业务逻辑与
输入变量说明

What JefurryAxis Does

When a buyer stops searching and simply asks an AI which one should I choose, the brands in that answer decide who gets looked at next. JefurryAxis measures exactly that: how consistently your brand appears in model answers, how high it ranks, how much of the answer is actually about it, and how competitors compare.

It is not a keyword rank tracker. A search tool tells you where a term places; JefurryAxis tells you whether you are in the answer when a real buyer brings a real situation to a model.

The product has two halves:

The console is the only context in the product scoped to a single brand, which is why the breadcrumb's brand switcher appears only there.

One line to remember: the workspace is where you watch trends, the console is where you run detections.

Quick Start in Five Steps

Five steps take a new organization from an empty workspace to a scored report.

1. Create a brand

Brands › New brand. A brand carries its domain, market, category and the competitors it is measured against. Everything downstream is scoped to it.

2. Open its console

The Console button on a brand card switches the whole frame into that brand's context.

3. Create a detection

Name the run, pick the answer engines, and set how many audience profiles to synthesise. Run settings holds the defaults, so the next run does not have to be described again.

4. Let the pipeline finish

Five stages run in order. Each can be advanced by hand, or left to run through with Auto continue pipeline.

5. Read the report

The report scores visibility, breaks it down by persona, competitor, engine and source, and lists what to change next. Put the run on a schedule to make the comparison repeatable.

The first run is the baseline. The value arrives on the second one: the same questions, the same personas, the same engines, asked again after the optimisation work, so the change can be measured rather than remembered.

The Three Engines

The platform is organised as three engines that hand work to each other.

They are not three separate products but one loop: detection defines the measurement space and finds the weak personas, scenarios, models and sources; optimisation turns that diagnosis into content and channel work; growth carries the work out and measures it. The report is not the end of the process — it is the layer that connects where a brand stands to what to do next.

One Detection Run, Stage by Stage

One detection run is five stages. A stage only starts once the one before it has produced something to work from, and every stage's output can be edited before the next one reads it.

A run can be stopped between stages and picked up later. Run history keeps each attempt with the settings it used, so two runs are only compared when they were asked the same thing.

Upstream decides downstream: brand information decides the tags, tags shape the personas, personas shape the scenarios, scenarios shape the queries, the queries decide the context the model answers in, and the answers decide the report. To change a result, go back up the chain — not to the report.

The Five Metrics at a Glance

Four measures are taken from the answers, and the fifth is their composite. Every chart, table and comparison in the product is switchable between them.

A mention is not a recommendation. Two brands can share a mention rate and sit a long way apart on rank and content ratio — which is the difference between being listed and being chosen.

Full definitions, how each is computed, and the edge cases are in chapters 23 to 29 below.

Answer Engines Covered

Each engine is queried separately, because they do not read the same internet. International models, Chinese-language models and more closed ecosystems draw on very different sources — what works on one may do nothing on another.

An engine can be switched off for a run; the report then says so rather than averaging it away. So before optimising for AEO, establish where the AI you care about actually gets its information — that decides which content to improve and which platforms to build visibility on.

Brands, Credits and Plans

Brands and competitors

A brand is the unit everything is scoped to: its own console, schedules, history and credit line. Brands can be pinned, paused and compared.

Competitors are the brands yours is scored against. They are tracked on the same queries, in the same period, with the same personas — the only way two scores mean the same thing. The console also surfaces potential competitors: brands the answers named that are not being tracked yet.

Credits

One credit covers one pipeline stage on one engine. A five-stage run across four engines therefore draws twenty credits.

Usage shows consumption across every brand in the organization, the charge ledger behind it, and the balance limit that stops a run before it spends more than is left. Plan credits and top-up credits are held as two pools and spent in that order.

Plans

Pricing depends on region, term and volume, so it is quoted rather than listed. What each tier includes:

Organization and roles

An organization holds its details, timezone, the topics its brands are read against, and its members. Three roles: Admin has everything, including members, billing and brand permissions; Member runs detections and edits the brands they have been given; Viewer reads reports and history without changing anything. Brand permissions are set per member, so a large organization can keep one team's brands out of another's console.

When a Report Looks Wrong

A stage is taking a long time

Every request in this build is delayed on purpose so loading states are visible. A stage that looks stuck is usually mid-skeleton; the running-tasks tray in the sidebar shows what is actually in flight.

Two runs disagree and nothing changed

Answers vary between models, between days, and between repeated asks of the same model. That is why a run is a sample rather than a lookup. Compare runs that used the same personas, queries, engines and period — run history records all four.

A competitor appears in answers but not in the report

It is not being tracked yet. The console lists brands the answers named under potential competitors; adding one there brings it into the next run's comparison.

