Search is no longer a scroll through blue links. Large language models and multimodal systems now answer questions by pulling from text, pictures, clips and audio in one pass. Generative Engine Optimisation (GEO) builds on traditional SEO by helping AI choose your material as a trusted source. Getting there means treating every file-image, clip or soundbite-as structured data the model can quickly recognise and reuse.
Why the shift matters
Visual and voice queries are not fringe cases. Google says Lens handles around 20 billion image-based searches each month, and one in four carries commercial intent. Meanwhile, GWI reports that 32 percent of people use a voice assistant in any given week, and a fifth rely on it to look things up. The takeaway is clear: if your content cannot be parsed across modes, it is invisible to a growing slice of the audience.
Images: feed the embeddings, not just the crawler
Generative engines embed pictures into vector space, matching them to user prompts. Help that process by
- naming files descriptively (reef-safe-sunscreen-spf50.jpg) rather than IMG_1234.jpg
- writing alt text that answers the “what” and the “why” in fewer than 140 characters
- surrounding each picture with a short caption and relevant heading so the model sees topical alignment
- adding ImageObject schema, including licence, creator and subject
Quality still matters. Grainy stock is easy to skip; high-resolution originals signal authority and reduce hallucinated artefacts when the AI renders snippets.
Video: transcripts, chapters and context
A clip is only as discoverable as its words. Upload a clean transcript, fold in closed captions and, if the platform permits, chapter markers. That text feeds the language components of generative search and boosts inclusion in AI Overviews. For on-site embeds, wrap the player in a VideoObject schema block with duration, description and key moments. Google’s advice is blunt: surround videos with helpful prose rather than dropping a lone iframe.
Video bitrate and aspect ratio also influence whether a snippet becomes the thumbnail inside an AI answer. Keep key visuals within the centre-safe area to avoid automatic cropping by chat interfaces on mobile.
Voice: speakable markup and conversational phrasing
Voice answers tend to surface a single source, so precision counts. Start pages with a direct, 30-word summary that can be read aloud without edits. Use question-style H2 tags that mirror how people talk-“How long does a battery backup last?”-and follow with concise replies. Add FAQPage or HowTo schema, and mark key passages with speakable where appropriate. Google’s documentation confirms the markup helps assistants decide what to read.
Localisation also matters. An Aussie retailer might add price in AUD, store hours in AEST and colloquial terms-“thongs” over “flip-flops”-to match local speech patterns.
Technical housekeeping that models love
Generative engines still crawl. Maintain an XML sitemap that lists image and video variants, allow GPT-based crawlers in robots.txt, and supply a JSON-LD feed for products. Vector databases on your side, or services like Pinecone, can serve embeddings via API, speeding up retrieval when an AI pings your site. Core Web Vitals have indirect value too: a sluggish page risks being down-weighted because user signals feed the same ranking feedback loops that inform generative answers.
For public pages, ensure the HTTP Accept-Language header reflects English-AU so the system returns the right spelling to users Down Under.
Trust signals in a post-link world
E-E-A-T guidelines did not vanish; they moved upstream. Cite primary data, show author credentials and date-stamp updates. When an AI composes a narrative, it weighs both semantic fit and source reliability. Forbes notes that brands perceived as authoritative are quoted more often in generative responses. Adding author bios with verifiable professional histories and linking to peer-reviewed sources strengthens that perception.
Measuring success without a ten blue links report
Traffic may arrive as brand mentions inside an AI answer rather than a click. Track
- inclusion rate in AI Overviews (via tools such as Authoritas)
- share of voice for prompt sets across text, image and voice
- citation frequency in chat transcripts collected from customer service logs
- traditional metrics-impressions, watch time, call-through rate-for context
Complement quantitative data with qualitative reviews: ask sales or support teams whether prospects repeat phrasing that matches your structured snippets. That anecdotal loop often flags wins long before analytics dashboards catch up.
Australian considerations and compliance
Privacy rules in Australia echo Europe’s in many respects. If transcripts contain personal data, ensure consent and offer deletion paths to meet the Privacy Act. Accessibility standards such as WCAG 2.2 overlap with GEO requirements, so captioning and alt text serve both compliance and discoverability. Local nuance also extends to spelling (optimise, behaviour) and units (kilometres, litres).
Looking ahead
Generative search is edging towards real-time multimodal conversations-point a phone at a product, ask a follow-up and hear the answer. Brands that treat images, clips and voice fragments as structured, contextualised assets will surface first. GEO is less a bolt-on tactic than a content discipline: describe things clearly, supply machine-readable hints, and keep the human value front and centre. That approach should travel well, whether the user types, taps the camera or simply asks out loud.
