AI visibility is not one metric. A brand can be cited by an AI system without being named, mentioned by ChatGPT but absent from Google AI results, or widely recognised by models even when it does not appear for a particular set of tracked keywords.
That is why our platform measures AI presence through three separate scores. Each answers a different business question, uses a different data source, and should be interpreted independently.
This guide explains what each score means, how it is calculated, what it does not measure, and how to use the results to improve your brand’s visibility in AI-powered search and answer engines.
Why we do not use one score
A single “AI visibility score” can be misleading.
For example, imagine a financial-advice site whose content is repeatedly used as a source in AI-generated answers. Its pages may be cited frequently, but the AI response may not explicitly name the business. The site has strong source authority, yet weak brand visibility.
Alternatively, a well-known consumer brand may be frequently mentioned by AI systems but rarely cited as a direct source. The brand is recognizable, but its website may not be earning source attribution.
These are different situations that require different actions. Combining them into one number would hide the insight.
Our platform therefore separates AI measurement into:
| Feature | Primary question it answers | Data source | Measurement scope |
|---|---|---|---|
| AI Visibility Tracker | Does Google’s AI experience mention and cite my brand for my tracked prompts? | DataForSEO | Tracked prompts on the configured Google AI surface |
| LLM Mentions | Which keywords, topics, and AI platforms mention my brand or domain? | DataForSEO | Keyword-level results across supported AI platforms |
| Brand Visibility Report | Do leading language models recognize, trust, and describe my brand positively? | Direct Gemini, Perplexity, and OpenAI API calls | Brand-level perception, independent of keywords |
DataForSEO treats AI mentions and cited sources as distinct signals, while its LLM-oriented tools are designed to analyse brand, domain, and keyword mentions across AI search contexts. docs.dataforseo
1. AI Visibility Tracker
The AI Visibility Tracker appears in the AI Visibility tab. It measures performance for a defined set of tracked prompts on the configured Google AI surface.
It is designed to answer:
“When people ask the questions we are tracking, is our brand visible, cited, and positively represented?”
The formula
The tracker calculates the AI Visibility Score (AIVS) as:
[\text{AIVS} =
0.40 \times \text{Brand Visibility Rate} +
0.30 \times \text{Sentiment} +
0.30 \times \text{Owned Citation Coverage}]
Each component is expressed on a 0–100 scale before weighting.
| Component | Weight | What it means |
|---|---|---|
| Brand Visibility Rate | 40% | Percentage of tracked prompts where the brand is explicitly mentioned |
| Sentiment | 30% | Tone of AI references to the brand: positive, neutral, or negative |
| Owned Citation Coverage | 30% | Percentage of prompts where the client’s domain is cited as a source |
Brand Visibility Rate
Brand Visibility Rate is the percentage of tracked prompts where the AI-generated response explicitly refers to your brand.
If your brand appears in 18 out of 50 tracked prompts:
[\text{Brand Visibility Rate} = \frac{18}{50} \times 100 = 36]
This is the most heavily weighted component because direct brand presence is the clearest form of visibility.
A cited webpage is not automatically the same as a brand mention. The source may appear in a citation list while the answer text does not name the company.
Sentiment
Sentiment evaluates the tone used when the AI response discusses your brand.
- Positive: The response associates the brand with trust, expertise, quality, leadership, usefulness, or a favourable recommendation.
- Neutral: The response mentions the brand factually without a clear positive or negative judgement.
- Negative: The response associates the brand with criticism, poor quality, complaints, risk, controversy, or an unfavourable comparison.
Sentiment is calculated only from responses where the brand is actually mentioned. If the brand is not mentioned, there is no meaningful sentiment to classify.
Owned Citation Coverage
Owned Citation Coverage measures how often the AI response cites a page from your own domain as a source.
If your domain is cited for 15 of 50 tracked prompts:
[\text{Owned Citation Coverage} = \frac{15}{50} \times 100 = 30]
This metric indicates that your content is being used as supporting evidence or source material, even if the brand is not named prominently in the generated answer.
DataForSEO’s AI-oriented products distinguish between mentions and sources/citations, which is why we retain both signals in the score rather than treating them as the same outcome. docs.dataforseo
Conservative and available-data modes
Sometimes sentiment cannot be calculated because the brand was not mentioned in any response.
Our default is conservative mode:
- Unmeasurable sentiment receives a score of 0.
- The full formula remains unchanged.
- The score does not increase simply because a difficult-to-measure component is excluded.
This is the recommended mode for client reporting because it avoids overstating visibility for brands with little or no observable presence.
We also support an available-data weighted mode:
- The unavailable 30% sentiment weight is redistributed proportionally between Brand Visibility Rate and Owned Citation Coverage.
- This is useful for diagnostic work, early-stage campaigns, and low-mention brands.
- It should not be compared directly with conservative-mode scores unless the report clearly labels the mode.