A run stopped before it finished

The balance limit stops a run rather than letting it spend past the credits left. Usage shows the ledger entry; topping up or lowering the engine count lets it continue.

The brand is strong in the market but VS is low

This is usually not a scoring error but a real signal: the brand's public content, third-party citations, structured data or coverage of the target contexts have not yet turned into AI visibility. Chapter 30 is about reading exactly this case.

The workspace looks different after signing in again

Edits, theme choice and the demo session live in the browser's local storage. Clearing site data resets the prototype to its starting state.

Something not covered here? Write to success@jeffery.asia or contact sales.

M1. The Behaviour Equation: B = f(P, E_t)

JefurryAxis is built on Kurt Lewin’s behaviour equation (1951): what a person does is a function of who they are and the environment they are in. We apply it to AI search. The equation is Lewin’s; its use for question design, content strategy and visibility measurement is ours.

The practical consequence: the same person asks different questions in different weeks, and different people ask different questions about the same product. A keyword list approximates E and ignores P; JefurryAxis models both.

M2. Personas: From Brand Material to Decision Profiles

Personas are derived from the brand’s own material, not written from intuition.

1. Collect — website, brochures, reports, media and social content, plus any customer data the client provides with permission.
2. Decompose — material is broken into tags and values across four segmentation dimensions (geographic, demographic, psychographic, behavioural), with no preset conclusion.
3. Cluster — an LLM-led clustering step finds tag combinations that recur and hold together.
4. Confirm — the resulting lead personas are reviewed with the client before any question is written.

Each persona carries at least a motive, a constraint and a timeline. A persona that could describe anyone is rejected.

M3. The P × E Matrix and the Question Set

Each persona is crossed with the situations it buys in. One cell (P_i, E_j) is one decision context, and carries a fixed chain of brand-free questions:

• Brand-free. No question contains the client’s name. A question that names the brand measures the prompt, not the market.
• Size. Questions = personas × situations × questions per cell. A standard slice is 5 × 2 × 3 = 30.
• Query freeze. The set is frozen before the first run. Re-tests use exactly the same questions, so two runs are the same measurement taken twice.
• New situations are added, never swapped in. When the environment changes, the new situation becomes a new column beside the tracked set, so the original trend stays comparable.

M4. The Run Protocol

1. Dispatch each question separately to every model in scope, in parallel, under the same conditions.
2. Repeat each question for n independent rounds. A model rarely gives the same answer twice; the score is built on the distribution, not on one screenshot.
3. Two tracks. International models (e.g. ChatGPT, Claude, Gemini) and Chinese models (e.g. Doubao, DeepSeek, Qwen, Hunyuan) read different corpora and trust different sources. They are always scored and reported separately, alongside a dual-track view — never replaced by an average.
4. Closed platforms. RedNote Ask and Tencent Yuanbao build answers from in-platform content and offer no equivalent open API. They are observed through compliant interface testing and reported apart from API evidence.
5. Archive every answer with its question, persona, situation, model, round, timestamp and every cited URL. Each reported number must trace back to answer text.

M5. Metrics and the AI Visibility Score

For a set of N sampled answers:

The four point in different directions and use different units, so the score is built in three steps:

The logistic map makes the score most sensitive in the middle of the range, where “sometimes mentioned” turns into “usually recommended”, and flat at the extremes. The weight vector and the map parameters are fixed before a run and printed with every report; a scale that can be retuned after seeing the result is not a scale.

VS describes the recorded question set on the recorded models. It is not market share, not a probability of sale, and not a promise that another user sees the same answer.

M6. Reading the Results: Radar, Tracks and Priorities

Results are drawn as a radar with one corner per persona, one line per brand. Read it one line at a time: first the client alone, then one rival, then the full competitive set.

Verdicts are set per track. The same persona can be an Attack on the Chinese track and a Defend on the international one.

M7. Why Content Gets Cited: Three Gates

The framework in Persona-Driven AI Engine Optimization in Zero-Click Search (Dr. Yun Jia, Jeffery Asia; accepted for an oral presentation at ANZMAC 2026 following peer review) describes three stages content passes before an AI names a brand:

The SMU Master of Applied Finance case is the study in the paper: articles written for one persona were cited only under that persona’s questions.

M8. HCI: Human-Centric Insight

HCI grades a piece of content before it ships: does it speak to this person, in this situation, with a solution that is credible and easy for an AI to lift? It has four layers and twelve checks.

The composite is normalised to 0–100; the layer scores show where a piece is failing. HCI grades the input; VS measures the result. In tracked work they move together — a working relationship, not a guarantee.

