# How to Measure AI Visibility With a Source-Layer Baseline

Machine Relations Research defines a source-layer baseline for measuring AI visibility without collapsing mentions, citations, retrieval state, and source quality into one opaque score.

Canonical URL: https://machinerelations.ai/research/measure-ai-visibility-source-layer-baseline-2026
Published: 2026-09-12
Research type: Practitioner Analysis
Tags: AI visibility, measurement, share of citation, source layer, Machine Relations

## Source Body

AI visibility should be measured as a source-layer system, not a single score. The minimum baseline is answer presence, brand mention rate, cited-source share, retrieval state, and source quality across a fixed query set. Without those separate denominators, teams cannot tell whether visibility changed because the brand became more credible or because the measurement collapsed different answer behaviors into one number.

Machine Relations Research uses this baseline because current AI search products expose source behavior in different ways. ChatGPT search can answer with links to sources, Google AI Mode organizes answers with links to explore on the web, Claude web search says every response includes citations, Perplexity positions its API around real-time web research, Gemini grounding connects model answers to Google Search sources, and Microsoft describes Copilot Chat as grounded in public-web data from Bing Search ([OpenAI Help Center](https://help.openai.com/en/articles/9237897-chatgpt-search), [Google AI Mode](https://search.google/ways-to-search/ai-mode/), [Claude Help Center](https://support.claude.com/en/articles/10684626-enable-and-use-web-search), [Perplexity Docs](https://docs.perplexity.ai/), [Gemini Google Search grounding](https://ai.google.dev/gemini-api/docs/google-search), [Microsoft Learn](https://learn.microsoft.com/en-us/copilot/privacy-and-protections)).

## AI visibility measurement starts with a fixed query universe

**AI visibility is only measurable against a declared query set.** A brand cannot have one universal visibility rate because answer engines respond differently to vendor-seeking, problem-aware, category-definition, and comparison queries.

The first step is to freeze the query universe before the run: the exact prompts, engines, date, locale, mode, and inclusion rules. AuthorityTech's September 11, 2026 AI-engine panel used 35 active queries and found portfolio presence in 31 of them, an 89% query-presence rate. That rate is meaningful only because the denominator is declared.

## Answer presence is not the same as brand mention rate

**Answer presence measures whether the brand or frame appears anywhere in the response.** Brand mention rate is narrower: whether the answer names the brand in the answer text.

This distinction matters because AI answers can carry a category frame without naming a company, can cite a company's page without recommending the company, or can mention a company without citing any supporting source. A measurement system that reports only a headline visibility score hides those differences.

## Share of citation measures the evidence layer

**Share of citation measures how often a domain or brand is used as cited source evidence.** Machine Relations defines [share of citation](https://machinerelations.ai/glossary/share-of-citation) as a descriptive citation-presence rate for a declared query set, engine set, and observation window.

That is why share of citation is not the same object as share of voice. [AI share of voice](https://machinerelations.ai/research/how-to-measure-ai-search-visibility-brand-share-of-voice) measures breadth of brand mention or presence. Share of citation measures whether the answer system used the brand, publication, or page as a source.

## Retrieval state is the missing baseline variable

**Retrieval state tells the team whether the answer was source-steered or model-memory-steered.** Retrieval-augmented generation was originally defined as combining parametric model memory with retrieved non-parametric memory, which is the operational reason source-layer measurement exists ([Lewis et al., 2020](https://arxiv.org/abs/2005.11401)).

If retrieval or web search did not occur, a missing citation is not necessarily a source-quality failure. If retrieval did occur and the engine selected competitor or generic sources, the program has a source-graph problem. The source-layer baseline records this before interpretation.

## Source quality belongs beside source count

**A citation count without source quality is an incomplete AI visibility metric.** One citation from a high-trust, category-relevant publication can be more useful than several citations from thin, off-topic, or duplicate pages.

The public [Machine Relations Index](https://machinerelations.ai/machine-relations-index) exists to make source analysis auditable across the market. The September 2026 MRI v2 release added a larger measurement base: 119,397 source events across 21,372 domains. That scale is useful because source behavior is sparse, uneven, and engine-specific.

Source quality also depends on whether the cited page is useful, original, and clearly structured. Google's helpful-content guidance tells publishers to create people-first content, and Schema.org's Article vocabulary gives machines a standard way to describe article entities and metadata ([Google Search Central](https://developers.google.com/search/docs/fundamentals/creating-helpful-content), [Schema.org Article](https://schema.org/Article)).

## A baseline separates five AI visibility denominators

**The source-layer baseline keeps the score from lying by compression.** These five denominators should be reported separately before any composite score is used.

| Denominator | What it answers | Why it must stay separate |
|---|---|---|
| Query universe | Which prompts were measured? | Changes in prompt mix can change every rate |
| Answer presence | Did the brand, domain, or frame appear? | Shows coverage, not evidence strength |
| Brand mention rate | Did the answer name the brand? | Captures prominence, not citation support |
| Share of citation | Did the answer cite the brand or domain? | Captures evidence-layer selection |
| Retrieval state | Did the engine search or retrieve? | Separates source work from model memory |

## Composite AI visibility scores should come after the baseline

**A composite score is an interface, not the measurement contract.** It can help executives read a dashboard, but it should never be the only exported number.

