ai visibility metrics
Every number Babel42 reports, defined
What goes in, what gets counted, how you compare, and what to do next.
Appearance rate
Appearance rate is the share of journeys in which your brand is named anywhere in the assistant's answer, whether or not it goes on to recommend you.
what goes in
The inputs
A visibility number is only as honest as the question behind it. These are the three things that decide what gets asked, and of whom.
Keywords
Keywords are the real Google searches your brand already gets found for, imported with their monthly search volumes.
Babel42 looks up the queries your domain already ranks for in Google, keeps the ones inside the top 30 positions and attaches each one's monthly search volume. Nothing needs connecting: it is a lookup against your domain, not an integration with your analytics. You tick which queries your AI Buyers open with, up to 50 are stored per brand, and you can add your own by hand.
This is the difference between measuring the questions your buyers actually ask and measuring the questions a tool guessed they might.
AI Buyer
An AI Buyer is a synthetic buyer persona that shops your category on a named AI assistant, the way a customer would.
It is a persona Babel42 runs, not a person we observed. What is real is the question it asks, drawn from your search demand, and the answer the assistant gives back live. A buyer can be pinned to a city or town, which is passed as the buyer's location to the assistants that accept one and stated in the question itself for the rest.
Buyer journey
A buyer journey is one complete multi-turn shop: an opening question, follow-ups, and a final decision naming a winner.
Babel42 pushes conversational assistants through the whole conversation rather than asking once, which is what makes a win rate possible at all. Every journey is replayable as a workflow diagram showing each prompt, each answer, the brands named and the sources cited. Each one runs twice, so a result can be checked for stability rather than reported from a single pass.
what gets counted
The six headline numbers
The scoreboard. Read them together: any one of them on its own can tell you a flattering story.
Appearance rate
Appearance rate is the share of journeys in which your brand is named anywhere in the assistant's answer, whether or not it goes on to recommend you.
Babel42 records it for every question an AI Buyer asks and reports it per assistant and per query, rather than as one blended figure. A brand can appear constantly on one assistant and be invisible on another, and an average hides exactly that.
A high appearance rate beside a low win rate is the most common pattern we see: AI knows you exist, and is still sending the buyer somewhere else.
AI win rate
AI win rate is the share of journeys where the buyer's final pick was your brand.
Babel42 counts decisions here, not mentions. This is the metric most AI visibility tools do not produce, because firing a static prompt at a model and tallying names never reaches a decision there is anything to count.
This is the number that moves with revenue, because a buyer acts on the recommendation rather than on the longlist.
Weighted win rate
Weighted win rate is your win rate after discounting the wins the buyer had to be coaxed into.
Babel42 grades every win before weighting it. A win counts in full when you are named unprompted in the assistant's first answer, counts less when you appear during normal follow-ups, and counts least when you only turn up after the buyer asks for more options. Three wins are not equal, and a raw win rate pretends they are.
It separates a brand the assistant reaches for from one it settles on once pushed. A wide gap between win rate and weighted win rate is a fragile position.
Shortlist rate
Shortlist rate is the share of journeys where your brand reached the buyer's final two or three candidates.
In a Babel42 run it sits between appearance and the win: the buyer took you seriously enough to carry you to the end, then chose someone else. Those journeys are the most recoverable ones you have.
how you compare
The competitive picture
Appearance, wins and share of AI voice are computed for the brands appearing beside you as well as for yours, because a visibility score with no rival in it is not a market position.
Rival leaderboard
The rival leaderboard ranks the competing brands by how often the assistant picked them, with the stated reason attached to each win.
Babel42 gives you the rival taking the most decisions off you, not just a list of names that came up. Journeys can be filtered to a single competitor, so you can read every conversation where a particular rival beat you.
Most brands can name their biggest competitor. Far fewer can name the one the assistants keep choosing instead of them, and those are often not the same company.
Win quality
Win quality grades each win as strong, standard or weak, according to how hard the buyer had to work to reach you.
A strong win means you were named in the assistant's first answer, unprompted. A standard win means you were named during normal follow-up questions. A weak win means you only appeared after the buyer asked for more options. These are the same three grades Babel42 uses to produce the weighted win rate.
Performance by AI model
Performance by AI model breaks appearance rate and win rate out separately for each AI assistant.
Babel42 reports the blended figure and the per-assistant split side by side, because the assistants disagree with each other constantly: they read different sources and search the live web differently. A brand that wins on one and loses on another has a source problem, not a product problem, and the blended number on its own hides that.
Stability
Stability records whether a buyer's result held across repeat runs: always appears, volatile, or never appears.
AI answers vary between runs even for an identical question. Because every journey runs twice, Babel42 can tell you whether a result is a reliable reading or a coin toss, which is the difference between a trend and noise.
Visibility trend
Visibility trend plots your appearance and win rate over time against the same set of questions.
