1
Add your prompts
A Prompt is one question phrased the way a buyer phrases it, not a keyword.
best invoicing software for freelancers is a prompt; invoicing is not.Start with ten to twenty prompts that map to how people actually shop for what you sell: category questions, comparison questions, and problem questions. Group related ones into a Topic so you can read a whole buying question as a single line later.Your plan sets how many Prompt slots you have: 50 on Starter, 150 on Growth, 350 on Business. A slot is occupied for as long as the prompt is tracked, and freeing one is as simple as removing the prompt.2
Confirm your competitors
Attensira reads every answer for your brand and for each competitor you track. Add the names buyers would put in the same sentence as yours, and check the spellings — a competitor named differently from how models write it will look absent when it is not.Competitor rates share the same denominator as yours, which means they will not add up to 100%. Two brands can both appear in the same answer, and often do.
3
Wait for the first readings
An hourly in-process ticker enqueues work, but each prompt × model × country combination runs at most once per workspace-local calendar day. So the first answers arrive within a day of setup, not within minutes, and adding a prompt after today’s run has already fired means its first reading lands tomorrow.A new workspace tracks two model surfaces: ChatGPT and Perplexity. Add more from settings if you want them.Sampling depth is the number of times each prompt × model is asked per day:
n = 1 on Starter, n = 3 on Growth and Business, issued concurrently. Failed runs are stored for inspection but excluded from n, so a bad day lowers your sample size rather than faking a zero.4
Read the board
Two numbers matter first. Mention rate is the share of successful runs that named your brand. Citation rate is the share of successful runs that attached a link to your own domain. They answer different questions: whether the model knows you, and whether it sends readers to you.Every rate arrives as a value plus
n, the number of successful runs behind it. A value of 0 with n > 0 is a real measured zero. A null value means not measured — no successful run exists for that cell, and it is not a zero.Per-model numbers also carry tracked and readable. tracked: false means that model is not configured for this workspace and was never queried, so reading it as 0% invents a failure that did not happen. readable: false means no reader exists for that surface.Changes over time pass a two-proportion z-test at 95%. Below that noise floor you get {real: false, value: null}, which means “no change we can prove” — never zero.