The source-layer baseline makes the composite falsifiable. If a score rises while cited-source share is flat, the gain probably came from mentions or query mix. If cited-source share rises while model-layer mentions stay low, the source graph may be improving before the model's latent answer behavior changes.

## The September 11 panel shows why the source layer matters

**The September 11 AI-engine panel separated presence from source selection.** It observed 31 present queries out of 35 active queries, 31% RAG share of citation, 0% model-layer share, and six Machine Relations propagation observations.

The practical read is not that one number is good and another is bad. It is that the near-term operating layer is retrieval and cited-source selection. The model layer may remain quiet while the source layer starts moving.

## TREG demand shows the measurement surface is still open

**Demand exists, but current answers do not yet cite owned Machine Relations pages reliably.** A fresh TREG topic-citations probe for the exact keyword `how to measure ai visibility` returned 482 cited domains and no owned-cited domains in the top result set. The same parent probe showed low but nonzero keyword volume for `how to measure ai visibility` and `ai visibility measurement`.

That is a useful demand signal, not a universal market truth. It says the measurement query is active enough to justify a clean answer and open enough that owned citation coverage still has room to improve.

## Machine Relations uses source-layer measurement because the reader changed

**Machine Relations treats visibility as a source system because AI engines are source selectors.** The Princeton-led GEO paper describes generative engines as systems that synthesize information from multiple sources, which gives content creators less direct control than ranking-based search while making source selection the central visibility problem ([GEO paper](https://arxiv.org/abs/2311.09735)).

That is why [AI visibility](https://machinerelations.ai/glossary/ai-visibility) should not be reduced to a mention counter. The answer system must be able to retrieve, parse, cite, and trust the evidence layer.

## How to run the baseline in practice

**A usable AI visibility baseline can be run in one afternoon.** Use the same prompts, same engines, same locale, and same timestamp window, then record one row per answer.

Each row should include: prompt, engine, mode, retrieval state if available, answer text, named brands, cited URLs, cited domains, source type, and whether the cited source supports the claim. If the page publishes structured data, record the schema type as a source-quality field; JSON-LD is a W3C recommendation for linked data in JSON ([W3C JSON-LD 1.1](https://www.w3.org/TR/json-ld11/)). Then calculate the five denominators separately before interpreting the composite.

## FAQ

### What is the best way to measure AI visibility?

The best way to measure AI visibility is to use a fixed query set and report answer presence, brand mentions, cited-source share, retrieval state, and source quality separately. A single score can be useful only after those denominators are visible.

### Is AI visibility the same as share of voice?

No. Share of voice is one visibility measure, usually focused on brand presence or mention share. AI visibility also includes cited-source selection, retrieval behavior, source quality, and engine-specific answer composition.

### Why does share of citation matter for AI visibility?

Share of citation matters because cited sources are the evidence layer AI answers expose to users. A brand can be mentioned without being cited, and a cited source can influence the answer even when the prose does not read like a ranking result.

### Can one AI visibility score compare two tools?

Not safely unless both tools disclose the same query universe, engine mix, retrieval state, denominator, deduplication rule, and uncertainty treatment. Without that contract, two scores may be measuring different objects.

### Where does Machine Relations fit in AI visibility measurement?

Machine Relations is the discipline that connects earned authority, entity clarity, source architecture, and answer-engine citation. It treats AI visibility as a source-selection problem rather than only a ranking, traffic, or mention problem.

## Attribution

This research is published by Machine Relations Research, the research program of machinerelations.ai — the public research and standards initiative that publishes the glossary, research, evidence, and measurements for the Machine Relations discipline. Provenance and editorial standards: https://machinerelations.ai/about

## Machine-readable related links

### Related concepts

- [Machine Relations Index (MRI)](https://machinerelations.ai/glossary/machine-relations-index)
- [Machine Relations (MR)](https://machinerelations.ai/glossary/machine-relations)
- [Share of Citation](https://machinerelations.ai/glossary/share-of-citation)
- [AI Visibility](https://machinerelations.ai/glossary/ai-visibility)

### Supporting research

- [Measuring Entity Chain ROI: How B2B Teams Quantify AI Visibility Gains in 2026](https://machinerelations.ai/research/entity-chain-measurement-roi-b2b-ai-visibility-2026)
- [AI Search Visibility Measurement Framework: Metrics, Tools, and Tracking Methods for 2026](https://machinerelations.ai/research/ai-search-visibility-measurement-framework-2026)
- [AI Visibility Methodology Transparency: What 16 Platforms Actually Disclose](https://machinerelations.ai/research/ai-visibility-methodology-transparency-benchmark-2026)
- [How to Measure Entity Chain Strength for AI Citation Eligibility](https://machinerelations.ai/research/how-to-measure-entity-chain-strength-ai-citation-eligibility-2026)

### Framework context

- [Machine Relations Index](https://machinerelations.ai/index)
- [Machine Relations Stack](https://machinerelations.ai/stack)
- [Evidence Base](https://machinerelations.ai/evidence)