Generated from your ICP
Stop writing the prompts you already win. Define your audience once and prompts are generated from those personas and from observed search demand, then suggested in bulk for you to edit or reject.
Be Chosen, the fifth rung
Track the prompts that hold still
AI answers vary between runs. A prompt whose brand list changes every probe measures noise, not visibility. Each carries a stability score so you know which is which.
Key takeaway
reproducibility.py computes the median pairwise Jaccard similarity across a prompt's recent brand lists. Below the floor, the prompt is flagged rather than charted, because a line drawn through noise looks exactly like a line drawn through a result.Prompts you did not have to invent
Guessing at prompts produces a list that flatters the brand. These come from your audience and from real demand.
Four question types
Fix the buying stage you are losing. Brand direct, category, comparison and pain point map to different stages, fail for different reasons and are scored separately rather than averaged into one number.
A reproducibility score per prompt
Know whether a trend is real before you act on it. The score is the median pairwise similarity of the brand lists returned across recent probes, so low means the answer is unstable and any trend drawn from it is noise.
A quality score per prompt
See which prompts are worth keeping. 0 to 1 across four axes: how well it discriminates you from competitors, how stable its mention rate is, whether it connects to a decision anyone acted on, and whether it has been probed recently.
How a prompt is run and read
Every number on a prompt traces to the probes behind it, and the responses are kept.
Two modes on the same schedule
Each prompt runs with live web search and without. The gap between the two separates what the model knows from what it just read, which is the difference between a brand problem and a content problem.
Full response retained
The answer behind every result is stored and readable. A mention rate you cannot open is a number you have to take on faith.
History per prompt
Mention rate, position and sentiment over time for each individual prompt, so a fall in the aggregate can be traced to the specific questions that moved.
Probe budget and prompt limits
50 monitored prompts on Measure through 400 on Prove, against monthly probe budgets of 3,000 to 24,000. The scheduler caps at the budget rather than billing you past it.
Bulk actions and priority
Prompts can be prioritised, paused and edited in bulk, because a prompt set is a living list rather than a one-time setup step.
Feeds Be Chosen
Brand mention rate across the prompt set scores rung 5,with the AI Overview gap count surfaced alongside it as evidence.
Why reproducibility is on the page
It is the metric that decides whether the rest of the numbers mean anything.
It marks which trends are real
Avoid presenting a convincing chart of nothing. A prompt with a low reproducibility score moves persuasively and means nothing, and publishing that score beside the trend is the difference between measurement and decoration.
It tests our own claim
Hold us to our own claim. We say persona-grounded prompts return more stable brand lists than invented ones, and reproducibility is the metric that proves or disproves it on your data, published either way.
It stops you optimising against noise
Spend the quarter on a question that will hold still. Chasing a mention-rate dip on an unstable prompt costs a sprint and produces nothing, because the dip was never there.
Against a prompt list you wrote yourself
The difference is where the prompts come from and whether their stability is known.
Prompts generated from ICP and real demand
TrustData
GA4
Other tools
Manual entryReproducibility score per prompt
TrustData
GA4
Other tools
Quality score across four axes
TrustData
GA4
Other tools
With-search and no-search on the same schedule
TrustData
GA4
Other tools
RarelyFull response retained and readable
TrustData
GA4
Other tools
Sometimes| Feature | TrustData | Google Analytics 4 | Other tools |
|---|---|---|---|
| Prompts generated from ICP and real demand | Manual entry | ||
| Reproducibility score per prompt | |||
| Quality score across four axes | |||
| With-search and no-search on the same schedule | Rarely | ||
| Full response retained and readable | Sometimes |
FAQ
Direct answers.
It is the practice of running a fixed set of questions against AI engines on a schedule and recording how your brand appears in the answers. TrustData runs each prompt in two modes, with live web search and without, across four question types, and keeps the full response behind every result. Prompt counts run from 50 on Measure to 400 on Prove, against monthly probe budgets of 3,000 to 24,000.
It is the median pairwise similarity between the brand lists an engine returned across a prompt's recent probes. If an engine returns roughly the same set of brands each time, the prompt is stable and a trend drawn from it means something. If the set changes every run, the prompt is unstable and its chart is noise that will still look like a trend. Publishing the score next to the trend is what stops a team spending a quarter chasing a movement that was never real.
From your own audience definition and from observed search demand, rather than from a brainstorm. You describe your ideal customer profile once, prompts are generated from those personas and from real query data, and they are suggested in bulk for you to edit, keep or reject. The reason is straightforward: a prompt list written from inside a company tends to ask the questions that company already wins, which produces a flattering baseline and no information.
Four axes on a 0 to 1 scale. Discriminance, weighted highest, is how far your mention rate diverges from your competitors' on that prompt: a question everyone wins tells you nothing. Stability is the variation in its daily mention rate. Actionability is whether the prompt is connected to a recommendation anyone actually followed. Recency is whether it has been probed in the last fortnight. Below three extracted probes the score is withheld rather than estimated.
Because they answer different questions. With search enabled, the engine retrieves current pages, so the result reflects your content and your crawlability. With search disabled, it answers from training data, so the result reflects what the model already believes about you. Strong with search and weak without means you are being retrieved but are not established in the model. The reverse means you are established but not being retrieved, which is usually a robots or extractability problem.
14-day free trial
Monitor the questions your buyers actually ask
14-day free trial. Prompt monitoring is part of AI Visibility, included in every plan.