M9. Measurement Rules and Limits

• Frozen configuration. Personas, situations, questions, models, competitors and scoring are frozen for a baseline. Any change goes through change control and is noted in the report.
• Like with like. A trend is only reported between runs on the same configuration.
• Variance is reported, not smoothed away. Answer wording varies between samples within an expected margin; a single run is a sample of that moment, not a fixed fact.
• Model updates move scores. A change of model version can shift results with nothing changed on either side; reports say so when it happens.
• Visibility is not revenue. Visits, enquiries and sales are measured in the client’s own systems, never claimed from a visibility score.

1. System Positioning

JefurryAxis analyzes how visible a brand is in large language model answers, where it appears in recommendations, how competitors pressure it, and what external content foundation supports the answers. These diagnostics can then inform content, channel, and brand cognition optimization.

Traditional brand analysis often focuses on website traffic, ads, search ranking, media volume, and sales conversion. JefurryAxis focuses on a different question: when the target audience asks a real decision question to an LLM, does the model include the brand, recommend it early, give it enough explanation, mention competitors instead, and rely on particular sources?

JefurryAxis therefore does not measure whether a brand is objectively excellent. It measures visibility in AI search and LLM recommendation environments. A brand can be strong in the traditional market but weak in LLM answers; this often means its public content, third-party references, structured materials, or target-market context have not yet converted into AI visibility.

In business terms, the system forms a loop of measurement, diagnosis, optimization, and continuous monitoring. Detection builds the measurement space and identifies weak audiences, scenarios, models, and sources. Later optimization and monitoring convert diagnosis into brand management signals. The report is not the business endpoint; it is the diagnostic layer between current brand cognition and future optimization.

2. Core Business Chain

Separate these core concepts first:

Business space is not a random keyword set and not the whole company. The core question is: which part of the brand is being measured this time? To judge business space, look at whether product or service scope, target customers, user decision chain, and competitor environment are consistent. If these factors clearly differ, there may be different business spaces.

The Brand Workspace is the system container. Brand Name tells the system which brand is being measured. Core Keywords tell it which part of the brand's business to measure. Business Description explains what that business actually is. Together, they define what the current Brand Workspace represents.

A Detection Run is a complete measurement inside that defined Brand Workspace. It uses the current Brand Name, Core Keywords, business knowledge, audience context, generated samples, Run Settings, and Tested LLMs to produce LLM answers and aggregate them into a report. A Detection Run should not be used to redefine business lines on the fly; the business space should be defined at the Brand Workspace level first.

The same public brand can have different Brand Workspaces for different business spaces. The same Brand Workspace can also produce multiple Detection Runs over time. Trend comparison is meaningful only when the business space and major measurement conditions remain consistent.

The JefurryAxis analysis chain can be summarized as:

Brand business information
-> audience tag space
-> audience personas
-> decision scenarios
-> natural language queries
-> multi-model answers
-> brand and competitor recognition
-> visibility, rank, content share, source, and competitive-pressure analysis

The key feature of this chain is that upstream inputs shape downstream outputs. Brand information shapes tag space; tags affect personas; personas affect scenarios; scenarios affect queries; queries define the context of LLM answers; answers ultimately determine report metrics.

The report is therefore not an isolated score. It is a measurement experiment formed by input variables, generated samples, and LLM answers.

The chain has two cooperating business lines:

Brand Name and Core Keywords jointly define the center of the measurement experiment. In plain language, Brand Name tells the system who is being measured, while Core Keywords tell it which part of the brand's product, business, or service is being measured. Other inputs add business knowledge, audience structure, regional context, competitive references, and sampling controls.

Core Keywords define the measurement scope, but the system does not reduce the business to keyword monitoring. It turns the scope into audiences, decision scenarios, and natural questions, then observes whether the target brand enters model answers. The real measurement unit is visibility of a brand inside one business space, for specific audiences and decision scenarios.

3. Object Types

Objects in the system fall into three categories.

Business inputs are causes, and report outputs are effects. Generated samples sit in the middle and convert business facts into a repeatable question space.

Generated samples are also reviewable and governable measurement assets rather than invisible black-box output. Their value is that they reveal how the system understands inputs as audiences, scenarios, and questions. When middle samples change, downstream answers and report conditions change. Governing generated samples is therefore governing the measurement question, not directly editing report scores.

4. Variable Importance Hierarchy

Variables are not equally important. The closer a variable is to the upstream foundation, the more it can redirect the whole report.

Errors in Brand Name and Core Keywords change who and what is being measured. This is directional bias. The measurement center is not Brand Name alone; it is jointly defined by Brand Name and Core Keywords, with Business Description explaining them. Bias in business knowledge or audience modeling changes how the system understands the target business and audience. Sampling controls mainly affect coverage, stability, and cost; they cannot correct a wrong foundation.

