I've watched the web reinvent itself three times.
This one is different.
I started in digital in 1999, in my first job out of the gate. Founded my own agency in 2001. Sold it. Bought it back. Built another. Somewhere along the way Top4 Technology grew into a platform serving 200,000+ businesses globally — from ASX-listed brands and government departments to small clinics and family trades. Property developers in Surabaya. Plumbers in Parramatta.
I have lived through Google ranking, the social era, mobile, voice. None of those shifts made the previous playbook worthless. They added a layer. The work compounded.
This one is different.
AI search hasn't added a layer. It has replaced the layer everyone was standing on.
If you sell anything to anyone, the rules just changed. Not in a 'we should look at this next quarter' way. In a 'your traffic is already moving and your team hasn't noticed yet' way. The brands and businesses that adapted in the last twelve months are pulling away. The ones still optimising for page-one Google are watching their organic traffic drop 30, 40, 50% and chalking it up to 'the algorithm.'
It isn't the algorithm. It's a new system.
Why I wrote this in 2026.
I wrote the first edition of this guide in September 2025. The shape of the opportunity was clear by then. The detail wasn't. I had to make some bets about how things would develop, and frankly, some of those bets were wrong.
llms.txt was supposed to matter. It doesn't, yet — and probably won't in the way most agencies sold it. ChatGPT Instant Checkout was supposed to be the agentic commerce on-ramp. OpenAI quietly killed it in March. AI Mode was supposed to be a slow rollout. It hit 75 million daily users by the time you're reading this.
So this edition is the version with the bets corrected. With the new mechanics — query fan-out, retrieval before ranking, embeddings, the centroid model that determines whether AI even sees you — explained in language a marketing director can act on.
None of this is theoretical. Every framework in this book is one we run for clients at Top4 Technology. Every number is one I have tested. Every recommendation is one I'd bet my own business on.
Which I am.
"The brands winning today aren't running clever AI strategies. They are running unsexy fundamentals with discipline, and they started six months before everyone else."
Michael Doyle · Sydney · April 2026
If this guide saves you from making the wrong bet at the wrong moment, it has done its job. If it gives you the language to brief your team, sell it to your CFO, or pick the right agency partner, even better. And if it convinces you to stop reading and start acting — that's the goal.
The next training cycle is coming. Let's get you in it.
— Michael
Eighteen months that broke search.
Your traffic is moving. Your team probably hasn't noticed yet.
In September 2025 I published the first edition of this guide. AI search was a thesis. Eighteen months later it is the channel. Faster than I expected. Messier than the analysts predicted. And the businesses that adapted in those eighteen months are already pulling away from the ones that didn't.
of all tracked Google queries now trigger an AI Overview, up 58% year on year.
BrightEdge · Mar 2026
of Google AI Mode searches end with zero clicks to any external website.
Seer Interactive · 2025
surge in traffic from AI sources to small business sites in the last 12 months.
Previsible · Jan 2026
Two things are happening at once. Traditional Google traffic is shrinking. AI-mediated traffic is exploding. They behave nothing like each other.
People who arrive from an AI answer convert at roughly 4× the rate of standard organic traffic. But only if your business is the one being recommended in the first place.
That's the whole game now. Not 'ranking on page one.' Being named in the answer.
What changed since the last edition.
If you read the 2025 edition and built against it, you got most of the foundation right. But six things have shifted in ways that demand a fresh playbook.
1. Google launched AI Mode — and it's a different beast
AI Mode is not the same product as AI Overviews. Overviews sit above traditional results and let people click through. AI Mode replaces the results page entirely with a conversational answer. It hit 75 million daily users by late 2025 and became publicly available across the US in March 2026. Sessions average 49 seconds — more than double an Overview session — and the people inside them are doing real research. They are also, overwhelmingly, not clicking out.
2. ChatGPT's commerce strategy reset
OpenAI launched Instant Checkout in September 2025, then quietly retired in-chat checkout in March 2026 in favour of merchant apps and dedicated discovery experiences. Google countered with the Universal Commerce Protocol at NRF in January 2026. The infrastructure is moving faster than most retailers can keep up with — but the underlying truth is settled: agents are doing the shopping, and your product feed is now your storefront.
3. llms.txt got hyped, then got real
The 2025 advice — 'implement llms.txt now' — was premature. As of early 2026 no major LLM provider officially honours the standard, despite 844,000+ sites adopting it. We give you the unvarnished take in Chapter 5.
