What a prompt monitor actually tracks
Four things per prompt, per assistant, per run: whether your brand is mentioned, where it sits among named brands, how the answer frames you, and which sources the answer cites. Aggregated over a fixed prompt set and time, those become mention rate, share of voice, sentiment trend and a citation map, the metrics that make AI visibility manageable instead of mystical.
Why prompts, not keywords
People do not type keywords into assistants; they describe situations. A prompt monitor built on keyword lists measures a conversation nobody is having. The useful unit is the buying moment, expressed in several natural phrasings, my basement flooded, who do I call and is this covered, not water damage restoration. Variation across phrasings is itself signal: brands with real presence survive rephrasing; lucky mentions do not.
Designing a monitoring setup that tells the truth
Fix a prompt set of twenty to forty moments weighted by revenue. Run each in multiple phrasings across ChatGPT, Gemini, Claude and Perplexity, because they disagree constantly. Schedule monthly at minimum, judge trends rather than single runs, and store the raw answers, screenshots settle arguments that summaries start. Track competitors in the same runs, since every answer that omits you names who is beating you and why.
Prompt monitoring tools, and the monitoring trap
Free version: do it manually each month with a spreadsheet, honest but tedious past a dozen prompts. Paid prompt monitors automate the runs and the scoring; the trap is subscribing to a feed of your own invisibility with no path to change it. Aethon's prompt monitoring is moment-based across all four assistants and wired directly to the execution layer, each weak prompt traces to its causes and lands in a fix queue. The free audit is a one-time run of exactly this monitoring on your top questions, screenshots included.
