Every metric in Aiso is derived from a panel of users who agree to share their AI assistant conversations in exchange for model access credits. No scraping, no synthetic prompts, no model-relabeled data.
Aiso operates a consent-based panel of users who share their real conversations with AI assistants. Panelists opt in explicitly and receive model access credits in return. The panel is the foundation of every dataset, benchmark, and visibility metric we publish.
Because we collect the prompt itself rather than reverse-engineer it from logs or paid keyword tools, we can show which prompts drive which AI answers, in the customer's own words.
The Aiso Conversation Database contains more than 10 million real human-AI conversations and grows continuously. Conversations are indexed for both keyword search and semantic similarity, so a brand can find every nearby prompt, not just exact-match ones.
The panel captures conversations across the major consumer-facing AI assistants: ChatGPT (OpenAI), Claude (Anthropic), Perplexity, Gemini (Google), and Microsoft Copilot. Each conversation is tagged with its source so brands can compare how they appear across models, not just on one.
The panel is multi-language and geolocated. We can isolate conversations by country and language, which is what makes market-by-market commercial intent benchmarks possible (for example, comparing how often users in different countries reach transactional intent inside an AI assistant).
Client-facing samples include the conversation hash, the user prompts, the brands mentioned, the inferred intent, the language, and the source assistant. Personally identifying fields, raw timestamps, and account-level identifiers are stripped before any data leaves Aiso.
Most AI-visibility tools either run synthetic prompts against the models or scrape public answer pages. Both approaches measure what the tool generated, not what real customers ask. The Aiso panel measures the actual demand side of AI search, which is the only honest way to tell a brand how visible it really is.
For more on how we run experiments on top of this dataset, see our note on scientific methodology.
Start a free trial or talk to us about an enterprise sample of conversations relevant to your category.