1. What we measure
In this notice, "AI Apps" means AI assistants and AI search features, such as ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Perplexity and Grok (xAI), in any form, including their consumer apps and websites. Namova measures AI Apps only through each provider's own API, directly or through an AI routing service, with web search turned on where we use it, and not through their consumer apps.
We send questions that a business's buyers might ask to AI Apps, through their providers' APIs as described above. Every result shows which provider and model it came from. We record the full answer, the brands it names and the sources it cites. We do this many times, across AI Apps, and show how often each brand is named.
2. What we do not measure
- We do not see what people see in consumer apps. ChatGPT, Claude, Gemini, Perplexity, Grok and others can give a different answer in their own apps and websites because of different models, settings, search tools, location, language, account history, personalisation and timing. Our numbers describe what the APIs returned to us.
- Web search may differ too. Where we turn on web search, the search results come through the provider's API or a separate search provider. For Gemini, they come from a separate search provider, not from Google Search.
- We do not see any real buyer's conversation. We do not know what any particular person asked or was told.
- We do not measure sales. If you connect Google Analytics (coming soon), we can show visits that analytics labels as coming from AI assistants. Those labels are incomplete, and many AI-influenced visits are not labelled at all.
3. Why answers vary
AI Apps are not fixed lists. The same question can get a different answer a minute later. Providers update models, search tools and sources without notice. So one answer, or one screenshot, proves very little. That is why we ask many times.
4. How to read our numbers
- Ranges, not single numbers. We show results as a range (for example "named in 54 to 83 percent of answers") worked out from the sample. The true rate could still fall outside the range.
- Labelled. Every number says which provider and model it came from, when, and how many answers it is based on.
- Failed answers are left out. If a provider's API fails, refuses or times out, we leave that answer out of the score and say how many were left out. We never count a failed answer as "not named".
- Model changes are marked. When a provider changes a model, results before and after may not compare directly, and we mark the change.
- Snapshots are small. The website snapshot asks 6 buyer questions, once each, on 3 AI Apps (OpenAI, Claude and Perplexity). It is a first look, not a measurement you should make decisions on.
- Names are matched by software. We detect brand names automatically. Unusual spellings, similar names or brands that share a word can be missed or miscounted. We check and improve this, but it is not perfect.
5. Fixes are suggestions
Fixes are drafted from what the answers and their cited sources show, partly with the help of AI. They can be wrong or unsuitable for your business. You decide what to publish, and you are responsible for checking that it is accurate, lawful and allowed by the platforms you publish on.
6. No guarantee of results
We cannot control AI Apps, search engines or other websites. We do not promise that any AI App will name, rank or recommend you, or that you will get more visibility, traffic, leads or sales. If results improve after a fix, we report what changed and what we checked, but a change after a fix does not prove the fix caused it.
7. Independence
Namova is independent. We are not affiliated with, endorsed by or sponsored by OpenAI, Anthropic, Google, Perplexity, xAI or any other AI provider. We name them and their AI Apps only to say where an answer came from.
8. Questions
If a number looks wrong, tell us at [email protected] and we will show you the answers behind it. The full legal terms are in our Terms of service.