Australian AI Search Visibility Index Methodology

This is the pre-publication methodology for a proposed Australian AI Search Visibility Index.

No first-edition ranking, winner, market score, sample size or reliability result is claimed on this page.

The protocol is public before collection so the domain set, prompts, weights and exclusions cannot be changed quietly around a preferred result. When evidence is unavailable, the index will say unavailable. It will not convert an access gap into a zero.

What the index is designed to answer

For a frozen set of commercially relevant Australian search decisions:

• which brands appear in organic search;
• which brands are included in AI-assisted answers;
• which sources are shown or named;
• whether the answer describes each brand accurately;
• whether independent public sources corroborate the brand’s category and material claims.

Those are separate observations. A brand can rank organically without appearing in an answer, appear without a visible citation, be cited inaccurately or receive visits that do not convert.

The index will preserve each component beside any composite score.

Proposed components and weights

These weights are Searchmaxxed’s declared research design. They are not platform ranking factors.

Version 0.2 will either be used unchanged for the first collection or replaced by a new published pre-collection version. We will not tune weights after seeing scores.

Unit of analysis

The atomic record is:

edition
× platform and product surface
× frozen prompt or query
× declared market and account state
× run number
× retrieval timestamp
× observed answer or search result

The record stores:

• exact prompt or query;
• platform and visible product surface;
• date and time;
• country, language and explicit location where available;
• device;
• account or personalization state where observable;
• answer text or search result evidence;
• brand inclusion;
• displayed sources;
• linked and unlinked brand references;
• accuracy annotations;
• unavailable or failed states.

This prevents a screenshot from being detached from the conditions that produced it.

Market and sample definition

Every edition will declare before collection:

• industry and buyer group;
• Australia-wide or explicit city/state market;
• English-language scope;
• included domains and exclusion rules;
• query and prompt lists;
• platform and product surfaces;
• device and browser;
• account state;
• run count;
• collection window;
• organic data provider;
• any provider-modeled metrics;
• score denominators;
• conflict and missing-data rules.

The proposed first edition concerns Australian SEO and AI-search providers. A benchmark for another vertical needs its own domain, query, prompt and evidence set.

Scores from different editions are not comparable merely because both are shown on a 0–100 scale.

How domains enter the benchmark

The candidate set should be established before scoring using a documented combination of:

• businesses visible for the frozen commercial query set;
• recognized category competitors nominated before collection;
• relevant providers found in current market directories or independent coverage;
• Searchmaxxed, if included, disclosed as the research owner.

No domain will be added or removed because its preliminary score is inconvenient.

Duplicate brands, acquired brands, subdomains and regional variants will follow a declared entity-resolution rule.

Commercial organic visibility

Organic search uses a bounded query set grouped by customer decision:

• category and provider discovery;
• service and system searches;
• location-specific searches;
• pricing and evaluation;
• comparison and alternative;
• AI search, AEO and GEO language where relevant.

Before retrieval, each query receives:

• a demand weight;
• a commercial-intent weight;
• a market definition;
• a position curve;
• a missing-result rule.

Provider keyword volume and visibility estimates will be labeled as modeled. They are not first-party demand.

Google Search Console evidence can be used only for the property that authorises access. It will not be implied for competitors or used to make incompatible domains look directly comparable.

Answer inclusion

Prompts are grouped by customer job:

1. provider recommendation;
2. comparison;
3. alternative;
4. pricing or investment;
5. risk or objection;
6. category education;
7. local selection.

A brand receives inclusion credit when it appears in the observed answer, not because its name was placed in the prompt.

We record:

• exact wording;
• whether the brand appears;
• whether its owned domain is linked;
• other displayed sources;
• whether the inclusion repeats;
• whether the interface presents an ordered list.

Prose mention order will not be treated as ranking unless the product explicitly presents an ordered recommendation.

Every brand and competitor receives the same declared number of runs for the same prompt and conditions. A single appearance is an observation, not stable visibility.

Platform boundaries

The index will not pretend every AI search product works the same way.

Product features can change between editions. Every edition therefore records retrieval dates and uses versioned surface definitions.

Citation quality

Visible sources are classified before scoring:

• government, regulator or official register;
• academic or primary research;
• independent editorial;
• industry association;
• customer or partner;
• owned website;
• review platform;
• directory;
• community;
• unclear or low-quality network.

The score considers relevance to the answer, not only source class.