5. Brand Name

Brand Name answers: who is being measured this time? It is the entry point for identifying the target brand and the foundation for report attribution.

The most important property is stable reference. In reality, a brand may have a Chinese name, English name, abbreviation, legal company name, parent company name, sub-brand name, or product name. Users, media, websites, and models may use different names. Brand Name does not always have to be the most formal legal name; it should point to the brand entity that users and models actually recognize in the market.

Brand Name is uniquely important because it connects the entity the system understands with the entity the report counts. If name reference drifts, upstream business understanding and downstream metric attribution may both drift.

Brand recognition in report analysis is not simple exact string counting of the main name. It semantically recognizes the target brand entity and possible aliases. The system currently uses Brand Name as the main entity anchor and does not have a separate aliases business input field. Therefore, name ambiguity, too many aliases, complex parent-child brand relationships, or product names stronger than brand names can still cause missed recognition or wrong attribution. Brand Name stability affects business understanding and the credibility of MR, OF, and AR.

6. Business Description

Business Description answers: what exactly is this business? It is not a fixed template and does not require every background detail. Its role is to help the system understand the real business meaning behind Brand Name and Core Keywords.

A good Business Description helps the system understand what the brand does in the current business space, whom it mainly serves, what problem it solves, and how this business differs from other lines. It does not need to follow fixed dimensions or become a full company profile. If it lets the system clearly understand the current measured business, it can stay concise.

A description that is too short or slogan-like leaves the system without enough basis for judgment. A long but unfocused description can cause multiple business spaces to blend together. The value of Business Description is clarification and supplementation, not replacing Brand Name and Core Keywords as the measurement center.

When a brand has both enterprise or institutional customer businesses and individual user or consumer businesses, or multiple product lines for different customers, Business Description should make those differences clear. The goal is not to add complexity, but to prevent different businesses from being mixed into one generic space. If two businesses clearly differ in served audience, use case, decision criteria, delivery model, and competitor environment, they should usually be treated as different business spaces rather than mixed into one Detection.

7. Core Keywords

Core Keywords answer: which part of this brand's business is being measured? They are not ordinary SEO keywords and not a complete business introduction.

Put simply, Brand Name defines the subject, while Core Keywords define the scope. Brand Name tells the system who the target brand is. Core Keywords tell it which type of product, business, or service should drive Tags, Personas, Scenarios, and Queries.

Core Keywords should jointly describe one coherent business space:

Core Keywords are not supplementary notes. They are the scope definition running through the generation chain. If too broad, they expand the question space and pull downstream samples away from the brand's real strengths. If too narrow, the report covers only a small part of the business. If incoherent, they mix multiple business spaces in one measurement and make results hard to interpret.

Avoid two extremes:

• Do not ignore business-line differences. If one brand serves enterprise, institutional, or team users in one line and individual users or consumers in another, and these lines involve different product capabilities, use cases, decision criteria, delivery models, or competitor environments, consider separate business spaces or Detection Runs. The point is not to label the business, but to decide whether these are genuinely different core businesses for different audiences. If yes, they should usually be measured separately.
• Do not split too finely. If several keywords are only different phrasings of the same business, they do not need separate spaces.

Core Keywords usually work best as a small set of clear phrases that jointly define the business space. Use 3-4 related words or phrases as a reference. The goal is not to fill a quota; it is to tell the system what part of the brand is being measured.

For one public brand, do not split business spaces too much. In most cases, start with 2-3 clear business spaces so reports remain interpretable. Split further only when more business lines truly have different target customers, decision chains, and competitive environments. Failing to split when needed makes Detection too generic; splitting too finely makes each report lose enough business breadth.

8. Audience Geographic Regions

Audience Geographic Regions describe where the target audience actually is. It describes the user or buyer location, not the brand headquarters.

Business meaning:

• Affects persona regional identity.
• Affects query language and regional context.
• Affects model ecosystem selection.
• Affects price, regulation, availability, and cultural background.

The essential question is: who is asking, deciding, or buying? When brand location, user location, and user target region differ, Audience Geographic Regions still follows the user location, not the brand location.

9. Audience Preferred Regions

Audience Preferred Regions describe the regions the target audience cares about, compares, consumes in, applies to, invests in, or wants to enter.

Business meaning:

• Adds cross-region business context.
• Affects regional targets in scenarios.
• Affects whether queries naturally express cross-border comparison, application, procurement, or selection.
• Separates where users are from where users care about.

Geographic and Preferred can be understood as:

For local businesses, Preferred may match Geographic or may have no independent meaning. The key is not whether the fields repeat, but whether region materially affects comparison and model answers. If region affects supply, regulation, price, language, or service scope, it is important context. If it does not affect the question, it is only background.

