The AI visibility dashboard you put in front of a board.
Every week we run your buyer prompts through the assistants and record what came back: whether the answer named you, which sources built it, whether it recommended you, and how much of the category conversation your competitors took. Each number points at a different thing to fix.
What arrives each month.
Named prompt set
The questions your buyers ask while they are choosing, agreed with you and then frozen. A set that keeps changing has no trend line in it.
Continuous tracking
Every prompt runs across the assistants on a fixed weekly cadence. Monthly misses real movement and daily buries it in model noise.
Live dashboard
Month-over-month movement with your named competitors sitting beside you, updated as the runs land.
Monthly readout
Thirty minutes that ends in decisions: what moved, what caused it, and what changes next month.
Six AI visibility metrics and the work each one triggers.
Mention rate
Prompts where the answer names you, divided by prompts tracked. A low number means the model does not know you exist, which is an entity problem before it is a content problem. It sends us back to schema, Wikidata, and the profiles models read to decide who you are.
Citation share
Answers that cite your domain, divided by answers carrying any citation, scored against named competitors. When the cited sources belong to competitors, the answer about your category is getting built from work you do not own. That routes to topic ownership.
Recommendation rate
Mentions where the model actively puts you forward, divided by mentions. A brand that gets named and never recommended has a framing problem. We go after reviews, comparison pages, and the third-party proof a model weighs before committing to a name.
Share of voice
Your mentions divided by every brand mention in the category, yours and your competitors together. Share climbing means the work is landing. When it falls while your mention rate holds, a competitor is moving faster and we name whose pages are taking the citations.
Visibility score
One number for a board, weighted: mention rate 35 percent, citation share 30, recommendation rate 20, share of voice 15. We publish the weights because a composite you cannot audit is a number you should not trust.
AI-attributed traffic
Clicks and leads arriving from AI surfaces, plus modeled branded-search lift after a visibility win. This one is correlational. We write the assumptions down every month and never hand it to you as causal proof.
Choosing the prompts worth tracking.
A prompt earns its place when a real buyer would type it while deciding, and when the answer could plausibly change. Best CRM for a 40-person sales team qualifies. What is a CRM does not, because every assistant answers it the same way for everyone and nothing you do will move it.
We build the set with you across branded, category, comparison, problem-solving, and recommendation questions, weighted to how your category gets bought. Branded questions tend to land around a third of the set. Early-funnel categories carry more problem-solving. Each quarter we revisit the weighting and leave the prompts themselves alone.
How we measure brand visibility in ChatGPT, Gemini and Perplexity.
ChatGPT, Gemini, Perplexity, and Google AI Overviews are the default set. Those four cover most of how buyers reach an AI answer in 2026, and each one retrieves differently enough to be worth its own column. Claude, Copilot, and Google AI Mode go in on request. A given engagement runs three to five of them.
Every prompt runs in a clean session with no history, no personalization, and no system instructions, which is how an anonymous buyer meets the model. For each prompt and assistant we keep five things: the answer text, the cited sources in the order they appeared, whether your brand is named, how the mention frames you, and where you land in any list or comparison. Every number above is built from that record.
What this reporting cannot tell you.
Voice is not covered
Everything here runs on text prompts. Voice assistants retrieve differently and need their own framework, which we have not built.
Multimodal is partial
We capture answer text and note when images appear. Visual brand visibility goes unscored, which matters most in retail and creative categories.
Multilingual sentiment is harder
The three-tier rubric holds up in English and Bahasa Indonesia. Other languages produce more disagreement between scorers, especially where honorific or indirect speech is the norm.
Attribution stays correlational
When branded search rises after a visibility win, we model the link and show our assumptions. Causation is beyond what this method can prove, and beyond every public method we have read.
Results from AI visibility clients.
Common questions
What do buyers complain about most with AI search visibility services?
Prompt sets that quietly change between reports, which turns any trend line into noise. Composite scores whose weighting the vendor will not show you. A number like that can rise while everything underneath it falls. And correlational attribution presented as causal proof. We freeze the prompt set with you, publish the weighting behind the visibility score on this page, and label the modeled traffic number as correlational every month.
How is this different from an AI visibility tool we could buy ourselves?
Getting a visibility number is the cheap part of this work. What a dashboard will not do is agree the prompt set with your sales team, trace a falling citation share to the competitor page that took it, rewrite the schema, and get the fix through your release process. Several clients run a tool alongside us and we have no objection to it.
How do you measure traffic attribution from AI answer engines?
Direct referrals from AI surfaces are measured from your analytics, and those are real clicks. Branded-search lift following a visibility win is modeled, because an assistant that recommends you often produces a search rather than a click. We write the modeling assumptions into every readout. Anyone presenting the second number as measured fact is overselling it.
How do you tell a real change from model noise?
A frozen prompt set, a weekly cadence, and clean sessions with no history or personalization. Assistants vary their wording run to run. A single check proves nothing. Weekly sampling across a fixed set means a movement has to persist across several runs before it shows in the monthly number. Single-digit swings inside one month get flagged as noise rather than reported as progress.
How do you report AI search performance to a board?
One headline number with its weighting published, the four component metrics underneath it, and named competitors on the same chart. Boards discount a metric they cannot check, which is why the visibility score formula sits on this page rather than inside a black box. The readout is thirty minutes and ends in a decision about next month.
Find out where your brand stands today.
Request an AI visibility audit and we will benchmark your brand and three competitors across the major AI assistants. You get a 15-page report and a 30-minute walkthrough.