A single scan is a snapshot. The anchor questions carry between runs unless you change them, so movement in the trend is movement in your position and not a change in what was asked. It is available from Starter upwards.
why it went that way
The reasons behind the score
Numbers tell you where you stand. These tell you why, and they are where the work to change it actually starts.
Citations
Citations are the specific URLs an assistant quoted or linked when it composed its answer.
Babel42 lists every source with its URL and marks which cite you and which cite a rival. The rival-citing pages are the practical target list: they are the places already shaping the answer, so they are where coverage and backlinks change it.
Perception
Perception is the position your brand is being given in the answers: market leader, value pick or niche specialist.
Babel42 derives one label per run, from how often you appear, how often you win, and whether price is what rivals are compared with you on. It is a read of your position across the answers rather than a quote of any single assistant's wording. Two brands can share an appearance rate and be positioned in ways that send completely different buyers their way.
Sentiment in AI answers
Sentiment in AI answers is whether an assistant's description of your brand is positive, neutral or negative.
This is not social sentiment, which measures what people post about you. Babel42 measures both, in its two products, and keeps them apart for exactly this reason: what an assistant hands to a buyer who asked for advice is a far shorter path to a purchase than what the internet says in general.
Objections
An objection is the specific stated reason an assistant gave for not recommending your brand.
Babel42 keeps the conversation for as long as your plan's retention allows, so you can read the sentence that turned a shortlisting into a loss: the price, the missing integration, the thin documentation, or a rival's stronger proof.
Of everything on this page, this is the part you can act on this week.
what to do about it
The output
A measurement you cannot act on is trivia. Every run ends in a ranked list of specific things to change.
Recommendations
Recommendations are the ranked, specific actions drawn from your own results: the page to publish, the source to earn a mention on, the objection to answer.
Babel42 sorts them into three groups. Beat the rivals targets the competitor taking the most decisions off you. Fix the model gaps targets the assistants where you underperform your average. Win the decision targets weak wins, lost finals and the objections behind them. Where a journey produced a quote worth reading, the recommendation carries it, so you can check the reasoning instead of taking it on trust.
A visibility report that ends in a number leaves you roughly where you started. This one ends in a list.
how it is measured
Where the numbers come from
A metric is only worth as much as the method behind it, so here is the method in full, including the part that is simulated.
The questions are real
Prompts are anchored to the queries your domain already ranks for in Google, with monthly search volumes attached, and you pick which ones matter. Where you pick none, the buyer opens on the highest-volume intent found rather than on a prompt written for you.
The buyers are synthetic
An AI Buyer is a persona Babel42 runs, not a real person we observed. What is real is the question it asks and the answer the assistant gives back, live.
The journeys reach a decision
Conversational assistants are shopped multi-turn until there is a recommendation to record, which is what makes a win rate possible at all.
Every journey runs twice
AI answers vary between runs. Each journey is repeated so a result is checked for stability rather than reported from a single pass.
Babel42 tracks seven AI assistants: ChatGPT, Claude, Perplexity, Gemini, Grok, Google AI Overviews and Google AI Mode. How many a single AI Buyer uses at once depends on your plan, and the pricing page sets out which.
Metrics, answered
What is the difference between appearance rate and win rate?
Appearance rate counts whether your brand was named in an AI answer. Win rate counts whether the assistant actually recommended you at the end of a buyer journey. A brand can be named in most answers and win almost none of them, which is why appearance rate on its own is a misleading measure of AI search visibility.
What counts as a good result?
Babel42 does not roll your results into a single score out of a hundred, because categories differ enormously in how many credible brands an assistant has to choose from and a composite would hide that. The readings that matter are the gap between your appearance rate and your win rate, the gap between your win rate and your weighted win rate, and the trend in all three against the same set of questions.
How is share of AI voice different from share of voice?
Traditional share of voice measures your slice of the human conversation, in press coverage or social posts. Share of AI voice measures your slice of the brand mentions inside AI assistants' answers. They can diverge sharply: a brand can dominate social chatter and barely register in the answers an assistant gives a buyer who asks for a recommendation.
Do you measure competitors as well as my brand?
Yes. Every journey records each brand the assistant named and which one it finally picked, so rivals get the same appearance, win and citation figures you do. You can filter the journeys to a single competitor and read every conversation in which it beat you, with the reason the assistant gave.
Where do the questions come from?
From your own demand. Babel42 looks up the queries your domain already ranks for in Google, filtered to the top 30 positions, with monthly search volumes attached, and you choose which ones your AI Buyers open with. If you pick none, the buyer opens on the highest-volume intent Babel42 found rather than on a prompt written for you.
Are these metrics measured the same way on Google AI Overviews?
Not quite. Five of the assistants Babel42 tracks are conversational, so they are shopped multi-turn to a decision and produce a win rate. Google AI Overviews and AI Mode are not conversational in the same way, so they are probed once per question for appearance and citation, with an honest “no AI Overview was shown” recorded when Google returns none.
See your own numbers
Run a set of AI Buyers against your category and get every metric on this page for your brand and the rivals beside you. Free to start, no card required.