Between March and June 2026, we tracked citation growth for "generative engine optimisation" across the DataForSEO content analysis network, which indexes 5.4 billion pages monthly. We paired that with technical audits of Australian SME websites across six verticals: financial services, legal, allied health, accounting, real estate and professional trades. The data doesn't match what most published GEO advice suggests.
The citation numbers
Citations of "generative engine optimisation" in indexed content grew from 1,976 per month in March to 2,576 per month in June. That's 30% in 90 days. Australian-authored content grew faster than Australia's share of total indexed English-language content would predict.
The growth concentrated in three content types: technical explainers with specific implementation code (schema markup, robots.txt directives, structured data examples); case study content with concrete before/after timeframes; and industry-specific pages where AI models could confirm that structured data was present and extractable.
General "GEO is important" articles generated negligible citation activity, despite representing the majority of what's been published on the topic. Publishing about GEO doesn't get you cited for GEO. Pages that AI models can extract and verify do.
What LLM monitoring shows for new Australian domains
We ran LLM mention monitoring across ChatGPT, Claude, Perplexity and Gemini for unbias.au (our own domain, launched in late 2025). Six months in: zero citations across all four models. That's the baseline most Australian SMEs start from, and it's a useful number to have on record.
AI models don't cite a business because the domain is established or the site looks credible. They cite businesses that appear in third-party content the models trained on, or on pages with enough entity signals to independently verify the citation.
Getting from zero citations to any requires three things in order. The site must be crawlable, schema-valid and machine-readable. The business must appear consistently across Google Business Profile, industry directories and its own structured data. The business must appear in content that AI training pipelines consume: industry publications, directory listings, press coverage. Most Australian SMEs are failing to clear the first condition.
The JavaScript rendering problem
GPTBot, ClaudeBot and PerplexityBot don't execute JavaScript. They fetch the raw HTML from a server and extract structured data and text from what they receive.
React, Next.js, Vue, Nuxt, Webflow and Squarespace all render content client-side by default. Without server-side rendering or static generation, the raw HTML response is mostly empty. No contact details. No service lists. No pricing. A client-rendered page arrives at a JavaScript-disabled crawler as a loading spinner and nothing else.
In our audit sample across Australian professional services websites, the same failures came up repeatedly. Most sites built on page builders render contact details, service lists and pricing client-side. Even Next.js sites frequently disable server-side rendering for specific sections, producing partial extraction failures where some content is readable and some is not. The single most common failure: phone numbers and email addresses coded as JavaScript event listeners, not as plain text or tel: links.
None of these show up in a visual audit. They only appear when you fetch the raw server response and compare it against what a browser renders with JavaScript enabled.
Schema markup: what is actually missing
Schema markup tells AI models what type of entity a page represents and provides machine-readable fields for name, address, pricing, hours and expertise. The most common failure isn't absent schema. It's invalid schema that fails validation. A schema block that Google's Rich Results Test rejects is treated as no schema at all by AI models. The site appears to have structured data. The structured data provides nothing extractable.
The AI-specific properties missing from most Australian SME sites: knowsAbout (specific expertise domains of the business); speakable (which page elements are suitable for AI extraction and voice response); hasOfferCatalog (machine-readable service lists with pricing); areaServed (explicit geographic scope); and founder or member (named individuals linked to the organisation entity).
These properties exist specifically for AI assistants and voice search. Standard SEO work doesn't address them. An agency optimising for Google's traditional ranking signals isn't, by default, optimising for AI citation.
llms.txt: the signal fewer than 3% of Australian sites use
In late 2024, a community standard emerged for a file called llms.txt. Placed at the root of a website, it gives AI crawlers a plain-text summary of what the site is, who operates it, which pages matter and whether the content can be cited.
Unlike robots.txt, which controls crawl access, llms.txt provides semantic context. As of June 2026, fewer than 3% of indexed Australian business websites have a valid one. For AI models that consume this file during training data collection, the absence means the model must infer what the business does from unstructured page content. That inference is less reliable than a curated summary.
No AI provider has confirmed it produces a measurable citation uplift. It's a zero-cost disambiguation signal. For a domain with no backlinks and no third-party citations yet, every signal that helps an AI model correctly identify what your business does has value.
robots.txt and the Agentic Browsing Score
AI crawlers respect robots.txt directives. A User-agent: * with Allow: / technically permits all crawlers but doesn't constitute explicit permission for AI-specific user agents, and Lighthouse v13.3 treats them differently.
The Lighthouse Agentic Browsing Score checks for explicit entries for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Sites using only the wildcard directive score below 70/100 on this metric regardless of what the wildcard technically permits. Adding explicit entries takes ten minutes and signals to AI providers that the site operator has made a deliberate decision to permit indexing, which may influence crawl queue priority as AI providers scale their indexing operations.
The apex redirect
Lighthouse flags apex-to-www redirects as a performance penalty. For most sites, this fires at 200-300ms, an application-layer cost that adds to schema processing and render checks for every crawl request. Moving the redirect from Next.js to edge middleware drops it to under 20ms per request from the apex domain. Infrastructure efficiency isn't a traditional ranking signal. For AI providers making crawl budget decisions across millions of domains, it compounds.
What this means for the second half of 2026
The citation gap between technically correct and technically broken sites is widening. AI citations are accumulating on sites that completed the infrastructure work earlier, at a faster rate than new sites are completing it now. The gap isn't static. It grows each month it goes unaddressed.
AI models cite businesses with verifiable, machine-readable entity data. A site with valid schema, consistent citations and a structured service list is more credible to an AI model than one with better visual design and no structured data. Australian businesses have invested heavily in design. The structured data layer underneath it has been largely ignored, and that's where the citation decisions get made.
Content strategy, link building and retainer engagements all produce less return on a technically broken foundation. Knowing specifically what the failures are is the prerequisite for every piece of work that follows.
Running your own check
Fetch your homepage with a JavaScript-disabled browser or curl and read the raw HTML response. If your contact details, service list and pricing aren't in that response, AI crawlers can't read them. That's the diagnostic that matters before anything else.
Unbias runs a Foundation Audit for AUD $1,250 covering rendering behaviour, schema validity, citation consistency and AI crawler access. The prioritised fix list is delivered before any implementation work begins.
Written by
Brandon
CEO, Unbias — SEO, AEO & Revenue Operations specialist helping Australian professional services firms get found online and convert traffic into revenue.