For transparency, the platform should clearly state when sentiment is unavailable, for example:
“Sentiment could not be measured because the brand was not mentioned in the analysed responses. This score uses conservative weighting.”
2. LLM Mentions
The LLM Mentions section is a keyword-level measurement tool. It uses DataForSEO LLM Mentions data to evaluate visibility across supported AI platforms, including ChatGPT, Perplexity, and the configured Google AI product.
It is designed to answer:
“For the topics and keywords that matter to us, which AI platforms mention our brand—and where are the gaps?”
DataForSEO describes its LLM Mentions capabilities as providing metrics for keyword, brand, and website mentions in LLM environments, along with related sources and AI-search metrics. docs.dataforseo
What the report shows
The LLM Mentions view includes:
- Overall LLM Visibility: An aggregate summary across the selected keywords and available platforms.
- Platform-level visibility: Separate results for each supported AI system.
- Keyword-level results: Whether the brand or domain appears for each tracked keyword.
- Gap Matrix: A visual map of which keywords are missed on which platforms.
- Scan cost: Approximately $0.0006 per keyword per scan, subject to the active provider configuration and pricing.
The Gap Matrix is particularly useful because it turns an abstract score into an action list.
| Keyword | ChatGPT | Perplexity | Google AI | Priority |
|---|---|---|---|---|
| Best retirement planner | Mentioned | Missed | Mentioned | Improve Perplexity coverage |
| Tax-saving investments | Missed | Missed | Mentioned | High-priority topic gap |
| Mutual fund advisory | Mentioned | Mentioned | Mentioned | Maintain and expand |
A single aggregate score can indicate whether visibility is improving. The Gap Matrix identifies which platform-topic combinations need work.
What counts as a mention
A mention may refer to:
- An explicit brand-name reference.
- A domain reference.
- A recognised business/entity reference.
- A platform-specific target match, depending on the configured DataForSEO query and matching logic.
The report should always be read with its exact target definition in mind. A brand-name mention, a domain citation, and an unlinked textual reference are related but not identical signals.
Why platform-level reporting matters
AI systems do not retrieve, rank, cite, or generate answers in the same way.
A brand can perform strongly in one environment and poorly in another because of differences in:
- Retrieval sources and source-selection methods.
- Entity recognition.
- Freshness and indexing cycles.
- Geographic and language context.
- Prompt interpretation.
- Citation behaviour.
- Model behaviour and answer formatting.
That is why a cross-platform average should never replace the per-platform breakdown.
3. Brand Visibility Report
The Brand Visibility Report is a separate, proprietary measurement. It uses direct calls to Gemini, Perplexity, and OpenAI models to assess how those models recognize and describe a brand based on its domain.
It is designed to answer:
“If an AI system is asked about our company, does it know who we are, does it consider us credible, and how does it describe us?”
Unlike the AI Visibility Tracker and LLM Mentions, this report is keyword-independent. It does not test whether the brand appears for a particular collection of search queries.
The formula
[\text{Brand Visibility Score} =
0.55 \times \text{Awareness} +
0.25 \times \text{Credibility} +
0.20 \times \text{Sentiment}]
| Component | Weight | What it measures |
|---|---|---|
| Awareness | 55% | Whether models recognise the brand, business, domain, products, and category association |
| Credibility | 25% | Whether models associate the brand with expertise, legitimacy, authority, trust, or reliable information |
| Sentiment | 20% | Whether descriptions of the brand are positive, neutral, mixed, or negative |
Awareness
Awareness is the foundation of the report.
A brand cannot be considered visible in a meaningful sense if leading models do not recognize it, cannot identify its category, or confuse it with another business.
Awareness may be assessed through signals such as:
- Correct identification of the business or domain.
- Accurate category and service descriptions.
- Recognition of key products, expertise, or audience.
- Consistency of answers across model providers.
- Ability to distinguish the brand from similarly named entities.
Credibility
Credibility evaluates whether model responses portray the business as reliable and authoritative.
Relevant signals may include references to:
- Demonstrable expertise.
- Quality of information.
- Professional reputation.
- Established market presence.
- Verifiable trust signals.
- Third-party recognition.
- Clear and consistent business identity.
This is not a direct measure of Google rankings, backlinks, reviews, or revenue. It is a measure of the way models describe the brand when prompted.
Sentiment
Sentiment evaluates the tone of responses across the sampled models.
A high awareness score with poor sentiment is a reputational warning. A low awareness score with neutral sentiment often means that the brand is simply underrepresented in the data and sources models use.
Why results are cached on the same day
Direct LLM results can vary between refreshes because models, retrieval systems, safety layers, sampling behaviour, and provider-side configuration can change.
To prevent a client seeing inconsistent results during the same reporting period, we cache Brand Visibility Report outputs for the day.
This supports:
- Consistency: The same report produces the same result during the cache period.
