Methodology
What the data is, and where it comes from.
A monitoring product is only as good as your understanding of what it measures. This page documents what Pineprompt tracks, how often, what is stored, how each metric is computed, and where the data stops.
What is tracked
Every Pineprompt project has one or more prompts, the exact buyer queries you care about. Every monitor is scoped to a country (ISO 3166-1 alpha-2) and a language. Each prompt runs against the subset of these eight engines you enable on the monitor.
Each engine answer is captured under the configuration its production users see. Pineprompt does not strip system prompts or disable web search, so the answers you analyze match what a buyer would read.
Refresh cadence
Every monitor refreshes daily, on every plan. The cadence does not change with your tier. AI answers are stochastic: the same prompt returns subtly different answers each time. A daily cadence gives enough density to surface a trend while keeping noise manageable. Pineprompt does not advertise a "real-time" cadence or an interval it cannot hold.
What is captured per response
For each run of a prompt against an engine, Pineprompt captures the following.
- The answer text, stored verbatim for audit.
- Citations, the URLs the AI cited and the sources it scraped, with domain and position in the answer.
- Brand mentions, extracted brand entities matched against your registered brands and competitor list. Unmatched mentions become candidates for competitor discovery.
- Sentiment, each mention scored for tone.
- Engine, country, language, and timestamp.
How this captured data is stored, who it is shared with, and how long it is kept is set out in our privacy policy.
How metrics are computed
- Mention rate: the share of responses that name your brand at least once, over the window and engine set you select.
- Position: your brand's average place in the order an engine names brands, across the responses that name it. Position one means named first, and lower is better.
- Sentiment: each mention is scored by an analysis model. Aggregates are averaged across the mentions in the window.
- Citation share: the share of cited URLs that belong to your domain, over the window.
- Share of voice: mention rate across a tracked prompt set, aggregated by engine. See share of voice in AI.
The limits of the data
- AI answers are non-deterministic. Mention rate fluctuates from day to day without any real-world change. Trend over multi-day windows for reliable signal.
- Sentiment scoring uses an analysis model. It is not perfect, and a small share of mentions may be mis-scored. Spot-check by reading the underlying response content, which is stored verbatim.
- Brand matching has a threshold. Very close spellings and aliases are matched aggressively, which can occasionally group two similarly named brands. Register aliases explicitly to disambiguate.
- Engine behavior changes. When a model provider ships an update, answers can shift overnight. Pineprompt does not smooth the shifts away; you see them as they happen.
- No pre-signup history. Data begins when the monitor is created. There is no archive to back-fill from.
Start tracking.
Every monitor records every mention, citation, and sentiment score across all eight engines, daily, in the country and language you choose.