4. Local AI visibility is the new local SEO
Perplexity is Bing-first. Gemini draws from Google Business Profile. ChatGPT pulls live data from Bing too. Local businesses with clean directory citations and proper schema are getting cited at rates that didn't exist 12 months ago.
5. AI citation tracking became a category
Profound, Otterly, AthenaHQ, Rocketito, Scrunch, AIclicks — there are now real tools that show you where you appear in AI answers and where you don't. Treat them like the early Search Console. Imperfect, occasionally weird, but essential.
6. Bottom-of-funnel content beats top-of-funnel
Generic how-to content has been cannibalised by AI Overviews at almost 100%. Comparison pages, product pages, case studies with real numbers, and pricing pages are now the formats earning citations and the trickle of clicks that remain.
From search to synthesis.
The job of a search engine used to be to find things. The job of an AI engine is to decide things — what's true, what's relevant, what's worth recommending, and what's not worth mentioning at all.
That distinction sounds philosophical. It is in fact the entire commercial difference between the old web and the new one. A ranking system rewards the most relevant page for a query and lets the user adjudicate. A synthesis system makes the adjudication itself, then presents a single, confident answer with three to ten cited sources baked in. If you are not one of those sources, you are invisible. There is no second page.
The four engines that matter.
Google AI Mode & AI Overviews
~48% of queries · 75M daily users
Two products in one ecosystem. AI Overviews are the quick AI answer above the blue links; AI Mode is the full conversational replacement. Driven by Gemini, deeply integrated with Google's index, Knowledge Graph, and Google Business Profile.
- Cites organic results that already rank well
- Heavy weighting toward schema and structured data
- Strong local signal via Google Business Profile
ChatGPT (OpenAI)
700M+ weekly users · 1B+ daily queries
The behavioural default for 'ask AI' worldwide. Live web access via ChatGPT Search is powered by Bing. Cites sources in only ~16% of responses — which makes the slots that exist disproportionately valuable.
- Bing index is the gateway
- Listicles, articles, and product pages dominate citations
- Brand authority and entity clarity matter more than backlinks
Perplexity
~97% of answers carry citations
The most citation-friendly engine by a wide margin. Real-time crawl, fresh content bias, draws heavily from Reddit, Quora, and authoritative third-party publications. Bing-indexed under the hood.
- Rewards visible publish/update dates
- Loves original data, statistics, and proprietary research
- Reddit presence in your category genuinely moves the needle
Microsoft Copilot & Claude
Enterprise reach + premium reasoning
Copilot is the workhorse inside Microsoft 365 and Edge — quieter than ChatGPT in consumer mindshare, but shipping commerce features faster. Claude (Anthropic) is the long-context model used heavily by professionals and developers.
- Copilot Checkout is live with Shopify, PayPal, Etsy
- Both reward genuinely well-structured, factual content
- Claude weights authoritative sources particularly heavily
How synthesis actually works.
It helps to understand the mechanism, because it tells you what to optimise for. When you type a question into Google AI Mode, the system uses a technique called query fan-out: a single question is decomposed into roughly sixteen related sub-queries, each one searched in parallel, the results filtered for credibility and relevance, and the whole lot synthesised into one coherent answer with citations.
"Ranking is no longer the prize. Inclusion is. And inclusion is decided by whether the model believes you are the consensus."
The 2026 reality, in one line
The zero-click economy is real, and it's fine.
The most common objection to all this: 'if 93% of AI Mode searches end without a click, what's the point?' Fair question. The answer is uncomfortable but mathematically clean.
The traffic you used to get from informational queries — 'how does X work,' 'what is Y,' 'guide to Z' — is gone. Permanently. That click was always low-intent and low-value; you just didn't notice because volume hid the conversion problem. What's left in the click stream is high-intent traffic: comparison pages, 'best X for Y' queries, transactional searches, branded searches. Those still click through, and they convert at 4× or higher.
How AI actually sees your brand.
It doesn't read your homepage. It does the maths.
AI engines do not see your business the way you describe it on your About page. They build their own version of your brand from the content you have published — every page, every blog post, every product description, every PR mention — and they store that version as a mathematical object. Not a sentence. A position in space.
