Cross-platform consistency
LinkedIn, Crunchbase, Wikipedia and on-site Schema use a consistent brand description.
Technical foundations
The technical foundations behind AI Search visibility
Being cited by ChatGPT, Perplexity and Google AI Overviews is less about publishing more pages and more about whether AI systems can access, parse and trust your pages. Most websites have problems in all three areas.
The problem
Why most SEO work does not convert directly into AI visibility
Traditional search optimisation and AI Search optimisation share the same technical foundations—then diverge at a critical point. AI systems are not ranking pages. They extract passages, verify entities and resolve citations in real time from a filtered set of trusted sources.
Even with strong Google rankings, a brand may still fail to enter AI-generated answers if any of the following is true:
- robots.txt is misconfigured and blocks AI retrieval crawlers.
- Content depends on JavaScript that AI crawlers cannot render.
- There are no machine-readable entity signals connecting the brand to verified knowledge-graph representations.
- Page structure buries answers deep in the page instead of stating them clearly at passage level.
- Schema markup is missing, outdated or implemented incorrectly.
47 is a service check scope—not a performance statistic. Actual priorities are confirmed against site architecture, content risk and available access.
For Hong Kong brands, these technical gaps often sit beside bilingual entity problems. A crawlable English page with weak Traditional Chinese counterparts—or Schema that names only one language variant—can leave the model able to fetch HTML while still unsure which company is being discussed. Technical access is necessary; it is not sufficient on its own.
Request a free audit →Layer 1: Access
Can AI systems reach your content?
Before any optimisation strategy can work, AI Search bots must be able to reach the page without unnecessary friction. By 2026, more than a dozen AI crawlers request content from every public site—and the differences between them matter.
Some robots ingest content for model training; others retrieve content live to cite while answering users. Treating them identically in robots.txt is a common and costly configuration mistake.
| Robot / User-Agent | Type | Audit treatment |
|---|---|---|
OAI-SearchBot | Live citation | Included in audit |
PerplexityBot | Live citation | Included in audit |
Claude-SearchBot | Live citation | Included in audit |
Google-Extended | AI Overviews + model training | Included in audit |
GPTBot | Model training | Included in audit |
Bytespider | High-frequency crawler | Included in audit |
This table lists six named robots inside AI Search Lab’s audit framework. A full audit also reviews other relevant robots based on site and platform conditions. User-Agents, purposes and platform policies can change; check each vendor’s latest official documentation before changing settings.
Key risk: Your current robots.txt may block robots that would cite you, while allowing robots that only harvest content without returning citations.
Audit your robots.txt →Layer 2: Trust
Does AI clearly understand what your brand represents?
AI systems do not trust a single website. They trust entities: brands, people, products or concepts with verified, consistent, unambiguous representations across multiple authority sources.
Without a clear entity definition, AI systems may skip your content—or attribute it to a competitor with stronger entity signals. There is no error message, ranking drop or alert when this happens.
sameAs declarations
Structured data connects your domain to verified external knowledge-graph entries.
Author entity profiles
Connect individual expertise signals to published content to build topical authority.
Topic cluster architecture
Concentrate topical authority through a coherent content map—not scattered one-off pages.
AI Search Lab’s entity audit maps a brand’s current knowledge-graph footprint across 11 external source types, finds entity gaps that cause attribution loss, and identifies competitor entities that replace you in AI answers.
“11 external sources” is audit coverage—not a performance statistic. Specific sources adapt by brand and industry; detailed method is provided in the engagement brief.
Layer 3: Extraction
Can AI extract a citeable passage from the page?
AI citation happens at passage level, not page level. Perplexity or ChatGPT Search does not simply cite a domain—it extracts specific sentences or paragraphs and attributes them to a URL.
Successful extraction depends entirely on content structure. Most content is written for humans scanning top to bottom. AI looks for answer units that can stand alone, then verifies surrounding context. Even when the information fully answers the user, content not built this way may never be extracted.
Every content audit checks
Does each section lead with a stand-alone answer?
AI often extracts the first coherent statement in a block. Front-loading context and delaying the answer can block extraction.
Are headings phrased as questions that match user query patterns?
Query-aligned headings create passage anchors. Vague headings such as “Overview” give weak citation signals.
Is comparison content presented in tables rather than pure paragraphs?
Where comparison fits, tables make fields, conditions and differences easier for people and machines. Citation still depends on quality, sources and the query.
Do how-to sections use numbered steps that start with action verbs?
Procedures in numbered-list form extract cleanly into AI answers; pure paragraphs are harder.
Are statistics accompanied by named sources and publication years?
Unsourced statistics look unverifiable. Named, dated sources increase citation trust.
