The Two Scales Behind an AI Visibility Score: Why a 55 Composite and a 0.29% Citation Rate Describe the Same Brand

A brand visibility score is only readable once you know which of two scales it is on, and the two are orders of magnitude apart. On a 0-to-100 composite, the largest published benchmark of 200 brands across five assistants averages 55.8, so the mid-50s is ordinary and above 70 is strong. On a citation-rate scale, the share of monitored answer runs in which an engine cited the domain, the graded median across the Machine Relations Index is 0.29% and the highest rate any domain reaches is 12.67%. A composite reading 55 routinely sits on top of a citation rate below 1%, because presence and citation are different layers and no normalization reconciles them.

Two further properties decide whether a number means anything: the engine set it was computed over, and the minimum evidence required before the score is shown at all. The Index publishes no rate below 10 observations across 7 distinct run dates; a platform with no floor prints single observations as movement. Those three properties, scale, engine set and evidence floor, are the measurement contract behind a score. Whether the vendors state their own contracts consistently is a separate audit, run across nine platforms by AuthorityTech, a related party to this site; what follows here is what the platforms query and expose, read against the three properties.

Most AI visibility tools measure the wrong thing. They count brand mentions in AI responses while ignoring whether answer engines actually cite — link to — your content as a source. In Q2 2026, cross-engine citation overlap sits at just 18% (Foglift Research), meaning a brand visible in ChatGPT may be invisible in Gemini, Perplexity, or Google AI Mode. Measuring one engine tells you almost nothing about the others.

This analysis compares 20 AI visibility measurement tools across three dimensions that determine whether a platform delivers actionable intelligence or dashboard theater: engine coverage depth, API capability, and attribution methodology.

Why Citation Tracking Requires Multi-Engine Measurement #

The fundamental challenge of AI visibility measurement is engine disagreement. Foglift Research's Q2 2026 benchmark tested 75 brand-neutral buyer-intent prompts across 25 verticals and found:

• 1,119 distinct domains cited across 375 total responses from five engines (ChatGPT, Claude, Gemini, Google AI Overview, Perplexity)
• Cross-engine Jaccard similarity of 0.18 — engines agree on sources less than one-fifth of the time
• 61.7% of top-25 cited domains are engine-exclusive — appearing in only one engine's top list
• Only one domain (healthline.com) appeared in all five engines' top-25 lists

Source: AI Search Citation Benchmark: Q2 2026

Note, September 21, 2026. Most published engine-overlap shares are computed over every cited domain, counted once each, which makes them largely a measure of sampling depth: on the Machine Relations Index release of September 21, 46.2% of 22,377 cited domains had been cited exactly once in 128 days and are single-engine by arithmetic. The benchmark above is not that measurement — it is computed over each engine's top-25 cited domains, so its exclusivity figure is about frequently cited sources and stands. That distinction is the one to apply when reading any tool's overlap number: ask whether the share is conditioned on how often a source was cited. See Cross-Engine Citation Agreement.

This means a tool tracking only ChatGPT and Perplexity misses citation patterns in engines that may drive the majority of a brand's AI-sourced traffic. The Machine Relations Index addresses this by scoring source authority across all six major answer engines — but most commercial tools cover three to four.

The Citation vs. Mention Distinction #

A brand mention means an AI response names your company. A citation means the response links to your content as a source. The difference is structural: mentions indicate awareness, citations indicate authority. Tools that conflate the two produce misleading visibility scores.

As SeekLab's analysis documents, effective measurement requires storing prompts, full responses, source URLs, timestamps, and competitor presence for each tracked query. Without this evidence chain, "the report becomes a screenshot exercise."

Engine Coverage Across 20 Measurement Platforms #

Engine count is the first measurement property to establish, because it bounds what any score on the platform can describe. The inventory below records how many answer engines each platform queries. It is a coverage census, not a recommendation: this publication does not rank vendors or compare prices.

The market segments into three tiers by primary capability.

Tier 1: AI-Visibility-Native Platforms #

These tools were built specifically to track brand presence in AI-generated responses.

Tier 2: SEO Platforms With AI Visibility Layers #

Traditional SEO suites that added AI tracking features on top of existing infrastructure.

Tier 3: Adjacent Tools With Partial AI Tracking #

Platforms not built for AI visibility but offering limited overlap.

