The Black Box of AI Search: Decoding the New Algorithm
As tools like Perplexity, ChatGPT, Claude, and Google AI Overviews rapidly become the default way people search for information, New Zealand business owners are asking a critical, urgent question: How do these AI models actually decide which websites to cite?
While traditional SEO relies heavily on keyword density, backlinks, and domain authority, AI search engines use a fundamentally different set of criteria to determine trustworthiness and relevance. Understanding this shift is the difference between thriving in the AI era and becoming digitally invisible.
How Perplexity and ChatGPT Actually Work
When a user asks a generative AI engine a question, the AI doesn't just rely on the static data it was originally trained on. Modern models perform live, real-time web searches. They read the content of the top-ranking pages, synthesise that disparate information, and generate a completely new, conversational answer, explicitly citing the sources they used.
To be chosen as a source by an AI model, your website must meet three critical, non-negotiable criteria:
1. Factual Consensus and Accuracy
AI models are designed to look for consensus. If your website provides clear, factual information that aligns with other highly trusted sources, the AI is far more likely to trust it and cite it. If your website makes wild, unsubstantiated claims or relies on hyperbole, the AI will flag it as unreliable.
You must stick to verifiable facts. If you state a price, state it clearly. If you claim to be the "best," you better have the aggregate review data to back it up, because the AI will check.
2. Machine-Readable Structured Data (Schema)
AI bots prefer data that is explicitly structured and easy to parse. Implementing comprehensive JSON-LD Schema Markup acts as a direct translation layer between your human-readable content and the AI's processing engine.
You must use LocalBusiness schema to define your core entity details, FAQPage schema to highlight your questions and answers, and Article schema for your blog posts. A website with flawless schema markup will almost always be cited over a website with slightly better content but zero structured data.
3. Unshakeable Entity Authority
AI models assess the overall reputation of your business "entity" across the entire internet. They don't just read your "About Us" page; they read your Google Business Profile, your reviews on NoCowboys, your mentions on local news sites, and your social media presence.
A strong, consistent entity with a high volume of positive sentiment is far more likely to be cited. If the AI finds conflicting information (e.g., your address on your website doesn't match your address on Yellow.co.nz), its confidence in your entity drops, and it will choose a competitor instead.
Formatting Content for the AI Reader
AI models are essentially highly advanced skimmers. They look for direct answers to specific questions. To optimise for this, you must adopt a strict Q&A formatting style.
Use H2 and H3 headings to pose the exact questions your customers ask, and immediately follow the heading with a concise, factual, 2-3 sentence answer. Once the direct answer is provided, you can expand on the topic in subsequent paragraphs for the human reader.
Adapting to these new ranking factors is the core methodology behind Generative Engine Optimisation (GEO). At Arise SEO, we ensure your business is structured perfectly for the machines that now control the flow of information.
The Impact of User Engagement Signals on AI Rankings
While structured data and entity authority are the foundational pillars of AI search rankings, generative models are increasingly looking at post-click user engagement signals to validate their recommendations. It is not enough to simply be cited by an AI; your website must actually satisfy the user once they arrive.
When a user clicks a link provided by an AI Overview or a Perplexity answer, the system monitors what happens next. If the user immediately hits the "back" button (a high bounce rate) or spends less than five seconds on your page (low dwell time), it sends a massive negative signal back to the AI model. It tells the algorithm: "I recommended this site, but the user found it unhelpful or difficult to use."
If this pattern repeats, the AI will permanently drop your website from its source list. To prevent this, your website must load instantly, be flawlessly optimised for mobile devices, and present the most important information "above the fold" (visible without scrolling). You must also ensure your content is visually broken up with clear headings, bullet points, and high-quality images to keep the user engaged and reading. In the AI era, user experience (UX) and SEO are completely inseparable.