1 · Retrieval comes before ranking
In classical SEO, you fight to move from page two to page one. In AI search, the fight happens earlier. Before anything is ranked, the system decides which content is even eligible to be considered. That step is called retrieval. If you're not retrieved, you don't lose. You just never entered the contest. This is the single most important mental model shift in 2026. Optimisation is no longer about position. It's about admission.
2 · Pages don't compete — passages do
AI doesn't treat your page as one unit. It breaks it into chunks: paragraphs, sentences, sections. Each chunk is evaluated on its own merit against the query. A paragraph buried halfway down a guide can win on its own. A homepage hero can be ignored entirely. The competition is passage versus passage, not page versus page. This is why the format guidance in this book is so specific — direct answers in the first 80 words, question-shaped headers, FAQ blocks, TL;DRs. Every chunk has to earn its own place.
3 · Your "brand" is a centroid
Here's the bit most marketing teams aren't ready for. Every chunk you publish is converted to a vector — a position in high-dimensional space that represents its meaning. All those vectors cluster together. The centre point of that cluster is what AI thinks your brand is. Not your tagline. Not your brand guidelines. The mathematical average of everything you've ever written. Two implications. First, every page you publish either reinforces that centre or pulls it sideways — drift is a real, measurable phenomenon. Second, brands that follow the same playbooks and write about the same things end up in the same place in meaning-space. The technical term is 'cluster collision.' The plain-English version: you and your three competitors all sound identical to AI, so the model picks one of you and ignores the rest.
The framework researchers call this 'the centroid model' of brand visibility. The practical version is simpler. Every chapter that follows is, ultimately, a different way of pulling your content into a clearer, more distinct centroid that AI can recognise, retrieve, and recommend. Hold that picture in your head as you read on.
What this means for your team this quarter.
If you take only one thing from this chapter, take this: the unit of optimisation has changed from page to entity. Google AI Mode and ChatGPT do not rank pages. They model entities — your brand, your people, your methodology, your products — and decide whether to mention them. Pages still matter, but only as the evidence layer that proves the entity is real, current, and credible.
That reframe drives every chapter that follows:
- Chapter 2 shows you what citation actually looks like and what behaviours buy it.
- Chapter 3 breaks down each platform's preferences in detail.
- Chapter 4 covers entity-level positioning — naming your methodology, claiming a specific outcome, drawing a small clean line around your expertise.
- Chapter 5 handles the technical evidence layer: schema, structured data, what to do about llms.txt.
- Chapters 6–10 apply the model to local businesses, eCommerce, agents, measurement, and the verticals Top4 sees most often.
One more thing before we move on. The temptation in this market is to chase platforms — 'we need a ChatGPT strategy, a Perplexity strategy, a Gemini strategy.' Don't. The platforms differ in mechanics, but they all reward the same underlying behaviours: clear positioning, structured content, third-party validation, fresh activity, consistent identity, and original evidence. Build that foundation once and it pays across all of them.
The agencies and brands winning right now are not running four parallel optimisation programmes. They are running one good one and pointing it at four pipes.
Query Fan-Out. Why you don't have to win the whole question.
Here is the single most useful thing to understand about how Google's AI answers are built: you are no longer competing for one page to win one search.
For twenty years the game was simple to describe, even if it was hard to play. Someone typed a phrase. Google ranked a list of pages against that phrase. You fought to be the page at the top. One query, one winner, ten blue links.
That is not how AI Mode and AI Overviews work. When someone asks a real question — the messy, over-loaded kind people actually type — Google no longer hunts for a single page that matches the whole thing. It quietly pulls the question apart, answers each piece separately, and stitches the best bits back together into one response. The technique has a name, and it is Google's own: query fan-out.
What Google is actually doing.
The idea traces to a Google patent from 2024, but you no longer have to take a patent's word for it. In its official Search Central documentation, Google now describes fan-out plainly: when you ask a question, the model generates a set of related searches, runs them at the same time, and gathers results across all of them to build one answer. Alongside it sits a second technique Google calls retrieval-augmented generation, or grounding — pulling live, relevant pages from its search index so the answer is built on current sources, with clickable links back to them.
Google's own illustration is a homeowner asking how to fix a lawn full of weeds. Behind the scenes, the fan-out queries might be "best herbicides for lawns", "remove weeds without chemicals", and "how to prevent weeds in lawn". Three different pages could each win one of those, and the homeowner never sees the seams.
