AI search and SEO. How to stay visible when the answer is an AI, not a list of links.

What AI search actually is

AI search is the shorthand for any search experience where the front-end is a generated answer rather than a list of links. Four products dominate the picture in 2026. Google's AI Overviews sit at the top of the regular Google SERP. LLM-driven search inside ChatGPT pulls live web results and synthesises them. Perplexity built its whole product around cited answers. Google Gemini does the same inside the Google ecosystem. Each one reads pages from the open web, summarises them, and shows the user the answer plus a handful of citation links.

From an SEO perspective, that is both familiar and new. Familiar because the systems are crawling and indexing the same web pages we already optimise. New because the unit of visibility is no longer "rank position three on a page of ten blue links". It is "cited inside the synthesised answer", which is a very different rendering and a very different click pattern.

Two terms get thrown around for the work of optimising for this. AEO (Answer Engine Optimisation) is the older one, born around 2017 to cover featured snippets and voice. GEO (Generative Engine Optimisation) is the newer one, born around 2023 to cover generative AI specifically. The two overlap so heavily in practice that arguing about the definitions wastes time. We use both terms because clients use both terms; the underlying work is the same. The first cluster in this pillar, AEO vs GEO vs SEO, draws the lines properly for the people who want them drawn.

Worth saying out loud at the start: there is no public algorithm for how ChatGPT, Perplexity or Gemini choose their citations. Vendors publish high-level documentation; nobody publishes a ranking function. Everything written in this pillar about citation behaviour is observation across client sites and the public research we trust, not algorithmic certainty. Treat anyone claiming a guaranteed method with appropriate scepticism.

Why SEO has to adapt rather than retire

The honest read on AI search and SEO in 2026 is that SEO is changing shape, not dying. Three reasons.

• AI systems are still reading the web you already optimise. ChatGPT Search, Perplexity, Gemini and Google's AI Overviews all crawl the same web pages your existing SEO is built around. The technical foundations (crawlable HTML, clean structured data, a fast site, indexable URLs) still matter. If anything they matter more, because the AI extraction step is less forgiving than the Google ranking step ever was.
• The traffic loss is uneven, not catastrophic. Our working position across client sites: AI Overviews reduce click volume by 20 to 30 percent on pure informational queries, where the user gets a usable answer in the Overview and never needs to click. Commercial intent (people ready to buy, call or book) is far less affected because the searcher still needs the business. Local intent with the local pack and Maps is barely affected at all. Most service-business revenue lives in those last two buckets.
• The work to win citations is the work to win rankings, sharpened. The pages we see picked up inside AI answers are the same pages we would have flagged as strong-on-traditional-SEO three years ago: clear topic coverage, named expert author, clean schema, decent backlink profile. AI search has not invented a new ranking factor; it has put a steeper price on the ones that already mattered.

Kinda unsexy way to say it. The agencies selling "GEO services" as a separate line item are mostly selling sharpened SEO with new branding. We do not run AI search as a separate service; we have folded the AI-specific moves into the existing SEO service and the website audit process. The detail of how those moves layer onto traditional SEO is what this pillar covers. For the broader question of "what is SEO" before we get into the AI-specific moves, the What is SEO pillar is the foundation.

How AI search behaves in 2026

Six patterns from running AI-visibility audits across Perth and WA client sites this year. None of these are algorithmic facts; all are observation.

AI Overviews appear on roughly half of informational queries

For the informational queries we monitor across client sites, Google's AI Overview shows up on something close to half of them. Coverage is higher on health, finance, how-to and general "what is" queries. Coverage is lower on local-intent, transactional and brand queries. The implication: if your traffic comes from informational content, expect a meaningful click-volume hit. If your traffic comes from local-service-plus-suburb queries (most Perth trade businesses), expect the AI Overview to barely touch you.

Citations cluster around a small number of trusted sources

Across all four AI clients, citations cluster heavily on a small number of sources. Wikipedia, Reddit, major news outlets, large authority sites in the topic, and the top three to five organic ranks on the underlying SERP. Below that, the long tail of sources is thinner than the equivalent SERP would be. The practical read: getting into the citation set requires the same authority signals that get you into the top of the regular SERP, with very little margin for "almost ranking".

