3D Chess Media, LLC develops and operates research properties focused on how consumers and businesses discover, compare, evaluate, and choose products, services, and companies. Our research activities can include:
- AI search measurement
- AI recommendation analysis
- Citation research
- Competitive benchmarking
- Consumer research
- Business research
- Industry reports
- Longitudinal datasets
- Comparative analysis
- Consensus studies
Different properties within The 3D Chess Media Portfolio may use different methodologies. This page explains the principles that generally guide our research. Individual studies and publications may provide additional methodology specific to the research being presented.
Start With the Question
Useful research begins with a clearly defined question. In AI search research, that question is often expressed as a prompt.
Examples might include:
- What are the best medical alert systems for seniors living alone?
- Which CRM platforms are best for small businesses?
- What are the best alternatives to Company A?
- Which mortgage lenders are best for first-time homebuyers?
- What companies do AI systems recommend for a specific commercial need?
The wording matters. Changing: best company, to: cheapest company, or best company for enterprise buyers, can produce a very different result. For that reason, prompt design is a material part of AI search research.
Focus on Commercially Relevant Questions
A company can appear frequently in AI-generated answers without appearing when the user is actually considering a purchase. 3D Chess Media research often places particular emphasis on high-intent prompts. These can include:
- Best-company questions
- Best-product questions
- Company comparisons
- Alternatives
- Pricing
- Cost
- Reputation
- Trust
- Buyer-specific recommendations
- Use-case recommendations
- Industry-specific recommendations
- Product-selection questions
- Provider-selection questions
These prompts are valuable because they occur closer to an actual buying decision.
Prompt Clusters
Rather than rely on a single question, research may organize related prompts into prompt clusters. A prompt cluster represents a group of questions addressing a similar commercial intent.
For example, a comparison cluster might contain questions such as:
- Company A vs. Company B
- Is Company A better than Company B?
- Should I choose Company A or Company B?
- Which is better for small businesses, Company A or Company B?
This provides a broader view than relying on one isolated prompt.
Multiple AI Systems
Where appropriate, our research may test the same or substantially similar prompts across multiple AI or AI-search systems. Platforms studied may include major: Conversational AI systems, AI assistants, Answer engines, AI-search platforms, and Search engines with generative features. The platforms included can change over time. A research report should identify the systems used when that information is material to interpreting the findings.
Why We Test More Than One AI Platform
Different AI systems can produce different answers to the same question. One platform may recommend: Company A, while another recommends: Company B, and another may not recommend either company. That disagreement is itself useful information. Testing across multiple systems helps us measure:
- Cross-platform visibility
- Recommendation consistency
- Competitive differences
- Platform-specific strengths
- Platform-specific weaknesses
- Areas of agreement
- Areas of disagreement
The objective is not to assume that one platform represents the entire AI search market.
Recommendation Measurement
When analyzing AI responses, we may distinguish among several different forms of visibility. These can include:
Mention
The company appears somewhere in the answer.
Recommendation
The company is presented as a relevant choice or option.
Ranked Recommendation
The company appears within an ordered list or identifiable recommendation position.
Rank One
The company receives the highest identifiable recommendation position.
Top Three
The company appears within the first three identifiable recommendations.
Comparison Inclusion
The company is included when the user asks for competitors, alternatives, or comparisons. These distinctions matter. A brand mentioned in passing is not necessarily receiving the same commercial visibility as the company an AI system recommends first.
Recommendation Share
Some research may calculate a form of recommendation share or visibility share. Conceptually, this asks: Of the relevant recommendation opportunities measured, what percentage included this company? Depending on the study, calculations may account for factors such as:
- Number of prompts
- Number of AI systems
- Recommendation frequency
- Recommendation position
- Prompt category
- Commercial intent
The exact calculation should be defined in the applicable methodology.
Position Matters
Two companies can both appear frequently while experiencing very different levels of visibility. For example: Company A appears in 80% of responses and is usually ranked first. Company B appears in 80% of responses but is usually ranked fifth. A simple mention count would treat those companies as equal. A ranking-sensitive analysis would not. For that reason, our research may separately measure:
- Overall visibility
- Rank-one visibility
- Top-three visibility
- Average identifiable position
- Recommendation concentration
Citation Research
AI-generated answers may include citations, links, references, or identifiable source material. Where citations are available, research may examine:
- Which domains are cited
- Which URLs are cited
- How frequently a domain appears
- Which sources appear across multiple prompts
- Which sources appear across multiple AI systems
- Whether competitors share common sources
- Whether particular publishers appear disproportionately often
- Whether first-party or third-party sources are being used
Citation research can help answer a different question from recommendation research. Recommendation research asks: Which companies are appearing in the answer? Citation research asks: Which sources appear to support or accompany the answer? Those are related, but they are not identical.