"Best split-system air conditioner for a small bedroom, quiet at night, under $1,500 installed."
One sentence, three separate searches running underneath it. The old system looked for one page carrying that entire phrase. The new system breaks it into underlying needs and searches each in parallel: which units suit a small room, which are genuinely quiet, what supply-and-install actually costs at that budget. A specification page might win the noise question. A trade blog might win the pricing question.
The shift that changes your content strategy.
Because the answer is assembled from pieces, you do not have to be the best answer to the entire question. You only have to be the best answer to one part of it.
"You have permission to stop writing one heroic mega-page that tries to be everything to everyone. Own one facet completely on each page instead."
The Shift in One Line
Read that again, because it inverts fifteen years of SEO instinct. That page was always a fantasy — too broad to be genuinely expert on any single point, and too generic for a machine to trust on the specifics. The winning move now is the opposite: identify the facets hiding inside a big question, and own one facet completely on each page, then tie them together so both the reader and the model can see they belong to the same body of work.
The facet play, step by step.
The line you must not cross.
The temptation, once you understand fan-out, is to mass-produce a page for every conceivable sub-query — a hundred thin pages, each targeting one variation, hoping to carpet-bomb the answer box. Don't.
Google names this exact move in the same documentation and calls it what it is. Spinning up separate pages for every variation, mainly to manipulate rankings or AI responses, breaks its scaled-content-abuse spam policy. In Google's own words, a high quantity of pages does not make a website higher quality — and its systems are now good enough to understand a page's relevance even when the words don't match the query exactly.
What Google has actually said. The official 2026 guidance.
Most of what you'll read about 'ranking in AI' is guesswork dressed up as expertise. When Google publishes something on the record, that beats anyone's theory — including mine.
In 2026, Google finally put a lot of it on the record. In May it published formal documentation — "Optimising your website for generative AI features on Google Search" — under a new Generative AI fundamentals section of Search Central, and it has been updating it since. Here's what it actually says, and what it means for you.
1 · There is no secret, and Google said so in writing
The fundamentals of good SEO are the fundamentals of AI visibility. Google's generative features are built on its core ranking and quality systems — the same crawling, indexing, relevance and quality signals that have always mattered. If your site earns its place in normal Search, you're already most of the way to being in the answer.
2 · Fan-out and RAG are confirmed, in Google's own words
The two techniques behind these answers are no longer inference. Google confirms it uses retrieval-augmented generation to ground answers in live pages from its index, and query fan-out to break a question into concurrent sub-searches. The facet strategy in the previous chapter maps directly onto how the system is documented to behave.
3 · The myths Google explicitly told you to ignore
For Google Search specifically, you can stop worrying about special AI files like llms.txt (Google ignores them), "chunking" content into fragments (there's no ideal page length), rewriting purely for AI, chasing inauthentic mentions, and special structured data — none unlock AI inclusion, though schema still earns rich results in normal Search.
4 · To be in the answer, you first have to be in the index
A page can only appear in AI features if it's already indexed and eligible to show in normal Search with a snippet. There is no separate AI index and no separate submission. The boring work — crawlable site, clean internal links, indexable pages, good page experience — is still the price of admission.
5 · On "AEO" and "GEO" — read this one twice
Google addresses the acronyms head-on. From its perspective, optimising for generative AI search is optimising for search, and thus still SEO — and it openly tells readers to be sceptical of third-party "AEO" or "GEO" services. That's worth sitting with honestly, because this book carries one of those labels.
6 · Local business and shopping get pulled in too
Google states its AI responses can surface product listings, product details and local-business information — and that keeping your Google Business Profile and Merchant Center data complete and accurate helps you appear in them. Your Business Profile is no longer just a map pin, it's a feed into the AI answer.
7 · Agents are starting to use your site like a user
Browser agents may read your page the way a person's assistant would: the rendered screen, the underlying structure, the accessibility tree. The test is simple: can an agent complete a real task on your site — find a product, fill a form, book a slot — without getting stuck? If it can't, that's tomorrow's lost sale.
8 · You can now be chosen and measured
Google brought Preferred Sources into AI Overviews and AI Mode — users can nominate sites they trust, which get highlighted plus a "Highly Cited" badge for original reporting. Search Console now reports how your pages appear in AI features, alongside a toggle to opt out entirely. For almost every business reading this book, opting out is the wrong move.