Source: Best AI Search Visibility Tools 2026 — 20-Tool Comparison

API Capability Matrix: What Platforms Actually Expose #

Feature lists obscure a critical gap: what data each platform makes accessible through APIs for integration into existing workflows. Surferstack's capability matrix classifies platforms into four API depth tiers:

The gap between Tier 1 and Tier 4 is not incremental — it is structural. A Tier 1 platform tells you that ChatGPT mentioned your brand. A Tier 4 platform tells you which of your pages AI crawlers visited, which queries triggered citations, and whether those citations drove attributable traffic.

The Attribution Problem: GA4 Captures 9% of AI Traffic #

The most significant measurement gap is not in the tools themselves but in downstream attribution. GA4 captures approximately 9% of actual AI-driven visits, with the remainder appearing as direct traffic (Surferstack). This means even brands running sophisticated AI visibility tools cannot reliably connect citation presence to website traffic using standard analytics.

Only two platforms — Promptwatch and Profound (limited) — attempt to solve this through proprietary traffic attribution layers. For most organizations, the practical workaround remains correlating AI citation gains with direct-traffic increases over time windows, not attributing individual visits.

How Brand Visibility Scores Differ Across AI Platforms, and What a Good Score Is #

Two questions follow every tool comparison, and the tables above answer neither: does a visibility score mean the same thing on each platform, and what number is good. Both have measured answers. Neither is the answer a dashboard implies.

Scores differ across platforms because the engines differ — but that difference has to be read against evidence depth. Cross-Engine Citation Agreement measures it on the September 21, 2026 Index release: counting every cited domain equally, 67.4% of 22,377 domains appear in only one of the six engines and 1.6% appear in all six. That headline figure is largely sampling depth rather than engine disagreement — 46.2% of the index was cited exactly once across 128 days, and a domain cited once is single-engine by arithmetic. Require ten citations and the single-engine share falls to 1.5%; among the 508 domains the Index grades on evidence, it is zero, and the closest engine pair shares 90.6% of its source set.

The consequence for tool selection is a sampling one, not an ideological one. At the citation volume a typical brand has, one engine is a poor estimator of six, so a score built on two or three engines describes two or three surfaces whatever the dashboard calls it. At the volume a major publisher has, the engines largely agree and engine count stops discriminating. Establish which population your brand is in before weighting any cross-engine number.

Window agreement is also not answer agreement. Measured answer by answer rather than aggregated over months, only 13.5% of the domains cited for a given question appear on more than one engine. A blended score cannot see that, because blending dissolves the unit where it happens.

What a good score is depends on which scale the tool is using, and the two common scales are orders of magnitude apart. On a 0-to-100 composite, the largest published benchmark of 200 brands across five assistants has a mean of 55.8, so the mid-50s is average and above 70 is strong. On a citation-rate scale — the share of monitored answer runs in which an engine cited the domain — the graded median across the Index is 0.29% and the highest rate any domain reaches is reddit.com at 12.67%. A composite reading 55 routinely sits on top of a citation rate below 1%, because presence and citation are different layers. Establish which of the two scales your number is on before comparing it to anyone's.

Three checks when comparing scores across the tools above. In this order:

1. Which engines, and at what citation volume for your brand? Weight the score accordingly. Two engines is two surfaces, and at low counts it is a bad estimate of six.
2. Is the number a citation rate, a mention share of the tool's own prompt set, or a normalized 0-to-100 index? Only the first is comparable to anything outside that tool.
3. What minimum evidence is required before a score is shown? The Machine Relations Index publishes no rate below 10 observations across 7 distinct run dates. A platform with no floor prints single observations as movement.

Which of these platforms document their method well enough to answer those three is measured separately in the AI visibility methodology transparency benchmark.

Machine Relations Framework: How This Connects #

The Machine Relations approach to AI visibility measurement differs from commercial tooling in three ways:

1. Engine breadth: The MRI scoring methodology evaluates source authority across six engines (ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, Perplexity), while most commercial tools cover three to four.

2. Source authority, not just brand presence: MRI measures citation concentration, source preferences by engine, and temporal consistency — the structural factors that determine whether a source maintains citation authority over time.

3. Cross-engine divergence as a feature: Where commercial tools aggregate across engines into a single score, the MRI framework treats engine-specific citation behavior as the primary signal. A domain cited by all six engines has fundamentally different authority than one cited by two.

What Makes a Citation Measurement Defensible #

The properties below determine whether a number produced by any of these platforms can be reproduced and defended. They are tests to run against a measurement, not purchase criteria, and they apply equally to a vendor dashboard and to an in-house measurement:

1. Real API calls vs. proxy models. Some tools simulate AI engine responses rather than querying live systems. A measurement is only of live engine behavior if each engine was called directly rather than modeled from historical data.