10. Competitors

Competitors define the competitive context in the report. They are not used to make the target brand look better or worse. They let the system identify real comparison objects.

Business meaning:

• Supports competitor radar and relative visibility interpretation.
• Helps judge whether the target brand is suppressed by competitors.
• Provides relative visibility reference.
• Reveals which competitors models know or trust more.

The competitor set determines which entities share the same analysis and comparison space with the target brand. Weak competitors make the report look optimistic but less realistic. Overly broad competitors put the brand into an unfair or irrelevant competitive environment. Generic categories cannot support brand-level entity comparison.

Competitors mainly belong to report analysis and relative interpretation, not the core generation anchor for Tags, Personas, Scenarios, and Queries. They are the competitive reference frame in the same measurement experiment.

This separation keeps generated questions relatively neutral while report analysis still recognizes and compares defined competitors one by one. The competitor set affects who the brand is compared with; it does not require user questions to mention them.

11. Custom Persona Traits

Custom Persona Traits express key target-audience characteristics. They supplement the system's understanding of real users, making Personas closer to real decision makers rather than abstract templates.

Business meaning:

• Improves persona realism.
• Reduces generic audience generation.
• Protects strategic audience signals already known by the brand.
• Guides scenarios toward real decisions.
• Makes queries fit target audience information needs.

Persona traits can carry several types of audience information:

The value of traits comes from truthful and specific audience facts, not from the number of dimensions. Distorted audience information pushes downstream scenarios and questions in the wrong direction.

Tags and Custom Persona Traits both describe audiences, but their roles differ. Tags are system-generated general audience dimensions used to discover and combine current or potential groups. Custom Persona Traits are known audience facts provided by the brand to protect key signals. The former is exploratory; the latter is business prior knowledge. Together they shape Personas.

12. Additional Business Files

Additional Business Files supplement brand, product, market, and differentiation facts. They strengthen the system's business-context understanding.

Business meaning:

• Fill product details that a short description cannot cover.
• Provide features, pricing, cases, FAQ, positioning, and industry background.
• Strengthen upstream business knowledge and tag space.
• Distinguish real capabilities from marketing slogans.

These files mainly answer what the business is, where capability boundaries are, and what market facts matter. Their value depends on information density, relevance, and freshness. Low-relevance, repetitive, outdated, or overly promotional materials dilute business signals.

13. Sample Personas Files

Sample Personas Files add real user research and audience materials. They affect the realism of Personas and Scenarios.

Business meaning:

• Make personas closer to real customers.
• Provide decision motivation, comparison criteria, and barriers.
• Add information sources, search behavior, and regional preferences.
• Reduce template-like personas.

These files mainly answer who real customers are, why they decide, how they compare, and what blocks them. They differ from Additional Business Files: persona files add audience facts, while business files add business facts.

14. Language Environment

Language environment is not mechanical translation from one language to another. In different markets, the realism of brand descriptions, keywords, audience traits, and query language depends on whether they reflect how the target audience naturally expresses questions.

Business meaning:

• Affects answer quality in the target market.
• Affects whether user questions are realistic.
• Affects whether brands and competitors enter the correct context.
• Affects regional and language judgment in cross-border businesses.

Cross-border business does not automatically mean all English or one single language. Proper nouns, question structure, motivation expression, and comparison style together form the target audience's real language habits. Language environment should follow the target audience and market, not internal brand wording.

15. Tags and Tags Count

A Tag is a structured segmentation dimension for a brand's current or potential audiences. It converts dispersed customer understanding into a composable audience-feature space, mainly covering demographic, regional, psychological, and behavioral segmentation. A Tag is not a brand label or a product feature label.

Business meaning:

• Tags are the upstream audience semantic space for personas, scenarios, and queries.
• Tags determine which customer differences downstream samples expand around.
• Too few tags cause insufficient coverage.
• Too many tags add marginal, repetitive, or off-topic signals.

Tags Count controls tag-pool size. It is not the more the better. Since Tags are upstream, cutting them too early can reduce downstream diversity; but excessive low-quality tags can make personas and scenarios unfocused.

Tags Count should match business complexity and downstream capacity. If the business space is simple, too many tags add noise. If the business has multiple audience segments, decision roles, regional contexts, and comparison criteria, more tags may be needed.

Tags Count should be judged by decision-chain length, not industry name. Short decision chains with low pressure, few roles, and simple comparison criteria usually do not need many tags. Long decision chains with high risk, many roles, or complex comparison criteria need more tags to carry differences. Differences in served audience, users, decision makers, and evaluation criteria are not ordinary tag differences; they suggest different decision chains. If those differences point to different product lines or use cases, split business spaces first, then judge tags for each.