Citation Occurrence vs. Unique Source
A single website may be cited many times. Accordingly, citation research may distinguish among:
- Total citation occurrences
- Unique domains
- Unique URLs
- Responses containing a citation
- Cross-platform citation coverage
For example, 100 citations from one domain represent something different from 100 citations spread across 80 independent domains. Both measurements can be useful.
Source Influence
Some research may explore whether particular sources appear repeatedly around certain companies, industries, or recommendation outcomes. This can help identify:
- Frequently cited publishers
- Frequently cited company pages
- Comparison websites
- Review sites
- Research sources
- Forums
- Communities
- News organizations
- Industry publications
- Other influential information sources
Where we discuss potential source influence, we should distinguish observed association from proven causation. A website appearing frequently alongside a recommendation does not automatically prove that the website caused the recommendation.
Consensus Research
Several properties within The 3D Chess Media Portfolio may analyze agreement among multiple AI systems. A consensus result asks: How many of the AI systems tested reached a similar recommendation or conclusion? For example: 6 of 7 systems recommended Company A. That is an observable result within the defined test. It can be useful. But it should not be overstated.
Consensus Does Not Mean Objective Truth
AI agreement is not the same as independent expert agreement. Different AI systems may rely on overlapping: Search indexes, Training information, Websites, Publishers, Reviews, Company information, Research, and Public data. For that reason: 6 of 7 AI systems recommended Company A, means: six of the seven systems tested produced a recommendation for Company A under the methodology used. It does not necessarily mean: six independent experts proved Company A is objectively the best company. This distinction is fundamental to our approach.
Preserve Disagreement
Research becomes less useful if disagreement is removed merely to create a cleaner story. If different AI systems disagree, we want the research to show that. For example: ChatGPT recommends Company A. Gemini recommends Company B. Claude recommends Company C. That may be more informative than forcing the result into a single winner. Disagreement can reveal:
- Market fragmentation
- Platform differences
- Source differences
- Prompt sensitivity
- Brand-positioning issues
- Weak category consensus
We consider disagreement a research result, not a research failure.
Repeatability and AI Variability
AI systems can produce different responses to the same prompt at different times. Responses may vary because of:
- Model updates
- Search updates
- Retrieval differences
- Changing web content
- Randomness
- Personalization
- Geographic differences
- Account settings
- Platform experimentation
- Prompt interpretation
For that reason, AI research should generally be treated as a measurement taken under defined conditions rather than a permanent statement about future behavior.
Date Matters
AI search changes quickly. A study conducted in January may not produce the same result in September. Where practical, material research should identify:
- Collection date
- Research period
- Version
- Update date
- Platforms tested
This helps readers understand when the observation was made.
Longitudinal Research
One of the most valuable uses of AI search measurement is tracking change over time. Rather than asking only: Who is visible today?
longitudinal research can ask:
- Who is gaining visibility?
- Who is losing visibility?
- Which competitors are improving?
- Which sources are becoming more important?
- Are AI systems becoming more consistent?
- Did a company's position change after its information environment changed?
Repeated measurement can reveal trends that are invisible in a one-time snapshot.
Baselines
Before evaluating improvement, it is useful to establish a baseline. A baseline records conditions before a significant intervention or time period. Depending on the research, a baseline may include:
- Recommendation share
- Prompt coverage
- Rank-one visibility
- Top-three visibility
- Competitive visibility
- Citation occurrences
- Unique sources
- Brand framing
- Platform differences
Future measurements can then be compared with that baseline.
Correlation Is Not Causation
This principle is especially important in AI search research. Suppose a company: Publishes new research., Earns new third-party coverage., Improves technical SEO., Adds clearer product pages., and Gains AI recommendation visibility.. The increased visibility occurred after those changes. That does not automatically prove which specific change caused the result. Other factors may also have changed. Accordingly, we try to distinguish language such as: visibility increased after the intervention, from stronger claims such as: the intervention caused the increase.
Causal claims require stronger evidence than simple before-and-after observations.