An owned page may be the strongest source for official price or capability. An independent source may be more appropriate for experience or comparison. Government and academic sources are not automatically relevant to every commercial claim.

Repeated links from one domain do not become authority breadth.

Entity accuracy

An answer can include a brand and still misrepresent it.

Where publicly verifiable, annotations cover:

• correct brand name and canonical domain;
• primary category;
• market served;
• intended customer;
• service or product scope;
• material differentiators;
• current public facts;
• unsupported or contradictory claims.

The brand’s own website is the source for facts it legitimately owns. Independent or official sources are used where the fact needs external verification.

Each item will be marked correct, incorrect, ambiguous, unsupported or unavailable. Reviewers will record the supporting source and date.

Authority breadth

Authority breadth counts unique relevant domains that corroborate the brand’s category or approved material claims.

The edition will disclose treatment of:

• sitewide links;
• duplicate domains;
• directory networks;
• syndication;
• irrelevant mentions;
• owned and related entities;
• links versus unlinked references.

This component is not raw backlink volume and does not claim to reproduce a search engine’s authority system.

Scoring and normalisation

Every component must define:

• numerator;
• denominator;
• missing-data rule;
• rounding;
• minimum sample requirement;
• reviewer-disagreement rule;
• exclusion condition.

The proposed composite is:

organic × 0.30
+ answer inclusion × 0.25
+ citation quality × 0.20
+ entity accuracy × 0.15
+ authority breadth × 0.10

A percentile or observed maximum will be labeled as relative to that edition. It will not be described as an absolute market truth.

If a stable denominator cannot be defined for a component, that component will be reported separately and excluded from the composite. The remaining weights will not be silently redistributed.

Repeat runs and uncertainty

AI-assisted answers can vary across runs and conditions.

For each prompt and platform, the edition will publish:

• declared repeat count;
• inclusion frequency;
• source frequency;
• agreement or disagreement across runs;
• collection window;
• failures and unavailable observations.

Where sample sizes are small, results will remain descriptive. The index will not imply statistical significance without a suitable design and calculation.

Manual accuracy and source classifications will use a written codebook. A second reviewer will examine a declared sample. Disagreement and resolution rules will be published with the edition.

Measurement beyond visibility

The index measures visibility and representation, not revenue.

Where Searchmaxxed has authorised first-party analytics, separate reporting may inspect:

• referral traffic identifiable by platform parameters or referrers;
• assisted journeys;
• qualified enquiries;
• CRM outcomes.

Google Analytics channel groups and attribution rules can change, and some AI referrals may be unidentifiable. Those gaps will be declared.

The benchmark will not estimate competitor conversions from public visibility.

What the index will not claim

• It does not measure every answer a customer could receive.
• It does not reveal private ranking or source-selection formulas.
• It does not guarantee that a mention or citation will recur.
• It does not treat a mention, citation, link, impression, visit and conversion as the same event.
• It does not prove causal revenue impact.
• It does not convert unavailable provider data into zero.
• It does not compare incompatible markets or versions.
• It does not claim the first edition already exists.

Publication threshold

The first edition will not be published until these artifacts exist:

• frozen domain, query and prompt sets;
• platform and collection specification;
• scoring workbook or reproducible code;
• source-class and entity-accuracy codebook;
• raw observation archive permitted for release;
• reviewer-disagreement record;
• exclusions and known gaps;
• versioned change log;
• reproducibility review.

If those conditions are not met, the benchmark remains proposed.

Frequently asked questions

Why combine organic search and AI answers?

They are different surfaces with connected web foundations. A combined framework lets a marketing leader see both while the component scores preserve the distinction.

Can a smaller brand lead the index?

The design permits it. Any lead must come from the frozen evidence and scoring rules, not from a size adjustment made after collection.

Will Searchmaxxed score itself?

If Searchmaxxed is included, the edition will disclose that Searchmaxxed owns the research and apply the same rules. Competitor first-party data will never be implied.

Can another analyst reproduce the result?

That is the standard. The public artifact set should expose enough inputs, rules and permitted observations to reproduce each score and see where evidence is incomplete.

When will the first edition be available?

Only after the publication threshold is met. Until then, this page describes a proposed methodology, not a market ranking.

Publish the rules before the winner

Explore the proposed Australian AI Search Visibility Index, read what to measure in AI search, or see how to track AI search visibility.

Searchmaxxed applies the same discipline to private market measurement through our AI search optimization system.

Show us the market you need to understand. We will freeze the decision set and evidence rules before collecting the first result.