How to Measure AI Visibility for a Web3 Brand
AI mention rate, share of voice, citation frequency, cited-domain distribution, sentiment, answer prominence, prompt coverage, competitor visibility, and content gaps. The nine metrics that define AI visibility, and why Google rankings cannot reveal any of them.
Joha J. Sulaymonov
Co-CEO, Lanesra AI
The short answer
Ask a Web3 marketing lead how their AI visibility is doing and you will usually get one of two answers. Either they do not know, or they cite a Google metric that has nothing to do with it. The second is worse, because it feels like an answer.
The confusion is understandable. For twenty years, discoverability had a shared measurement vocabulary: position, impressions, clicks, traffic. That vocabulary describes a system where a user sees a ranked list and chooses from it. Generated answers do not work that way, so the vocabulary does not transfer, and a new set of metrics is required.
Why Google Rankings Cannot Tell You Any of This
The core issue is mechanical. A generated answer produces no ranked list, no SERP position, no impression count, and frequently no click. There is nothing for a conventional analytics stack to capture, because the entire exchange happens between a user and a model in a session nobody else observes.
| Question you need answered | Google Data | AI Visibility Data |
|---|---|---|
| Do we rank for our target keywords? | Yes | Not relevant |
| How much organic traffic do we get? | Yes | Not relevant |
| Does ChatGPT name us in our category? | No | Yes |
| Does Gemini describe us accurately? | No | Yes |
| Which competitor is named instead of us? | No | Yes |
| Are we cited or only mentioned? | No | Yes |
| Which of our domains gets cited? | No | Yes |
| Is our framing positive or hedged? | No | Yes |
Google Search Console will show a healthy impression trend for a project that ChatGPT describes as something it stopped being two years ago. The two datasets are not in conflict, they are measuring different systems. Treating one as a proxy for the other is the single most common measurement error in this space right now.
The Nine Metrics
These are ordered from foundational to derived. Each one narrows the picture the previous one produced.
AI mention rate
The share of tracked prompts where your brand is named at all.
mentions / total prompts tracked
The foundational metric. Everything else is a refinement of it. A 12% mention rate across 200 category prompts means you are absent from 176 conversations your audience is having.
In practice: Track per engine, not pooled. A 40% rate in Perplexity and 4% in ChatGPT averages to something meaningless.
Share of voice
Your mentions as a proportion of all brand mentions in your category.
your mentions / all brand mentions
Mention rate tells you how often you appear. Share of voice tells you how much of the available attention you hold relative to everyone competing for it.
In practice: A rising mention rate with flat share of voice means the category is growing and you are keeping pace, not gaining.
Citation frequency
How often an engine links a retrievable URL from your domains rather than only naming you.
cited responses / total responses
Citations drive traffic and signal higher model confidence. A mention is recall. A citation is retrieval plus attribution.
In practice: The citation-to-mention ratio is the diagnostic. A low ratio suggests a content structure problem, not an awareness problem.
Cited-domain distribution
Which of your domains and subdomains actually get cited, and in what proportion.
citations grouped by source domain
Most Web3 projects assume their marketing site is the cited source. Frequently it is the documentation subdomain or the GitHub README, and the homepage barely appears.
In practice: Reveals where to invest content effort. Optimizing a page nothing cites is wasted work.
Sentiment
The qualitative framing of mentions: neutral, positive, hedged, or negative.
scored per response, then aggregated
Appearing in an answer with a caveat attached is not the same as appearing cleanly. Hedging language and references to past incidents materially reduce conversion.
In practice: Separates a tone problem from an absence problem. These need completely different fixes.
Answer prominence
Where in the response you appear: first named, mid-list, or a trailing afterthought.
average ordinal position
Position inside a generated answer matters the way search position does. Being named third in a list of five gets far less attention than being named first.
In practice: Tracked as an average position figure per prompt, which surfaces slow slippage that a binary mention count hides.
Prompt coverage
How many distinct prompts you track, and whether they reflect real query demand.
tracked prompts / relevant prompt universe
The most commonly neglected metric. A project tracking fifteen prompts it chose itself has no idea whether those are the prompts anyone asks.
In practice: Coverage gaps invalidate every other metric. Strong numbers on an unrepresentative prompt set are worse than no numbers at all.
Competitor visibility
The same metrics, computed for every named competitor in your set.
per-competitor mention rate and SOV
Absolute numbers are close to meaningless in isolation. A 15% mention rate is strong in a fragmented category and dire in a two-horse race.
In practice: Identifies which specific competitor holds each slot you want, which turns a vague gap into a target.
Content gaps
Prompts where competitors appear consistently and you never do.
competitor-present, you-absent prompts
This is the output metric. Every other number exists to produce this list, ordered by prompt demand.
In practice: The actionable deliverable. It converts measurement into a content and PR roadmap rather than a dashboard nobody opens.
What a Real Readout Looks Like
Numbers in isolation invite misreading. The value comes from reading them against each other, which is what a useful readout enables.
Illustrative monthly readout, hypothetical project
Read together: mention rate is up but share of voice is flat, so competitors grew at the same pace. Citation rate falling while mentions rise means new mentions are recall-based, not retrieval-based, which points at a content structure problem. Prominence slipping while coverage expanded suggests the newly tracked prompts are ones where the project ranks poorly. Priority: fix citation structure before chasing more mentions.
That reading is the actual work. A dashboard that reports six numbers without their interaction produces false confidence: mention rate up four points looks like a win in isolation and looks like treading water once share of voice sits next to it.
Engine Coverage and Why It Matters for Measurement
A visibility figure is only meaningful relative to which engines it covers. Measuring ChatGPT alone and calling the result "AI visibility" overstates coverage significantly, because engine behavior diverges substantially on the same prompt.
Core coverage:
Enterprise coverage:
The practical implication for measurement: a single blended visibility score across engines with very different behavior hides more than it reveals. Per-engine breakdown is not a nice-to-have, it is the only way the number means anything. A project can hold strong position in Perplexity, where grounded retrieval favors well-documented projects, and be nearly absent from ChatGPT, where training-weight recall favors projects with longer public histories.
Turning Answers Into Intelligence
The gap between having AI visibility data and having AI visibility intelligence is the analysis layer. Raw engine responses are text. Making them measurable requires parsing every response for brand presence, extracting ordinal position, resolving whether a URL was cited or a name merely recalled, classifying sentiment, attributing citations to specific source domains, and doing all of it repeatedly across engines so that variance is distinguishable from real change.
That last point is the one most often underestimated. Generated answers are non-deterministic. A single query returning no mention of your project proves nothing, and a single query returning a favorable mention proves equally little. Only repeated sampling over time separates signal from noise, which is why measuring AI visibility for Web3 is a monitoring problem rather than an audit problem.
This is the layer Lanesra provides: continuous prompt-level tracking across engines, with mention rate, share of voice, citation breakdown, cited-domain attribution, sentiment, prominence, competitor position, and the resulting gap list computed as structured metrics rather than raw text you have to interpret yourself.
Start measuring your AI visibility
Track mention rate, share of voice, citations, sentiment, and competitor position across every major AI engine. Sign up and see where your project actually stands.