Modeled Values
Some research may use models, estimates, or calculated opportunity values. Examples can include: Estimated market opportunity, Modeled traffic, Estimated commercial value, Projected visibility, Weighted scores, and Composite metrics. When a number is modeled rather than directly observed, it should be identified as modeled or estimated where material. A modeled value should not be presented as though it were:
- Audited revenue
- Actual traffic
- Guaranteed savings
- Guaranteed sales
- Observed financial performance
The assumptions matter.
Automated Research
Some research conducted by 3D Chess Media may involve automated systems. Automation can be used for:
- Prompt execution
- Response collection
- Data normalization
- Citation extraction
- Classification
- Competitive matching
- Report generation
- Longitudinal comparison
- Data visualization
Automation allows larger datasets to be studied than would be practical through completely manual research. It also creates its own risks. Those risks can include:
- Classification errors
- Parsing errors
- Duplicate records
- Incorrect company matching
- Platform failures
- Missing citations
- Formatting changes
For that reason, automation should be combined with validation and quality-control procedures appropriate to the study.
Human Review
Human review may be used to evaluate areas such as:
- Research logic
- Classification accuracy
- Company matching
- Outliers
- Material factual claims
- Industry terminology
- Interpretation
- Methodological compliance
- Results that appear inconsistent
Human review does not mean a reviewer should replace an observed AI result with the answer they personally prefer. If an AI system recommended Company B, human review should not simply change the raw research record to Company A because the reviewer believes Company A is better. The observed response should remain the observed response.
Subject-Matter Review
Some portfolio properties may involve industries where subject-matter expertise is particularly useful. A qualified reviewer may help evaluate:
- Context
- Terminology
- Consumer implications
- Industry-specific limitations
- Material factual issues
- Potentially misleading interpretations
The reviewer should not be presented as independently creating results produced by AI systems. Where material, individual properties should explain the role of the reviewer.
Data Quality
Research quality depends on the quality of the underlying data. Depending on the study, quality-control processes may include:
- Prompt validation
- Duplicate detection
- Company-name normalization
- Manual spot checks
- Outlier review
- Citation validation
- Platform-response validation
- Date tracking
- Version control
- Methodology documentation
Not every research project requires the same process. The level of quality control should be appropriate to the significance of the conclusions being drawn.
Company and Brand Matching
AI systems may refer to the same company in different ways. For example: ABC Software.
ABC Technologies, could potentially refer to the same organization. Research may therefore require entity normalization so that obvious variants are treated consistently. At the same time, normalization should not incorrectly combine separate companies simply because their names are similar. Entity matching can require both automated rules and human review.
Brand Families
Some companies operate multiple products or brands. A methodology should determine whether research is measuring:
- Parent company
- Individual brand
- Individual product
- Business unit
- Website
- Service line
That decision can materially affect the result. Where relevant, the unit being measured should be identified.
Prompt Selection Bias
The prompts selected for a study influence the result. A dataset containing only: cheapest provider, questions may favor a different group of companies than a dataset containing: best enterprise provider, questions. No prompt set represents every possible customer question. For that reason, research should be interpreted within the scope of the prompts actually tested.
Platform Selection Bias
The AI platforms selected also affect the result. If a study analyzes five systems rather than ten, the consensus result reflects those five systems. Adding or removing a platform can change:
- Recommendation share
- Consensus
- Rank distribution
- Citation coverage
- Competitive position
Accordingly, the platform set is part of the methodology.
Commercial Relationships and Research Integrity
3D Chess Media operates commercial businesses. Some companies included in our research may:
- Be clients
- Be advertisers
- Be affiliate partners
- Sponsor research
- Purchase data
- Have referral relationships
- Have no commercial relationship with us at all
Commercial relationships can create conflicts of interest. Pretending those conflicts do not exist does not eliminate them. Our preferred approach is: disclose the relationship, and preserve the underlying result.
Commercial Partners Do Not Receive Extra AI Votes
If an AI system does not recommend a commercial partner, the research should not manufacture a recommendation. If the observed result is: 2 of 7 systems recommended Partner A, a commercial relationship should not transform the result into: 6 of 7 systems recommended Partner A. Likewise, if a non-partner receives the strongest result, that result should remain visible.
Related 3D Chess Media Businesses
Research conducted within The 3D Chess Media Portfolio may occasionally concern industries in which another 3D Chess Media business participates. For example: CiteWorks Studio could appear in research about AI search agencies., and LLM Authority Index could appear in research about AI visibility platforms.. That relationship represents a potential conflict. Our preferred response is not to remove the related company automatically or to conceal ownership. It is to: Apply the defined methodology., Preserve the observed result., and Disclose the relationship..
