How NodellGraph measures AI answers
Last updated October 29, 2026
The engines
- Your prompts are asked of OpenAI, Anthropic, and Perplexity by default, through each provider's official API. We run no scrapers or headless browsers.
- Gemini is optional: check it on a prompt to include it. Each extra engine adds answers to every run, so a prompt on 4 engines counts for more of your quota (how your prompt quota works). Add it where your customers use it. Gemini is Google's AI assistant, not the AI answers shown in Google Search, which are measured separately.
- Google AI Overviews and AI Mode are measured from Google search data via a third-party provider, searched from your project's location; they're on by default when your project has one, and don't count toward your quota (how Google's AI answers are measured).
- The models are the ones people actually use, and every answer is stored with the exact model version that produced it. The versions for each run are listed under the engine scores.
How answers are analyzed
A separate AI model reads each answer and records which tracked brands it names, in what order, in what tone, and which sources it links to. That analysis model is never one of the engines being measured.
When runs happen
- Projects are measured weekly by default. Prompts you flag for daily tracking (a paid add-on) are also measured every day.
- Your first run starts on its own once your project has a plan, prompts, and competitors.
- Run now starts a full run on demand — after you change prompts or competitors, or to retry a run that failed. Your organization gets on-demand runs worth 4 full runs of your plan per billing period (4 during a free trial, 9 during a design partner's free month), and a project can run at most 3 times a day. Failed runs don't count. Answers rarely change within days, so the weekly schedule is what shows real movement.
- Runs go through each provider's batch processing, which is why results can take from minutes to a few hours.
- If an engine fails during a run, the run is marked incomplete and that engine is left out, never shown as a zero.
What's kept
Full answer text is deleted after 90 days. The mentions, ranks, sentiment, and citations extracted from it are kept, so your trend history stays intact.