- Comparability: Stakeholders can discuss one stable score rather than multiple fluctuating refreshes.
- Auditability: The system can retain the timestamp, prompt version, provider, and model metadata used to create the score.
- Cost control: It avoids unnecessary repeated calls to provider APIs.
DataForSEO similarly describes LLM-response analysis as a distinct capability for analysing how models respond to questions about brands, competitors, topics, and keywords. Our Brand Visibility Report applies that principle using direct multi-provider model calls. docs.dataforseo
The most important insight: cited but invisible
One of the most actionable patterns in AI visibility is the “cited but invisible” gap.
This occurs when:
- Your domain has high Owned Citation Coverage.
- Your Brand Visibility Rate is zero or very low.
- AI systems use your content as a source but do not explicitly identify your brand in the generated response.
In simple terms, your content has authority, but your brand is not receiving enough attribution.
Why it happens
This pattern can result from several issues:
- Content is useful but generic, so the answer uses the facts without naming the publisher.
- Page titles and headings do not clearly connect the content to the brand.
- Author, organization, and editorial information are weak or inconsistent.
- The site lacks distinctive research, proprietary data, tools, frameworks, or expert viewpoints.
- The business entity is poorly defined across the website and third-party sources.
- The tracked prompts are informational and do not naturally create a reason to mention a provider by name.
Why it matters
Citations are valuable. They signal that your content is useful enough to support an AI answer.
But if business growth depends on recognition, trust, direct traffic, leads, or demand generation, citation alone may not be enough. Users need to understand who produced the information and why that brand is worth remembering.
How to improve the gap
Prioritise actions that strengthen the connection between useful content and the business behind it:
- Add clear authorship, expert credentials, editorial policies, and organization information.
- Use consistent brand naming across titles, bylines, headers, footers, and company pages.
- Publish original research, calculators, proprietary frameworks, data studies, and expert analysis.
- Improve entity clarity through well-maintained Organization, Person, and author information where appropriate.
- Build credible third-party references, editorial mentions, reviews, and expert citations.
- Create topic clusters that connect informational content to your brand’s differentiated services, products, methodology, or point of view.
- Review the exact prompts where citations occur and determine whether the content gives AI systems enough reason to identify the brand explicitly.
How to interpret common patterns
| AI Visibility Tracker | LLM Mentions | Brand Visibility Report | Likely interpretation |
|---|---|---|---|
| High | High | High | Strong visibility, broad keyword coverage, and high model recognition |
| High citations, low mentions | Mixed | Moderate or low | Content is being sourced, but brand attribution and entity association need work |
| Low | High | High | Brand is recognized in some AI systems but weak for the Google AI prompt set being tracked |
| Low | Low | High | Models know the brand when asked directly, but it has weak keyword-level discoverability |
| High | Low | Low | Visibility may be narrow, prompt-specific, or driven by a small number of source pages |
| Low | Low | Low | Low AI-search presence and limited model recognition; build topical authority and entity signals first |
| High | High | Low sentiment | The brand is visible but may have a reputation, accuracy, or positioning issue to investigate |
What our scores do not claim
To interpret the tool responsibly, it is important to understand what these scores do not measure.
They are not direct measures of:
- Google organic rankings.
- Website traffic or conversions.
- Revenue, leads, or commercial performance.
- A guaranteed inclusion in future AI answers.
- Every prompt a user could ask.
- Every AI platform in the market.
- Model training-data inclusion.
- The objective truth of a model-generated statement.
AI outputs can change over time. The purpose of the platform is to provide a structured, repeatable way to monitor visibility, citations, coverage gaps, and model perception—not to promise a fixed position in a generative answer.
Methodology and transparency
Every AI visibility score should be interpreted alongside its measurement context:
- Data source: DataForSEO or direct model APIs.
- AI platform: Google AI Overview, Google AI Mode, ChatGPT, Perplexity, Gemini, OpenAI, or another configured source.
- Target: Brand name, domain, entity, or keyword.
- Prompt or keyword set: The exact list being measured.
- Location and language: Where applicable.
- Sample size: Number of prompts, keywords, or responses analysed.
- Date and scan time: AI results can evolve quickly.
- Scoring mode: Conservative or available-data weighted.
- Methodology version: The calculation rules and weights used for that report.
We recommend comparing scores over time only when these inputs remain broadly consistent. If the keyword set, location, language, model provider, matching logic, or scoring weights change, treat the result as a new baseline rather than a direct continuation of the previous trend.
The bottom line
Our platform does not ask one vague question—“Are you visible in AI?” Instead, it answers three specific questions:
- Are you mentioned and cited for the Google AI prompts that matter to your business?
- Which keywords and AI platforms are creating visibility opportunities or gaps?
- Do major AI models recognize, trust, and describe your brand positively when asked directly?
Together, these views provide a more complete picture of AI visibility: not only whether your content is present, but whether your brand is being seen, cited, trusted, and remembered.