LLM Authority Index

AI Market Discovery Research Standards

Overview

This page defines the universal quality, interpretation, versioning, correction, and publication standards applied across LLM Authority Index AI Market Discovery benchmarks.

Industry reports should focus primarily on unique category evidence and longitudinal findings. Universal standards belong here so they do not need to be repeated across every evergreen industry page.

For the research workflow, see AI Market Discovery Methodology. For metric formulas, see AI Market Discovery Metric Definitions.

Source-of-Record Principle

LLM Authority Index is the primary benchmark publisher for AI Market Discovery measurements.

An industry page should be the authoritative source for:

  • current and historical benchmark values;
  • tracked-brand standings;
  • collection and qualification context;
  • metric provenance;
  • public evidence summaries; and
  • measurement dates.

Downstream analysis on other properties may interpret the benchmark, but should link back to the relevant LLM Authority Index industry page as the source.

What the Benchmark Is Designed to Show

The benchmark can measure:

  • tracked-brand presence;
  • valid recommendation coverage;
  • top-three placement;
  • rank-one placement;
  • coded sentiment;
  • buyer-intent differences;
  • movement over time;
  • qualified surface breadth; and
  • citation/source patterns where the underlying AI/search surface exposes them.

These measurements are intended to show directional AI-mediated discovery behavior within a defined benchmark framework.

What the Benchmark Does Not Establish by Itself

The benchmark does not, by itself, establish:

  • total category market share;
  • revenue share;
  • attributable sales;
  • total AI prompt share;
  • every answer an end user could receive;
  • traditional search-engine market share;
  • causality from a metric increase or decline; or
  • guaranteed future recommendation behavior.

A benchmark movement identifies a change worth investigating. It does not automatically explain why the change occurred.

AI Outputs Are Dynamic

AI-generated answers can change because of:

  • model or system updates;
  • retrieval-system changes;
  • collection date and time;
  • geography;
  • language;
  • conversation state;
  • personalization;
  • logged-in status;
  • product interface;
  • source availability; and
  • stochastic model behavior.

A benchmark should therefore be read as a structured measurement of defined observations, not a claim that every user will receive the same response.

Repeated measurement is useful because it helps distinguish isolated observations from persistent patterns.

Raw Collection vs. Qualified Benchmark Size

The research process intentionally begins with a larger source collection and narrows it through relevance and buyer-intent qualification.

A final set of 87 qualified observations does not mean only 87 AI interactions were tested. It means 87 observations from the broader collection satisfied the benchmark's rules for the public category analysis.

Industry reports should state this distinction near the top of the page.

Small-Count Interpretation

For lower-visibility brands, one or two observations can materially change a percentage.

Public reports should therefore:

  • pair rates with counts where practical;
  • avoid overstating movements driven by a very small numerator;
  • explain when the underlying count is unusually small; and
  • prefer descriptive language over unsupported claims of statistical significance.

Example: A change from seven valid recommendations to one is strategically relevant, but the absolute counts are essential context.

Movement Labels and Statistical Language

Terms such as statistically significant, significant increase, or beyond normal variation should only be used when the benchmark has a defined, documented, and versioned rule that supports the label.

Until such a rule is publicly documented, prefer descriptive language such as:

  • largest increase;
  • largest decline;
  • notable movement;
  • new category high;
  • new category low; or
  • leadership change.

This avoids implying a formal statistical test when the public methodology does not define one.

Evergreen Industry Report Policy

Each industry/category benchmark should use one evergreen URL.

The same URL is updated as new measurements are added.

The page should preserve meaningful historical context such as:

  • the original baseline;
  • current measurement;
  • category leadership changes;
  • historical highs and lows;
  • major placement shifts;
  • persistent gains or declines;
  • reversals;
  • new entrants or exits from the tracked set; and
  • methodology annotations that materially affect comparison.

This creates a cumulative longitudinal research asset rather than a series of outdated monthly pages.

Longitudinal Integrity

Month-over-month comparison is only useful when the reader can distinguish benchmark movement from methodology movement.

Maintain an internal change log for:

  • AI/search surfaces added or removed;
  • query-universe changes;
  • geography or language changes;
  • collection-setting changes;
  • brand/entity-set changes;
  • buyer-intent changes;
  • recommendation-validity changes;
  • sentiment-coding changes;
  • modeled-value changes; and
  • pipeline or QA changes that could affect published values.