Entity identity matters more than it did

AI systems need to know which organisation and which person they are citing. Pages where the entity is ambiguous (no Organization schema, no consistent NAP, no Person schema for the author) get cited less than equivalent pages where the entity is locked down. See entity SEO for the page-level work and schema for AI for the structured-data layer.

E-E-A-T is now a real ceiling, not just a guideline

Pages without named expert authors, transparent organisational identity, and citable claims get summarised but rarely cited. The pattern is most obvious on YMYL topics (health, finance, legal) where AI systems are visibly more conservative about source selection. See E-E-A-T explained for the framework that ties this together.

Schema is now a confidence signal, not just a SERP feature

Structured data used to be mostly about triggering rich results in the SERP. In 2026 it is also a confidence signal AI systems use to confirm what a page is about and which entity to attribute it to. Organization, Person, LocalBusiness, Article, FAQPage and HowTo schema all carry weight here. See schema markup and the schema for AI chapter.

AI referral traffic is small but high-intent

Across client GA4 properties, AI referrer traffic (ChatGPT, Perplexity, Gemini, Copilot) ranges from less than 1 percent to around 4 percent of organic-equivalent sessions. The volume is small. The intent is unusually high: the user has already asked a specific question and is clicking through to verify or buy. Conversion rates on AI-referred sessions often run higher than on regular organic, which is why the channel matters even at small volume. See tracking AI referrals for the measurement method.

The 8 sub-topics that make up the pillar

This pillar splits into eight chapters. Each one covers a sub-topic you will hit the moment you start taking AI search seriously for an Australian business.

1. AEO vs GEO vs SEO. Definitions, where they overlap, where they differ, and what each one actually means for the work you do this quarter.
2. Google AI Overviews explained. How AI Overviews are generated, what triggers them, the click-volume impact we measure, and the on-page moves that increase your odds of being cited inside one.
3. ChatGPT Search and SEO. How ChatGPT's web search works, which signals it appears to weight, and the practical optimisation moves we observe correlating with citation pickup.
4. How to get cited in Perplexity. Perplexity's citation-first design, the source-quality patterns we observe, and how to give the system the structured signals it parses well.
5. Gemini visibility for SEO. Google Gemini's relationship with the wider Google index, entity identity, and what businesses can do to be picked up inside Gemini answers.
6. The llms.txt standard. What the proposed standard does, who has adopted it, and the cheap-insurance argument for adding it to every client site today.
7. Schema markup for AI search. The schema types that matter most for AI extraction, how to layer them properly, and how to validate the output.
8. Tracking AI referrals in analytics. A working method for surfacing AI-referred traffic inside GA4 plus server-log analysis, with the limits of each approach.

Our framework: the AI-visibility stack

Every AI-visibility audit we run for a Perth or WA client is built around four layers. Skip a layer or work them out of order and the AI moves stop reinforcing each other.

Layer 1: Traditional SEO foundations

The crawlable, indexable, fast, well-structured site that every AI client still reads. Without this layer, the AI-specific moves on top have nothing to attach to. The Technical SEO pillar covers the crawl and indexation layer, and the On-Page SEO pillar covers the page-level structure. Core Web Vitals covers the speed layer. None of this is optional. AI search penalises broken foundations harder than traditional Google did.

Layer 2: Entity identity and E-E-A-T

The layer that tells AI systems which organisation and which person they are reading. Organization schema, Person schema for authors, consistent NAP, named credentialled authors on YMYL topics, a transparent About page, the right linking pattern between Person and Organization. This is where most sites lose ground first. See entity SEO and E-E-A-T explained for the on-page treatment.

Layer 3: AI-specific structural signals

The AI-search-only moves that build on the first two layers. Schema layered properly for the page type (Article, FAQPage, HowTo, LocalBusiness, Service, Product). An llms.txt file at the root of the domain. Clean question-and-answer formatting on FAQ content so the AI extractor has something to lift. Section headings that match the actual sub-questions inside the topic. Citable claims with named sources where the page is making a factual assertion. See schema for AI and the llms.txt chapter.

Layer 4: Measurement and iteration

The feedback loop that tells you whether the first three layers are working. Custom referrer groups in GA4 for the AI clients, periodic server-log review for AI crawler user agents, manual citation checks across the four major AI systems, and a tracking schedule that catches change. AI search visibility moves more abruptly than traditional rankings; a site that was being cited last month may not be this month, and only a measurement loop catches it. See tracking AI referrals and the SEO Measurement pillar.

