivino feed), embeds easily onto a website via JavaScript, and focuses on shop-branded recommendations from active store stock.Measure the same for your organisation.