16. Personas and Personas Count

A Persona is a fictional audience archetype built from tag space. It represents current or potential customers in the brand's category or industry. It is not one real user and not a static marketing persona template; it connects audience understanding, decision scenarios, and real questions.

Business meaning:

• Splits the target market into user types.
• Tests brand visibility from multiple audience perspectives.
• Supports differentiated scenarios and queries.
• Helps locate which audiences are stronger or weaker for the brand.

Personas Count controls the number of simulated audiences. Fewer personas produce a more focused but thinner report. More personas expand coverage but raise cost and depend more on tag and trait quality.

Personas Count should also follow decision-chain complexity. If buyer, user, influencer, approver, or comparer differ, more personas may be needed. If the business decision is direct, forcing many personas makes profiles artificial.

Persona quality is not determined by quantity alone. The key is whether personas represent real buyers, decision makers, searchers, or influencers.

17. Persona Similarity

Persona Similarity controls how different personas are from one another. It is not a quality score; it is a business setting for audience similarity.

Low Similarity with few personas produces a small number of highly different audience views. This helps quickly reveal different user types but is less suitable for deeply describing one narrow group. High Similarity with high Regenerate Degree can create similar profiles, which may fit a concentrated market but reduces diversity.

18. Scenarios and Scenarios Count

A Scenario is the specific context and pain point that makes a persona ask or consult an AI. It connects who the user is with why the user asks now, and is usually shaped by both persona factors and external conditions.

Business meaning:

• Converts persona into a real decision context.
• Reflects comparison, evaluation, risk, region, budget, and use context.
• Helps locate weak decision backgrounds.
• Determines whether queries have real context.

Scenarios Count controls how many scenarios each persona has. More scenarios reveal weaknesses across different decision paths. Too few scenarios can make the report reflect only one problem background.

Scenario count should follow the decision process. Longer decision chains usually involve more information gathering, comparison, risk judgment, budget tradeoff, or internal persuasion. Shorter decision chains may make too many scenarios repetitive.

Scenario differs from Query: Scenario is decision background, while Query is the natural-language question asked in that background. Scenario expands where the question occurs; Query expands how the same situation can be asked.

19. Queries and Queries Count

A Query is the natural-language question actually submitted to Tested LLMs. It directly triggers measurement of AI visibility.

Business meaning:

• Determines answer context.
• Determines whether the brand has a reasonable chance to appear.
• Determines whether the report resembles real user search behavior.
• Determines whether results are repeatable and interpretable.

Queries Count controls how many questions exist under each scenario. Multiple queries under one scenario reduce randomness from a single wording but increase answer volume and cost.

In this business, Query mainly carries recommendation, ranking, comparison, and selection entry points within a category or industry. It does not directly ask about the target brand. Instead, it observes whether the brand naturally enters relatively neutral user questions.

Query quality is mostly determined upstream. Brand Name and Core Keywords define measurement object and business scope. Business Description adds business meaning. Regions define regional context. Personas define user identity. Scenarios define decision background. Query is the final measurement entry produced by the whole chain.

Induced queries change the nature of measurement. They make the model answer around the target brand and can raise some metrics, but this is closer to brand Q&A testing than real AI search visibility testing.

Neutral questions can also produce an important result: if a new, niche, or weakly recognized brand does not appear in relevant questions for a long time, that can indicate structural low visibility. This does not automatically mean measurement failure; it may mean the brand has not entered the natural recommendation set for that business space. Core Keywords define the legitimate visibility opportunity. Too broad a scope creates irrelevant zero visibility; focused scope with persistent zero visibility is closer to a real brand cognition gap.

20. Regenerate Degree

Regenerate Degree indicates how aggressively the system fills missing Personas after deduplication.

Business meaning:

• Controls the balance between quantity and quality when filling personas.
• Affects how strongly the system pursues enough coverage.
• Affects noise and forced-sample risk.

Regenerate Degree has tension with Persona Similarity. Similarity controls the direction of difference; Regenerate Degree controls fill intensity. When the required difference is high and fill intensity is high, the system is more likely to trade off quantity against quality.

21. Tested LLMs and Model Ecosystem

Tested LLMs define the model ecosystem covered by the report. Different models represent different language environments, training content, user markets, and answer styles.

Business meaning:

• Determines which AI search channels are covered.
• Affects brand and competitor appearance probability.
• Affects realism in Chinese, English, and cross-border contexts.
• Helps compare brand strength across model ecosystems.

Query language is strongly related to Tested LLMs. Query language is not an independent business anchor like Core Keywords; it is a measurement feature formed by Persona, region, and scenario context. Chinese audiences, Chinese questions, and Chinese-ecosystem models form one measurement context. English audiences, English questions, and global models form another. Language mismatch causes the report to measure the wrong market.

