Insight

AI Visibility Metrics: A Reproducible Sheet

Measure AI mentions, citations, referral sessions, and qualified actions with explicit denominators instead of a universal visibility score.

AI visibility metrics are dated observations that show whether an answer names a brand, cites a source, describes the brand accurately, sends a referral session, or precedes a qualified action. Measure each layer in its own field. Use explicit prompt-level denominators for mention and citation rates. Keep the prompt set, engine, date, and locale attached to every result.

My verdict: an inspectable measurement sheet is more useful than an invented universal AI visibility score. I would rather show the rows behind a rate than compress unlike signals into one number. That is an editorial position, not a claim that Jungle Roots tested every checker or index.

Measurement boundary: A visibility observation is evidence about a defined prompt set and run. It is not a prediction of revenue, rankings, conversions, or future citations.

Which AI visibility metrics belong in the report?

Use four layers: visibility observation, citation quality, owned-site analytics, and business outcome. They can sit in one report, but they should not become one blended score. Google, for example, defines Search Console clicks, impressions, and average position as Google Search performance metrics (Google Search Console Help). Google also says appearances in AI Overviews and AI Mode are included in overall Search Console traffic and reported within the Web search type (Google Search Central). Those fields do not document third-party assistant mentions.

The definitions below are Jungle Roots' measurement method unless a platform source is named.

Metric	Numerator or recorded value	Denominator or comparison boundary	What it establishes
Brand mention	Answers that name the tracked brand	Eligible answers in the fixed prompt set	Observed brand presence in that set
Citation rate	Answers that cite the tracked site or a defined tracked domain set	Eligible answers in the fixed prompt set	Observed citation coverage in that set
Citation quality	Cited URL, domain, relevance note, and support or contradiction flag	Review rule declared before assessment	Human review of the cited evidence, not a universal quality score
Answer accuracy	Accurate, inaccurate, mixed, or not assessable	Published review rule and facts available to the reviewer	A review flag for the captured answer
Referral session	Session attributed by owned-site analytics to a declared AI referrer rule	Analytics property, attribution rule, and reporting window	An owned-site visit under that rule
Qualified action	Action that meets a written qualification rule	Named action source and reporting window	A business event recorded under that rule

A brand mention can exist without a citation. A citation can exist without a referral session. A referral session can exist without a qualified action. The sheet preserves those gaps instead of filling them with a score. For the competitive ratio built from mentions, see AI Share of Voice.

What should the minimal weekly measurement sheet record?

Jungle Roots' minimal weekly sheet uses one row per eligible answer and keeps downstream evidence beside, not inside, that observation. “Weekly” is the cadence of this editorial protocol; it is not a platform recommendation or a claim that every team needs that frequency.

Prompt set and version: the frozen question text and the version that defines the run.
Engine, date, and locale: the answer surface, capture date, and language or market condition.
Brand mention: yes, no, or not assessable under the written rule.
Cited or not cited: a separate yes, no, or not assessable field.
Cited URL and domain: the visible source attached to the captured answer.
Answer accuracy flag: accurate, inaccurate, mixed, or not assessable, with a short review note.
Referral session: the owned-site analytics record when the declared referrer rule identifies one.
Qualified action: the downstream event only when it meets the written qualification rule.

The prompt set version prevents a silent wording change from entering the same trend. Engine, date, and locale preserve the observation boundary. The mention, citation, and accuracy fields describe the answer. Referral session and qualified action come from owned systems and remain downstream.

Operator note: Keep “not measured” distinct from “no.” A missing capture, blocked run, or unavailable analytics record is not a negative observation.

See the proof wall for the evidence-first reporting style behind this sheet.

How do you calculate mention rate and citation rate?

These formulas are Jungle Roots editorial definitions, not formulas published by an answer engine.

Mention rate = answers that name the tracked brand ÷ all eligible answers in the fixed prompt set × 100.

The numerator is the count of eligible captured answers that meet the declared brand-mention rule. The denominator is every eligible captured answer in the same fixed prompt set, engine scope, locale scope, and time window. If an answer could not be captured, apply the prewritten eligibility rule and disclose any exclusion instead of silently treating it as a non-mention.

Citation rate = answers that cite the tracked site or declared domain set ÷ all eligible answers in the fixed prompt set × 100.

The numerator is the count of eligible answers with a visible citation that meets the declared domain rule. The denominator is the same eligible answer set used for the stated citation rate. This is answer-level coverage. If the team wants citations per answer or unique cited domains, those are different metrics with different denominators and labels.

Do not merge rates produced from different prompt sets, engines, locales, or time windows. A changed denominator creates a different observation. Show the numerator and denominator beside every percentage so a reviewer can reconstruct it.

How should the four evidence layers stay separate?

The cleanest report moves from what the assistant displayed to what the business recorded. It does not claim that one layer caused the next. The following list is Jungle Roots' review order:

Visibility observation: inspect the exact prompt and captured answer; record the brand mention and accuracy flag.
Citation quality: inspect the displayed URL or domain and record whether it supports, contradicts, or does not clearly support the answer under the declared review rule.
Owned-site analytics: look for a referral session under the analytics property's written referrer and attribution rules.
Business outcome: look for a qualified action under the CRM or operating system's written acceptance rule.

