Prompts
The four prompt groups, how to choose a mix, and how to read per-prompt performance.
Prompts are the measurement instrument. Everything the platform reports is an aggregate over the answers your prompt set produced, which makes this the highest-leverage screen in the product.
The four groups
Every prompt sits in exactly one group, formed by crossing two axes: subject (your brand, or one of your products) and intent (visibility, or perception).
| Group | Asks | Example |
|---|---|---|
| Brand Visibility | Category questions where you want to be named | best AI visibility platform for SaaS brands |
| Brand Perception | What engines say about you by name | is Geonimo any good for enterprise teams |
| Product Visibility | Category questions for one product | best AI crawlability checker |
| Product Perception | What engines say about that product | how accurate is Geonimo's AI tracker |
Four sections, always, whatever you sell. The product is a property of the row rather than its own heading — otherwise ten products would mean twenty-two sections, and every customer's table would be shaped differently.
The brand-name rule
Visibility prompts must not contain your brand name. The rule is a property of intent, and it is enforced when prompts are created.
The reason is that a prompt naming you guarantees an answer naming you. Measuring visibility on a branded question measures your own name back at yourself, and it inflates every downstream number. Perception prompts are the opposite: naming you is the point, because you are asking what the engines say about you specifically.
Keeping the two apart is what makes the Overview's visibility figure mean something.
Choosing your mix
Rather than a fixed plan-imposed list, you choose how the budget splits:
- Brand Visibility and Brand Perception counts, set directly.
- Products — chosen by count, not by prompt count. Each tracked product gets a fixed allocation: 4 visibility prompts and 2 perception prompts.
Products are counted rather than tuned because the arithmetic does not leave a choice. Collection produces roughly 2 answers per prompt per day; over 30 days one prompt yields around 60 answers and a confidence band about 22 points wide — not a number anyone can act on. Four prompts yield around 240 answers and a band near 11 points. So a product gets 4 visibility prompts or it is not worth tracking. The free variable is how many products you cover, never the depth per product.
The dialog quotes the totals — including scrapes per day, which is the number that recurs on the bill — before you spend anything. Bounds: 150 prompts total, 10 products, 100 prompts per brand group.
Generating a prompt set
The generator audits your website, identifies your category and the products you sell, and proposes prompts to your chosen mix. Each proposal carries its intent and, for product prompts, the product it belongs to — so accepting a set wires the Products tab up at the same time.
Review before accepting. The generator is good at coverage and cannot know which questions your buyers actually ask; a set that mixes generated breadth with ten questions from your sales calls beats either alone.
You can also add prompts by hand at any time, and archive ones that stop earning their place.
Reading performance
Each prompt row shows how it performs over the selected window, sorted by visibility within its group, with its category tag and the country it is tracked in.
- Consistent wins — your defensible ground. Understand why, then reproduce the pattern.
- Consistent losses — the work queue. Open the answers and see who wins and what is cited.
- Volatile prompts — flipping between mentioned and absent means you are borderline. The cheapest to convert.
- Wide intervals — not a result yet. More evidence needed before you act.
Open a prompt for its detail panel: trend, per-engine breakdown, and the captured answers themselves. Read the answers. The number tells you that you lost; the text tells you why.
While a prompt is analyzing
A newly added prompt is dispatched on the next hourly pass rather than waiting for the nightly run, and the tab polls while rows are analyzing — every 8 seconds, for up to 20 minutes — so a row resolves while you are still looking at it.
Maintenance
Review monthly, prune quarterly. Drop prompts that never move and never matter, promote the questions that keep appearing in your fan-out data, and add what buyers started asking this quarter.
Note that changing the prompt set changes your denominators. Adding ten weak prompts lowers blended visibility without anything having got worse — one more reason to read groups separately rather than one headline number.