Cross-border business is complex because users may be in one region, care about another, and search in a third language. Language, Geographic, Preferred, and Tested LLMs together define the measurement context.

22. Run Settings and Answer Volume

Run Settings are the measurement sampling structure. They mainly include Tags, Personas, Scenarios, Queries counts, Persona Similarity, Regenerate Degree, and Tested LLMs. They do not define business facts; they define how much coverage, difference, and model-channel observation is applied to the same business space.

Approximate answer volume is:

Personas x Scenarios x Queries x Tested LLMs

Business meaning of answer volume:

• Determines report coverage.
• Determines cost and time.
• Determines how much single-run fluctuation affects results.
• Determines whether results fit light exploration, standard benchmark, or deep research.

Larger answer volume does not automatically mean a better report. It adds value only when upstream inputs are accurate, sample quality is reliable, and the task truly requires broader coverage.

Run Settings should be judged from business space and decision chain, not from the idea that larger parameters are always more complete. Simple businesses need focus. Complex businesses can increase coverage, but only when added samples correspond to real audiences, roles, scenarios, or question forms. If the brand has core business lines with clearly different served audiences, use cases, or decision criteria, first consider separate Detection Runs instead of forcing all coverage into one run.

Answer volume is the main proxy for cost and time, but it is not the full cost formula. Actual resource use also depends on model type, answer length, web retrieval, failed retries, and report-analysis calls. The same answer count can therefore have different time and resource costs.

23. Report Metrics Overview

JefurryAxis report metrics are not one single score system. They are complementary observation dimensions.

The atomic unit for metric calculation is one answer produced when one Query is measured by one Tested LLM. If the same Query is sent to multiple Tested LLMs, it creates multiple independent answers. Each answer is analyzed for brand, competitors, ranking, and content, then aggregated into model, Persona, Scenario, and overall report metrics.

These metrics are mainly for relative comparison: target brand versus competitors in the same report, the same Brand's historical trend under the same configuration, and differences across persona, scenario, and model. They should not be interpreted as fixed industry thresholds outside the task and configuration.

24. VS: Visibility Score

VS is a 0-100 composite visibility score reflecting overall brand performance across multiple dimensions.

Business meaning:

• Provides an overall report entry point.
• Aggregates mention, frequency, rank, and content-share signals.
• Supports trend observation and competitor comparison.
• Cannot explain causes by itself.

VS represents overall visibility quality, not a pursuit of one exposure count. It combines signals such as MR, OF, and AR, but the total score alone does not explain why. High VS usually means the brand performs well in relevant answers. Low VS can come from many causes: the brand is not mentioned, is mentioned but ranked lower, competitors are stronger, certain personas or scenarios are weak, or external content foundation is insufficient.

25. MR: Mention Rate

MR is the proportion of answers that mention the brand, from 0-100%.

MR = answers mentioning target brand / all answers

Business meaning:

• Measures whether the brand enters model recommendation or comparison scope.
• Reflects basic visibility in the question space.
• Is the key signal for whether the model thinks of the brand.

Low MR usually means the brand does not enter the answer candidate set. Common business causes include unstable brand naming, unclear business description, keywords drifting away from the strong category, mismatched query context, insufficient public content, or competitors being learned and cited more often in the relevant context.

26. OF: Occurrence Frequency

OF is the length-normalized average occurrence frequency of the target brand across all answers. It starts from 0 and has no fixed upper bound. It is not calculated only from answers that mention the brand.

Brand mention count in each answer is adjusted by answer length so longer answers do not automatically gain higher OF. The adjusted values are averaged across all answers. Answers without the brand enter OF as zero.

Business meaning:

• Measures repeated presence across the full answer space.
• Distinguishes being merely listed from being sufficiently explained.
• Works with MR to judge whether the brand is only occasionally listed or has stable presence.

If MR is high but OF is low, the brand may often be included in lists but not explained. If OF is high but MR is low, the brand may be discussed deeply in only a few answers while overall coverage remains limited. Abnormally high OF should be checked against query context for over-induction.

27. AR: Average Rank

AR is the mean standardized relative position of the brand in valid ranking answers. Its direction is special: lower values mean better relative ranking.

For each answer, the analysis model identifies brand rank and total list size from answer semantics. Only answers where the brand is actually ranked and both rank and list size are valid enter AR. Answers where the brand is absent, no clear ranking exists, or no valid list can be recognized do not enter the AR average. Valid ranks are converted into comparable relative positions, so lists of different lengths can be aggregated.

Business meaning:

• Measures whether the brand is recommended early.
• Separates being mentioned from being ranked highly.
• Reveals competitive position in comparison scenarios.