Research Sponsorship
A company may financially support a research project. When sponsorship is material, it should be disclosed. A sponsor may fund activities such as:
- Data collection
- Surveys
- Research production
- Analysis
- Visualization
- Publication
- Distribution
Sponsorship should not purchase a predetermined finding. Research should remain capable of producing a result the sponsor does not prefer.
Original Research
3D Chess Media believes original research is one of the most valuable assets a digital business can create. Strong original research can:
- Answer questions that were previously difficult to answer
- Create proprietary datasets
- Identify market trends
- Support journalists
- Support business decisions
- Reveal competitive changes
- Create historical benchmarks
- Improve public understanding of emerging markets
Our objective is not simply to produce more content. It is to create information that did not previously exist in the same form.
Research Publication
Depending on the project, research may be published as:
- Industry reports
- Benchmark studies
- Citation indexes
- Company reports
- Competitive reports
- Articles
- Data tables
- Interactive dashboards
- Charts
- Research briefs
- Longitudinal updates
- Public datasets
- Commercial research products
Different formats may present different levels of detail. Where a summarized publication relies on a larger dataset, readers should interpret the summary within the limits described by the methodology.
Updates and Corrections
Research may be updated when:
- New data becomes available
- AI platforms change
- A material error is identified
- Classification is corrected
- A methodology changes
- A new research period is completed
Where practical, material methodological changes should be distinguished from routine data updates. If a correction materially changes a conclusion, the publication should be updated accordingly.
Methodologies Can Evolve
AI search is a rapidly developing field. There is no reason to assume that the best measurement framework used today will remain unchanged indefinitely. As our datasets grow, we may revise:
- Metrics
- Weighting systems
- Prompt taxonomy
- Citation classifications
- Platform coverage
- Data collection methods
- Quality-control systems
- Reporting structures
Methodological improvement is expected. Material changes should be documented where they affect comparisons with earlier research.
Proprietary Metrics
3D Chess Media and its businesses may develop proprietary metrics or frameworks. These may include measures relating to:
- Recommendation share
- Citation visibility
- Recommendation concentration
- Cross-platform agreement
- Source influence
- Competitive coverage
- Rank distribution
- Recommendation proximity
- Citation centrality
Unless specifically established otherwise, these should not be presented as universally accepted industry standards or official ranking factors used by AI platforms. They are research frameworks designed to help interpret observed data.
What Our Research Can Tell You
Depending on the study, our research may help answer questions such as:
- Which companies are AI systems recommending?
- Which competitors appear most frequently?
- Who receives rank-one recommendations?
- Which brands dominate high-intent prompts?
- Which websites are cited most frequently?
- Where do AI systems agree?
- Where do they disagree?
- Which companies are gaining or losing visibility?
- How has a market changed over time?
What Our Research Cannot Automatically Tell You
A dataset does not automatically prove:
- why an AI system made a particular recommendation
- that a cited source caused a recommendation
- that an AI recommendation is objectively correct
- that historical results will continue
- that correlation proves causation
- that every possible prompt would produce the same outcome
- that every consumer should make the same purchase decision
Research should be used for what it actually measures.
Measurement Before Assumption
The principle underlying much of our research is straightforward: Measure what the systems actually do before constructing a theory about what they might do. We prefer observed data over confident assumptions. Then we use the observed data to develop hypotheses that can be tested again.
The cycle is:
- Measure
- Compare
- Identify patterns
- Develop hypotheses
- Make changes where appropriate
- Measure again
This approach is central to the research businesses operated by 3D Chess Media.
Research Businesses Within The 3D Chess Media Portfolio
LLM Authority Index
AI search intelligence, recommendation measurement, citation research, competitive benchmarking, industry datasets, and longitudinal tracking.
The 3D Chess Media Portfolio
Category-specific research properties applying structured comparative and consensus methodologies across consumer and business markets.
Research Questions and Corrections
Questions concerning a specific study should ideally identify:
- Website
- Research title
- URL
- Publication date
- Question about the methodology
- Data point being questioned
- Supporting information, if applicable
3D Chess Media, LLC
Email Valerie@3DChessMedia.com
Telephone 949-309-0024
Related Pages
Portfolio & Ownership Disclosure