If a material change affects comparability:

  1. annotate the relevant report;
  2. update the methodology version;
  3. recalculate historical measurements where feasible; or
  4. clearly identify the break in comparability.

Qualified Surface Breadth

Reports should distinguish:

  • surfaces tested — part of the benchmark design; and
  • qualified surface breadth — tested surfaces contributing at least one qualified observation after filtering.

A change in qualified surface breadth should not automatically be described as a change in platform coverage methodology.

Entity and Category Naming

Category names and tracked-brand names should be normalized consistently across:

  • page titles;
  • headings;
  • report cards;
  • tables;
  • structured data;
  • internal links;
  • exported data; and
  • downstream strategic analysis.

A tracked brand should never be presented as the category itself unless the research category genuinely uses that name.

This is important for both human interpretation and machine/entity understanding.

Source and Citation Interpretation

Where AI systems expose citations or attributable sources, those sources can provide useful context around a recommendation.

However:

  • citation presence does not prove causality;
  • a source can support more than one brand;
  • different AI/search surfaces can retrieve different evidence for the same source query;
  • source frequency should be read alongside buyer intent and recommendation outcome; and
  • unavailable citations should not be treated as evidence that no external retrieval occurred.

Use language such as:

  • "was cited in";
  • "appeared around";
  • "was associated with";
  • "coincided with"; or
  • "was present in the evidence set"

unless stronger causal evidence exists.

Modeled Value Interpretation

Modeled AI Authority Value is a comparative benchmark estimate rather than measured commercial performance.

Reports should:

  • always label the metric as modeled;
  • distinguish modeled value from revenue;
  • disclose material input changes that affect comparability; and
  • link to Modeled AI Authority Value.

Percentage Movement

Changes between percentage rates should be reported in percentage points.

Public convention:

  • prose: "increased by 17.5 percentage points"
  • table/chart: "Up 17.5 points"
  • decline: "Down 8.0 points"

Avoid ambiguous shorthand when a plain-language label is clearer.

Report-Specific Limitations

Industry pages should not paste this complete standards page into every report.

Instead, they should include only limitations that materially affect the current category, such as:

  • very small counts for a particular brand;
  • missing source attribution;
  • a first measurement period;
  • a tracked-brand-set change;
  • a methodology version change; or
  • another local constraint that changes how the results should be read.

Publication Quality Standards

Data Integrity

Before publication:

  • use stored numerators and denominators;
  • reconcile all percentages with source counts;
  • remove duplicates and malformed observations;
  • normalize brand/entity mappings;
  • preserve historical values;
  • confirm model/value inputs;
  • confirm collection dates; and
  • remove test or illustrative data.

Editorial Integrity

  • The H1 must describe the actual category.
  • The executive summary must reflect the current data.
  • Observation must be distinguished from interpretation.
  • Causal claims require evidence.
  • Small-count movements must be contextualized.
  • Percentage-point changes must be labeled correctly.
  • Unsupported statistical language should be avoided.
  • Generic methodology should be linked rather than duplicated in full.

Human Readability

  • Put the most important current findings near the top.
  • Use explicit table labels and units.
  • Avoid unnecessary acronyms.
  • Provide textual equivalents for charts.
  • Make the benchmark denominator understandable without requiring a methodology deep dive.

Machine Readability

  • Keep important facts in crawlable HTML text.
  • Use descriptive headings.
  • Render tables with proper header cells.
  • Use descriptive internal links.
  • Keep structured data consistent with visible content.
  • Use accurate published/modified dates.
  • Keep each report self-canonical.
  • Include important evergreen URLs in XML sitemaps.

Corrections

Material corrections should update the page's modified date and be retained in an internal audit trail.

Examples include:

  • incorrect denominator;
  • coding error;
  • duplicate observation;
  • wrong brand mapping;
  • incorrect historical value;
  • source-attribution error;
  • material methodology-description error; or
  • modeled-value calculation error.

Minor stylistic edits that do not change the research result do not require the same correction treatment.

Relationship to Downstream Analysis

LLM Authority Index industry pages should remain the neutral benchmark source.

When CiteWorks Studio or another downstream property interprets the benchmark:

  • link to the source industry page;
  • avoid copying the benchmark article wholesale;
  • add independent strategic analysis or company-specific value;
  • preserve the meaning of benchmark metrics; and
  • distinguish source facts from downstream interpretation.

Related Resources