High MR but poor AR means the brand appears often but is not necessarily preferred. This usually means the brand has basic awareness but lacks enough comparative advantage in the model's view, or competitors have clearer evidence of strengths.

AR describes position after the brand is ranked; it does not describe the probability of being ranked. A brand can have low MR and good AR if it appears rarely but ranks well when it appears.

28. BCR: Brand Content Ratio

BCR for a single answer is the number of sentences containing the target brand name or alias divided by the total number of sentences in that answer. Sentence boundaries include not only periods, but also question marks, exclamation marks, paragraphs, and structured list boundaries. Multiple mentions in the same sentence still count as one brand-containing sentence.

Report-level BCR is the average of positive single-answer BCR values. Answers that do not mention the brand, with BCR equal to zero, do not enter this average. Therefore, BCR mainly describes how much content space the brand receives after it appears, not brand coverage across all answers.

Business meaning:

• Observes whether the brand receives enough explanation after being mentioned.
• Separates being named from being actually discussed.
• Works with MR, OF, and AR to judge visibility quality.
• Helps identify whether queries over-induce the brand.

BCR is not simply higher is better. Too low means insufficient brand content. Too high can mean the query over-induced the brand or the answer structure is unnatural. Its business meaning is whether brand content share is reasonable, not whether more brand content is always better.

28.1 Metric Combination Diagnostic Matrix

29. Source/Citation

Source/Citation describes observable information sources, citations, or content evidence in model answers.

Business meaning:

• Provides observable evidence of external sources linked to answers.
• Reveals competitor advantages in external content.
• Helps locate the target brand's public-content gaps.
• Supports direction for external content optimization.

Source/Citation focuses on external information foundations visible in answers. It can support source-structure analysis, but it is not the model's full knowledge source and cannot alone prove complete causality for brand mention or absence. Structured website content, FAQ, product comparison pages, cases, third-party coverage, rankings, reviews, and industry materials can all affect brand visibility in model answers.

Observable Source/Citation coverage depends on model channel, retrieval capability, search results, and whether the answer returns sources. Some answers may have no available citation data. No observed source does not mean the brand has no external content foundation. Source counts across models should not be directly compared without coverage context.

30. Strong Brand but VS Below Expectation

A strong brand with low VS is one of the most important JefurryAxis business scenarios. It does not automatically mean system error or real brand weakness. It requires separating two issue types.

The first type is measurement-configuration bias. Brand name, aliases, keywords, region, language, tested models, personas, scenarios, or queries may not match the real business, causing the report to underestimate the brand.

The second type is a real AI visibility gap. Traditional awareness, advertising, and offline influence may not have converted into LLM-visible content. Model answers may still favor competitors with richer public content, more third-party references, or stronger structured materials.

The value of this scenario is that it separates a strong brand from a brand visible in AI search. They are related but not the same.

31. Long-Term Trends and Measurement Uncertainty

Long-term trend value comes from repeatability. The same Brand is suitable for trend comparison only under the same input, same query space, and same Tested LLMs combination.

Business meaning:

• Observes whether AI visibility improves.
• Judges whether external content optimization has effect.
• Detects competitor improvement in certain scenarios.
• Identifies impact from model-ecosystem changes.

LLM answers are non-deterministic. Even if inputs and model names remain the same, wording, brand choice, ranking structure, and citation sources can change. Web search results, available sources, model service updates, and analysis-model judgment can also cause fluctuation. A single run is a sample of the AI visibility environment at that time, not an immutable fact.

The same model name does not guarantee the same underlying model. External model services may update weights, retrieval strategy, safety rules, or answer style, causing baseline drift. Trend changes may come from real brand changes or from model-ecosystem changes.

Configuration changes, query changes, model-combination changes, target-market changes, and model-service changes all alter experimental conditions. Trend analysis should keep the measurement baseline as consistent as possible and judge changes using multiple runs, change magnitude, several metrics, and Persona, Scenario, Model, and Source details. Small single-run movement is closer to noise; persistent cross-dimensional movement is more business-explainable.

32. Summary

JefurryAxis core business logic can be summarized in three sentences:

1. Brand Name and Core Keywords define who is measured and around which business space.
2. Audience and scenario chains convert business space into a testable user-question space.
3. Report metrics reflect brand visibility, competitive position, and content foundation in LLM answers.

Each variable's importance depends on its position in the chain and its business role. Brand Name and Core Keywords are foundation anchors. Business knowledge and audience variables determine how the system understands those anchors. Sampling parameters determine measurement coverage. Report metrics explain what happened. The system's core value is not manufacturing higher scores, but explaining why scores form and where the brand has real opportunities or gaps across audiences, scenarios, models, and sources.