Extraction-ready architecture can improve citation odds by adjusting content structure alone—without changing the underlying information.
Limits: Content format is a supporting signal. Tables alone do not produce rankings or citations. They only help when the material itself is complete, accurate and relevant to the query.
See what extraction-ready content looks like →Layer 4: Evidence
What the evidence points to
The following summarises technical and content conditions we check repeatedly in audits and open research. They are diagnostic directions—not a formula that guarantees citation.
| Signal | Observed influence on AI citation |
|---|---|
| Correctly allow required retrieval robots | Avoids technical settings that block platforms from accessing public content |
| Organization Schema matches visible brand facts | Reduces machine misunderstanding of names, services and official relationships |
| Organise content as answer units | Lets each section independently answer one clear question |
| Build interlinked topic clusters | Helps people and crawlers understand relationships between pillars, subtopics and commercial pages |
| Mark update dates and maintain outdated material | Helps readers judge whether content still applies—without meaningless date refreshes |
| Original data with named methods | Provides verifiable, repeatable evidence with real information gain |
Interpretation limits: Platforms, timing, language, industry and queries produce different results. These items support systematic checking. They should not be read as a single ranking factor, causal proof or performance guarantee.
The audit
47-point AI visibility audit
Every AI Search Lab engagement begins with a five-layer structured technical review that identifies which pages are already citation-ready, which have fixable issues, and which need structural rework.
Layer 1 · 7 checks
Crawl access
Confirm AI citation robots are correctly allowed, sitemaps are clean, and indexable content is not blocked by robots settings or render failure.
- robots.txt robot-permission review
- Sitemap completeness and freshness
- JavaScript render-dependency check
- 4 additional checks in the full audit
Layer 2 · 11 checks
Structured data
Review Schema presence and accuracy across page types, plus entity-declaration quality—including Organization, Article, FAQ and BreadcrumbList.
- Organization Schema and
sameAsmapping - Article / BlogPosting author entities
- FAQ and HowTo Schema implementation
- 8 additional checks in the full audit
Layer 3 · 12 checks
Content extractability
Review heading structure, answer units, table use, FAQ format and passage-level coherence on highest-priority pages.
- Per-page answer-unit detection
- Heading-to-query alignment scoring
- Table-to-paragraph ratio for comparison content
- 9 additional checks in the full audit
Layer 4 · 9 checks
Entity and knowledge graph
Map brand entity consistency across external platforms, validate sameAs declarations, and review topical-authority signals in author entity profiles.
- 11-source knowledge-graph footprint
- Brand-description consistency review
- Competitor entity substitution analysis
- 6 additional checks in the full audit
Layer 5 · 8 checks
Performance and rendering
Inspect Core Web Vitals, server-side rendering status, render-blocking resources and image optimisation—because slow or broken pages cannot be crawled reliably.
- Core Web Vitals baseline
- SSR vs CSR dependency mapping
- Render-blocking resource review
- 5 additional checks in the full audit
Five layers total 7 + 11 + 12 + 9 + 8 = 47 checks. This page lists representative checks per layer; actual scope is confirmed against site architecture and the engagement brief.
Request a free AI visibility audit →Your site may have problems—and many of them can be fixed
The first audit ranks issues that may block AI citation by impact, dependencies and implementation cost. Some items can be fixed on the current site; others may involve content, topic architecture or external sources. Whether a rebuild is needed should follow evidence.
47Technical checks per audit
PriorityOrdered by impact, dependencies and implementation cost
RoadmapSeparates immediate fixes from mid-term build work
Actual timeline and outcomes depend on tech stack, issue severity, access, content approval and implementation capacity—confirmed separately in engagement scope.
Hong Kong GEO full guide · AIO five-step method · AI Search guide
FAQ about How AI Search Optimization Works: Signals & Process
Is Schema mandatory?
Schema is not the only lever, but it helps machines parse entities and page types when accurate and aligned with visible content.
Should content or technical work come first?
Start with a baseline and find blockers. If pages cannot be crawled or rendered, fix technical access first. If access is fine but answers are weak, prioritize content, entities, and evidence.
What is an extractable answer?
A short passage that can be understood without heavy surrounding context—usually a direct answer first, then conditions, evidence, limits, or next steps.
What role do external sources play?
Trusted third parties can corroborate brands, people, products, and claims. They must be real and relevant—not fake reviews or low-quality stands-ins for weak official pages.
How do we know optimization improved?
Retest with the same queries, platforms, regions, and logging fields. Compare mentions, citations, accuracy, page coverage, and referral—not one favorable screenshot.
Want a baseline of how AI answers describe your brand?
Request a free Audit