This order creates a traceable chain without claiming a causal chain. An answer may cite a page without a visit, and a recorded visit does not by itself identify the prompt that preceded it. Keep any match between answer evidence and owned analytics labeled as observed, inferred, or unknown according to the evidence available.

Jungle Roots does not offer a universal AI visibility index.

What do platform controls and reports actually establish?

Platform documentation can define access controls and reporting scope; it cannot turn access into a citation guarantee.

OpenAI says OAI-SearchBot is used to surface websites in ChatGPT search features, while GPTBot concerns content that may be used to train generative AI foundation models (OpenAI). OpenAI also says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers except as navigational links, but its recommendation to allow the search bot is not a promise that a particular page will be cited (OpenAI).

Perplexity says PerplexityBot is designed to surface and link websites in its search results and is not used to crawl content for foundation models (Perplexity). It describes Perplexity-User separately as supporting user actions rather than web crawling (Perplexity). That separation supports a crawler-access check; it does not establish selection, citation rate, or traffic.

For Google AI features, a page must be indexed and eligible to appear in Google Search with a snippet to be eligible as a supporting link (Google Search Central). Google explicitly says meeting requirements and best practices does not guarantee crawling, indexing, or serving (Google Search Central). Bing's webmaster guidelines are likewise a source for Bing crawl and index guidance, not a universal answer-engine score (Bing Webmaster Guidelines).

Crawler boundary: Record access as an eligibility check. Never convert “allowed” into “included,” “cited,” or “converted” without the corresponding observation.

What is a good AI visibility score?

There is no universal good AI visibility score in this methodology. A useful benchmark is a declared baseline measured on the same fixed prompt set, engine scope, locale, eligibility rule, and time window. That comparison shows change within one defined observation boundary. It does not turn the result into a platform standard or a promise of future performance.

Before comparing one score with another, require the numerator, denominator, captured answers, and vendor formula. Also verify that both results use the same inputs and exclusions. If those records are missing or the boundaries differ, label the scores as non-comparable rather than treating the larger number as better.

This is Jungle Roots' editorial method, not a platform standard. It favors a reviewable baseline over a universal threshold because a reviewer can trace each rate back to the eligible answers that produced it. For the page-level implementation checklist, see how to optimize a website for ai search.

How do you audit an AI visibility score or checker?

Ask whether the output can be reconstructed from prompt-level records. A useful checker may automate collection, but its score remains the vendor's calculation unless the formula, inputs, exclusions, and denominator are disclosed. The audit question is not whether the number looks precise. It is whether a reviewer can identify what was observed.

To check AI visibility, run a fixed prompt set, capture answers and citations, apply declared mention and citation rules, then keep owned-site referrals and qualified actions separate.

Require the prompt set and version, engines, dates, locales, captured answers, mention rule, citation rule, eligible-answer rule, and any weighting method. Keep vendor-calculated values labeled as vendor metrics. Keep Jungle Roots formulas labeled as editorial definitions. Do not rename a brand-presence rate as a citation rate, or a citation count as referral traffic.

The same discipline applies to an AI visibility audit. The audit guide preserves query-level evidence, while AI visibility monitoring explains how to retain comparable captures. Neither method needs a universal index to produce an action.

What should the weekly review decide?

The review should decide which evidence gap deserves an owner. A non-mention can lead to a page or entity review. A weak citation can lead to a source check. An inaccurate answer can lead to a fact and positioning review. A referral session with no qualified action can lead to a landing-page or qualification review. These are editorial routing choices, not claims that the observed answer caused the gap or that a change will improve a future result.

Preserve the sheet before changing the prompt set. Start a new version when the questions, engines, locale scope, eligibility rule, or qualification rule changes. Compare only rows that share the stated boundary, and show unknowns rather than manufacturing zeros.

The final report should let a reader move from any rate back to its eligible answers, then forward to any separately recorded referral session or qualified action. That is enough to operate without pretending Jungle Roots has a universal AI visibility score.

See the proof wall

Written by Tileo, an operator who measures how AI assistants cite brands, on his own portfolio first.

What are the frequently asked questions?

What are AI visibility metrics?

AI visibility metrics are dated records of brand mentions, citations, answer accuracy, referral sessions, and qualified actions for a defined prompt set, engine, locale, and time window. Each layer needs its own field and evidence source.

How is AI mention rate calculated?

Jungle Roots defines mention rate as eligible answers that name the tracked brand divided by all eligible answers in the same fixed prompt set, engine scope, locale scope, and time window, multiplied by 100.

How is AI citation rate calculated?

Jungle Roots defines citation rate as eligible answers that visibly cite the tracked site or declared domain set divided by all eligible answers in the same fixed prompt set, multiplied by 100.

Does an AI visibility score predict revenue or citations?

No prediction is supported here. A score describes only its declared inputs and calculation. Revenue, rankings, conversions, future citations, and crawler inclusion require separate evidence.