NodpickHow we measure AI visibility
Nodpick asks AI assistants buyer questions, once a week, via each provider's API (directly or through the OpenRouter API gateway) with web search on, and reads Google's AI answers for the same questions. We record which brands each answer names and where yours appears, and turn that into a score from 0 to 100, share of voice and a list of the sources cited. Nodpick is in pre-launch: this page describes the method we run at launch.
1. Questions
For each account we use 5 to 200 questions in the buyer's language (the limit counts across all your brands and depends on your plan), written by you or generated from your category, country and language. Questions never name your brand, so we measure organic mentions.
2. Engines
Every paid plan measures six core engines. The free check asks four of them, once. Grok and Mistral (with web search) and DeepSeek and Kimi (without web search) are included in Agency and available as an add-on on Starter and Growth. We tell each assistant the user's country and language. Results can differ from what a logged-in user sees in an app with personal history.
| Engine | Web search | How we ask | Where |
|---|---|---|---|
| ChatGPT (OpenAI) | Yes | Provider API, directly or via OpenRouter | Free check and every paid plan |
| Perplexity | Yes | Provider API, directly or via OpenRouter | Free check and every paid plan |
| Gemini (Google) | Yes | Provider API, directly or via OpenRouter | Free check and every paid plan |
| Claude (Anthropic) | Yes | Provider API, directly or via OpenRouter | Free check and every paid plan |
| Google AI Overviews | Yes (Google Search) | Logged-out Google search | Every paid plan |
| Google AI Mode | Yes (Google Search) | Logged-out Google search | Every paid plan |
| Grok (xAI) | Yes | Provider API, directly or via OpenRouter | Agency, or add-on |
| Mistral | Yes (by default through OpenRouter with Exa web search; with a direct key, Mistral's own search) | Provider API, directly or via OpenRouter | Agency, or add-on |
| DeepSeek | No, model knowledge only | Provider API, directly or via OpenRouter | Agency, or add-on |
| Kimi (Moonshot AI) | No, model knowledge only | Provider API, directly or via OpenRouter | Agency, or add-on |
DeepSeek and Kimi answer without web search, and they run as the open models on hosting providers based in the USA, not in the vendors' own apps. Their answers come from model knowledge, so they can be out of date and cannot cite sources. Read them as a view of what the model has learned about your brand, not of what it finds today.
3. Google AI Overviews and AI Mode
We do not ask Google a chat question. We run the same question as a logged-out Google search in your project's country and language (desktop by default) through DataForSEO, and read the AI Overview or the AI Mode answer, with its sources, plus the organic results for your domain's position.
Google shows an AI Overview only for some queries. Questions without one are not counted in your score, because that would punish you for something Google did not show. Instead, the project shows the coverage: for how many of your questions Google showed an AI Overview. AI Mode answers are measured the same way.
4. Weekly runs and confirmation
AI answers vary between runs. Paid plans ask each question once a week per engine. If a brand's mention flips compared with the previous week, we ask again right away and count both answers. The free check asks once. Failed or timed-out answers are excluded and reported.
5. Analysis
From every answer we extract the brands named, in order. Your brand counts as mentioned when its name, an alias or your domain appears as a whole word (case, accents and German umlauts are normalized).
6. Scoring
| Your position in the answer | Points |
|---|---|
| 1st brand named | 100 |
| 2nd | 80 |
| 3rd | 60 |
| 4th or later | 40 |
| Not named | 0 |
The visibility score is the average over all answers. Share of voice is your mentions divided by all brand mentions. Cited sources are the domains the assistants linked to.
7. Search Console import
In paid projects you can upload a Google Search Console export of your AI impressions. These are real impressions from real searches, not questions we asked. Google reports AI Overviews and AI Mode combined, so we show them combined. We show them next to your measured results and do not mix them into the score. Only the aggregated numbers in the file you upload are stored.
8. Recommendations
Recommendations combine your results (missing questions, dominant competitors, citation gaps, assistants that never name you) with a scan of your homepage and /llms.txt for structured data, FAQ and comparison content.
Limits
- We measure what the engines answered at a point in time, not a guarantee of future answers.
- API answers can differ from consumer apps (memory, personalization, model versions), and a logged-out Google search differs from what a signed-in person sees.
- We do not measure Microsoft Copilot, voice assistants or featured snippets.
- Brand extraction can miss unusual names; add aliases in your project settings.