2. Citation source extraction, not just mention counting. A measurement of citation requires the actual URLs cited in AI responses to be captured and stored, not merely whether the brand name appeared.

3. Prompt coverage model. Fixed prompt sets create blind spots for unmeasured query categories. Platforms allowing custom prompt sets or continuous monitoring cover more of the query space relevant to your brand.

4. Multi-engine coverage depth. Three engines measured deeply (with full response capture, source extraction, and historical trending) is more valuable than nine engines measured shallowly through basic mention counting.

5. Evidence chain storage. Every tracked query should produce a stored record containing the prompt, full response, source URLs, timestamp, engine version, and competitor presence. This is the audit trail that makes measurement defensible.

2026 Category Shifts #

Three developments reshape the AI visibility measurement landscape:

• Reddit as dominant citation source. Approximately one in five citations across ChatGPT, Perplexity, and Google AI Mode now reference Reddit content — a signal that earned media placements in community platforms carry citation weight.
• Google AI Mode replacing AI Overviews in many search surfaces, changing what "AI visibility in Google" means for measurement tools tracking the older format.
• MCP server adoption. Four platforms (Profound, Peec AI, GEO Tracker AI, Brand24) now ship MCP servers, enabling direct integration with AI development workflows rather than dashboard-only access.

FAQ #

How many engines does a measurement need to cover before its score generalizes? That depends on the citation volume of the domain being measured, not on a fixed engine count. Among domains cited ten or more times on the September 21, 2026 Index release, the single-engine share is 1.5%, and zero among the 508 domains the Index grades — so at that volume the engines largely agree and additional coverage adds little. At the volume a typical brand has, where 46.2% of indexed domains were cited exactly once across 128 days, a two- or three-engine score describes two or three surfaces and does not estimate the rest. Establish which population the domain is in before weighting any cross-engine number.

How accurate are AI visibility scores? AI outputs vary by prompt wording, location, account context, model version, and retrieval behavior. Visibility scores are directional indicators — useful for trend tracking and competitive benchmarking — not absolute measurements. Consistent tracking over time produces more reliable signals than single-point snapshots.

Why does cross-engine citation overlap matter for tool selection? With Jaccard similarity at 0.18, the sources that ChatGPT cites are largely different from those that Perplexity, Gemini, or Claude cite. A tool tracking only one or two engines gives an incomplete and potentially misleading picture of a brand's AI search presence.

How does a brand visibility score differ across AI platforms? Structurally, because the engines cite different sources — but read the gap against evidence depth, not headline counts. On the September 21, 2026 Machine Relations Index release, 67.4% of the 22,377 cited domains appear in only one of six engines; require ten citations and that falls to 1.5%, and among the 508 domains the Index grades it is zero. So at the citation volume a typical brand has, a score covering two engines is a poor estimate of six; at publisher volume the engines largely agree. Compare the engine lists before comparing the numbers, and see cross-engine citation agreement for the conditioned measurement.

What is a good score for AI brand visibility? It depends which scale the tool reports on, and the two common scales are orders of magnitude apart. On a 0-to-100 composite, the largest published 200-brand benchmark means 55.8, so mid-50s is average. On a citation-rate scale the graded median across the Index is 0.29% and the highest rate any domain reaches is reddit.com at 12.67%, so a composite of 55 often sits on a citation rate under 1%. Establish which scale your number is on first.

Which AI visibility platforms explain how their visibility score is calculated? Disclosure varies widely, and three questions separate the platforms that document it from the ones that do not: what the denominator is (a share of what — runs, prompts, responses?), which engines are queried, and what minimum evidence is required before a score publishes. Platform-by-platform documentation quality is measured in the AI visibility methodology transparency benchmark. As a published baseline, the Machine Relations Index states both its definition and its evidence floor: at least 10 observations across at least 7 distinct run dates before a segment publishes a rate.

What is the difference between a brand mention and a citation in AI search? A mention means the AI response names your brand. A citation means the response links to your content as a source. Citations indicate source authority; mentions indicate awareness. The distinction determines whether your content is being used as a reference by AI engines or merely referenced in passing.

Last updated: October 4, 2026. Vendor pricing and positioning were removed from this analysis on October 4, 2026: provider choice, vendor comparison and pricing are outside this publication's scope, and this publication measures how engines select sources rather than recommending who to buy from.