9 · Mind the crawler controls — they don't do what people think
Blocking Google-Extended does not pull you out of AI Overviews or AI Mode, because those features draw on the normal search index via Googlebot. The usual ways of "hiding" from AI either don't work or cost you your normal visibility too. Don't let a well-meaning developer quietly switch you off.
10 · The scale is the reason this matters
Google states AI Overviews now reach over 2.5 billion monthly users, and AI Mode has passed one billion, with AI Overviews appearing on a large share of everyday searches. People are asking longer, stranger, more specific questions than they ever typed into a search box — exactly the multi-part questions fan-out was built to answer.
"Be genuinely useful, be technically accessible, be specific enough to own a facet without farming pages, keep your business data clean, and be trustworthy enough to cite — everywhere the answer gets built, not just on Google."
Google's 2026 guidance — the through-line
The citation economy, by the numbers.
The 2025 edition called this the '3 Rs' — Reputation, Recency, Authority. The instinct was right. The data is now sharp enough to be specific.
The numbers that should be on your strategy slide.
conversion rate of AI-referred traffic vs traditional search.
eMarketer · 2025
organic CTR uplift for brands cited in an AI Overview vs not cited.
Seer Interactive
more frequent crawling by LLM bots than Googlebot, by April 2026.
Search Engine Journal
of URLs cited by ChatGPT, Perplexity, and Copilot rank in Google's top 10 for the same query.
Ahrefs
of all LLM citations come from the first 30% of a page's text.
Position Digital
of all ChatGPT citations within a topic concentrate in just the top 10 domains.
Growth Memo · Mar 2026
Read those numbers carefully.
Three of those stats are particularly worth chewing on, because they overturn long-held SEO assumptions.
Only 12% of cited URLs rank in Google's top 10. Translation: ranking and citation are completely separate scoreboards. You can be invisible on Google and prominent in ChatGPT. You can be number one on Google and ignored by Perplexity. The scoreboards do not talk to each other.
44.2% of citations come from the first 30% of the page. Translation: if your answer is buried below a 600-word introduction, the model never finds it. The TL;DR section, the lead paragraph, and the H2 directly under the title are now the most valuable real estate on any page you publish.
The top 10 domains take 46% of citations. Translation: this is a winner-take-most economy. The wide middle of the SEO market — pages 2–10 of Google for years — barely exists in AI. You are either the consensus or you are background.
What buys a citation in 2026.
1 · Entity clarity: Your business needs to be a coherent entity in the model's worldview. That means consistent name, consistent description, consistent founder/leader attribution, and a Wikidata or Knowledge Graph footprint where possible.
2 · Structured directness: Headers that match real questions. Direct answers in the first 60–80 words. FAQ sections written in the voice your customer actually uses. Tables for comparisons. Bullet lists for processes.
3 · Original evidence: Proprietary data, original surveys, internal benchmarks, named methodologies, real client outcomes with real numbers. Models reward the source they can't get elsewhere.
4 · Distributed presence: You cannot win AI visibility from your website alone. You need consistent appearances across podcasts, industry publications, directories, review platforms, and Reddit and Quora threads in your space.
5 · Recent activity: Visible publish dates. Recent updates. Active social signals. Press in the last 90 days. Models heavily favour recency for almost every category.
The contrast point is also worth naming. Brands that are losing AI visibility right now share a profile: weak entity definition, generalist positioning, content optimised for keywords rather than questions, no original data, all activity concentrated on their own domain, and last-updated dates from 2023.
Eight channels. One website at the centre.
Every channel in this guide hangs off the same hub. Your website is the entity; the eight rings around it are the surfaces that prove it to humans, search engines, and AI models.
Read the wheel inside-out. The centre is your site — the entity AI models must recognise. The inner ring is the eight Top4 channel categories. SEO, AI & Automation, Reputation, and Google Business Profile are where AI visibility is won or lost; the others are the corroboration layer that makes the model believe you.
The compounding growth system.
The framework we run for 65,000+ business subscribers — from first AI mention to loyal client.
The 2026 GEO stack — what each platform actually rewards.
The acronym soup matters less than people think. SEO. AEO. GEO. LLMO. They describe the same thing from different angles. What matters is knowing the specific levers each platform pulls.
