AIready has published findings from 158 website audits showing many New Zealand sites are difficult for AI crawlers to read. The median score was 55 out of 100.
The results, based on scans carried out between August and October, found that 21% of audited sites scored below 40. These lower-scoring sites were often affected by client-side JavaScript rendering, bot-management settings and missing structured JSON-LD entity data.
The Auckland-based company assessed websites across three areas: crawler access and permissions, machine readability, and structured entity clarity. Checks covered issues such as robots.txt settings, bot blocking, semantic HTML structure, server-side rendering versus heavy client-side scripting, and the use of schema markup to describe a business and its services.
Of the 158 sites reviewed, 28% scored 70 or above, indicating clear crawler permissions, structured semantic content and machine-readable data. A further 51% scored between 40 and 69, suggesting partial readability with technical friction on some checks.
That left 33 sites, or 21% of the total, in the lowest-scoring band. These sites tended to present more serious obstacles, including security settings that blocked automated scanners, scripts that prevented key information from loading on first crawl, and content structures that made core offerings harder to parse.
Common issues
The benchmark identified three recurring patterns across the audits. The first was crawler access restrictions, including web application firewalls, bot-management rules and misconfigured directives that can block AI search crawlers.
The second was client-side rendering gaps. In these cases, important business information was embedded behind JavaScript execution, while lighter AI parsers often prioritised plain server-rendered HTML.
The third was missing structured context. Some sites relied on broad marketing language without Schema.org JSON-LD markup to define services, operating locations and leadership.
Denvor Snowden, Co-founder of AIready, said many organisations overestimate how easily AI systems can interpret their websites.
"Business leaders frequently assume that having an aesthetically modern website means AI assistants like ChatGPT, Claude, and Perplexity can easily read and cite their business," Snowden said.
He added: "In practice, automated scanners frequently encounter basic hurdles - from heavy JavaScript frameworks that don't render on first crawl, to blocked user agents, to a lack of explicit schema markup describing what the business actually sells and where."
Measurement limits
The scoring measured accessibility for automated scanners rather than a business's wider reputation or authority. Successful crawler access also does not mean an AI model will necessarily cite, recommend or surface a company in responses.
Even so, the findings point to a growing technical issue as conversational search and large language model citation systems play a larger role in digital discovery. If crawlers cannot access or interpret website content clearly, businesses may face reduced visibility in AI-generated answers regardless of how polished the site appears to human users.
The dataset also suggests the problem is not confined to outdated websites. Even modern sites can be opaque to fast, stateless AI crawlers when content depends too heavily on front-end scripts or when security controls are applied without allowing for legitimate machine access.
That matters because many websites are built primarily for human browsing, search engine optimisation and visual presentation, rather than machine interpretation by newer AI systems. A central theme in the audit results is the gap between what looks complete to a visitor and what is legible to an automated parser.
For organisations reviewing their digital estates, the benchmark points to a relatively narrow set of technical factors that can materially affect AI readability: whether content is visible in initial HTML, whether crawler permissions are correctly configured, and whether business details are clearly defined in structured data.