The order matters. Most agencies sell Layer 3 (the AI-specific moves) as a separate product because it is the most visible. The stack falls over without Layers 1 and 2, and without Layer 4 you cannot tell whether any of it is working.

Where most businesses get this wrong

From auditing Perth and WA businesses through 2025 and 2026, the same six failure modes come up.

The single biggest mistake we see in 2026 is the panic-purchase of "GEO services" from agencies that have not earned the right to sell them. The same client who needed a basic technical SEO audit six months ago suddenly wants AI search optimisation; the agency happily sells the latter without fixing the former, and the client gets neither. The honest sequencing is: traditional SEO foundations first, entity and E-E-A-T second, AI-specific moves third. The audit framework we use for new clients is the website audit service, and the entry-level diagnostic is the free SEO audit tool.

Tools and a checklist

You do not need a separate "AI search stack" to run this work. You need six inputs, most of which you already have if you run SEO at all.

1. Google Search Console. Free. The Performance report still surfaces the queries and pages picking up impressions inside AI Overview-modified SERPs. The new "Insights" view occasionally flags AI Overview presence. See the GSC glossary entry.
2. Google Analytics 4 with a custom AI-referrer channel. Free. GA4 is where the AI referral traffic actually lands, but you need a custom channel to surface it cleanly. The tracking AI referrals chapter includes the setup steps.
3. A schema validator. Google's Rich Results Test plus Schema.org's validator. Both free. Used to confirm that the structured-data layer is parsing cleanly before the AI clients see it.
4. A manual AI-citation check rhythm. Pick ten priority queries. Run them through Google (with AI Overview), ChatGPT Search, Perplexity and Gemini once a month. Record which sources get cited. Cheap, manual, and impossible to automate reliably yet.
5. Server log analysis for AI crawlers. A monthly check of the access logs for user agents like GPTBot, PerplexityBot, ClaudeBot, Google-Extended and the Bing/Copilot variants. Tells you whether the AI clients are actually crawling the site, even when no referral traffic shows up.
6. A llms.txt file. A static file at the root of the domain. Cheap to create, harmless if ignored, useful if adoption broadens. See the llms.txt chapter for the template.

For the wider Hub context that this pillar plugs into, the robots.txt chapter covers the file that sits next to llms.txt, the log file analysis chapter covers the server-log side, and the SEO Measurement pillar covers the broader tracking discipline. Perth businesses with a local-intent mix should also read Google Business Profile and Local SEO Perth for the local-pack side.

A 10-point AI search readiness checklist

1. Does the site have valid Organization schema with sameAs links to its brand profiles?
2. Does every published page have a named author with Person schema and a real bio?
3. Has the site's E-E-A-T layer (About page, named experts, citable claims) been audited in the last year?
4. Is the structured-data layer (Article, FAQPage, LocalBusiness, Service, Product) clean and validating?
5. Do FAQ blocks use a question-and-answer format the AI extractor can lift cleanly?
6. Is there a llms.txt file at the root of the domain?
7. Is robots.txt configured intentionally for the AI crawler user agents (allow or disallow, by design)?
8. Is there a custom AI-referrer channel in GA4 surfacing ChatGPT, Perplexity, Gemini and Copilot traffic?
9. Is there a monthly manual citation check across the four major AI systems on priority queries?
10. Is there a quarterly review rhythm that catches changes in AI visibility before they become a trend?

What to read next

Once you have read this pillar, the natural next steps are:

• AEO vs GEO vs SEO. Start here if you are still untangling the terminology.
• Google AI Overviews explained. The single AI-search surface that touches the most traffic for most Australian businesses.
• Schema markup for AI search. The structured-data layer that most sites get wrong first.
• Tracking AI referrals in analytics. So you can actually see whether any of this is working.
• Entity SEO. The cross-pillar piece that ties entity identity together at the page level.
• E-E-A-T explained. The cross-pillar piece that ties expertise and authority together.
• Topical authority. The cross-pillar piece that explains why depth of topical coverage matters as much as page-level optimisation.
• SEO Glossary. The AI-search-related entries (AEO, GEO, AI Overviews, LLM, llms.txt, schema markup) are all defined in plain English.