GOOGLE AI MODE & AI OVERVIEWS
Schema markup now non-negotiable (Organization, Person, Service, Product, Article, FAQPage, Review, LocalBusiness). Top-10 organic ranking still matters — pages ranking #1 have 58% chance of AI Overview citation; outside top 20 are 3.5× less likely. Knowledge Graph entity status is gold. Branded queries get 18% CTR uplift.
CHATGPT (OPENAI)
Bing Webmaster Tools — submit sitemap, fix indexation errors, use IndexNow. Check robots.txt for OAI-SearchBot and ChatGPT-User — 70% of citations come from sites not blocking these bots. Listicles and comparison content — 21.9% of all citations are listicles. Brand authority over backlinks.
PERPLEXITY
Visible dates on every page. Bing Places, not just Google Business Profile. Reddit and Quora presence — Perplexity scrapes both heavily. TL;DR blocks at top of pages — 4× lift in citations observed.
MICROSOFT COPILOT, GEMINI & CLAUDE
- Copilot integrated into Microsoft 365, Edge. Copilot Checkout live with Shopify, PayPal, Etsy. Same Bing-index as ChatGPT.
- Gemini draws from Google Search, Google Business Profile, Google Maps, YouTube. For local businesses path = excellent local SEO — fully optimised GBP, 50+ recent reviews, regular Google Posts.
- Claude used by developers, professionals. Weights authoritative sources heavily — research papers, official documentation. Requires documentation quality, GitHub presence, trusted reference sources.
The unified stack.
- Foundation — Indexable site, clean robots.txt, complete schema, claimed entity
- Content — Question-shaped headers, direct answers in first 80 words, comparison and listicle formats, original data
- Distribution — Directories, reviews, podcast appearances, guest articles, Reddit/Quora
- Recency — Quarterly refresh cycles, PR drumbeat, active social, regular case studies
- Measurement — Citation tracking, bot analytics, AI referral tracking in GA4
The unfair advantage hiding in your robots.txt.
# Block training-data harvesters
User-agent: GPTBot
Disallow: /
User-agent: Google-Extended
Disallow: /
# Allow live retrieval bots
User-agent: ChatGPT-User
Allow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: ClaudeBot
Allow: /Positioning & frameworks that get cited.
If the model can't tell what you do or who you do it for, it won't recommend you. Generalists lose in AI search the way generalists lose in any market with infinite supply.
The narrowness paradox.
Most founders resist tight positioning because it feels like leaving money on the table... AI engines are optimised to provide the most relevant answer to a specific query. When someone asks 'marketing strategy,' the model surfaces general resources. When they ask 'lead generation for cosmetic clinics with a single location and under $100k in monthly revenue,' the model picks the specialist who has spent two years writing about exactly that.
"The narrower your focus, the more queries you become the definitive answer to. Cumulative wins beat broad wins every time in synthesis search."
The Narrowness Paradox
The 2026 positioning template.
Generic: 'We help businesses grow through digital marketing.'
Sharp: 'We help Australian cosmetic clinics generate qualified bookings through the Clinic Demand Engine — a SEO, AI Search, and review-management system designed for single-location practitioners.'
Frameworks: still the highest-leverage IP move.
Named methodologies remain the most efficient way to compress your expertise into something AI engines can recognise, attribute, and recommend.
- Frameworks need to be schema-marked up (HowTo or DefinedTerm)
- Frameworks need their own dedicated page — a canonical URL
- Frameworks need to appear in third-party content
The framework playbook in eight moves.
- Identify your repeating process
- Name it with intent — "The Demand Engine," "The Pipeline Acceleration Method"
- Trademark it
- Build the canonical page
- Create supporting content
- Use it everywhere
- Get other people saying it
- Refresh it annually
Two warnings.
Don't trademark a framework you don't use. Don't proliferate frameworks.
Case study: framework as moat.
A Sydney HVAC contractor... built a framework called the 'Five-Point Comfort Audit.'... Twelve months in, asking ChatGPT, Perplexity, or Gemini 'what's the best HVAC inspection process in Sydney' returns the Five-Point Comfort Audit as a named methodology... in roughly 60% of responses. Their phone-call volume from AI-attributed sources has tripled.
Action list: Audit your homepage... Identify the repeatable process... Draft three candidate names... Trademark search... Build the canonical framework page... Brief every customer-facing team member...
Technical foundations — schema, llms.txt, the honest version.
Most of this chapter is non-controversial. Schema works. Structured content works. JSON-LD works. The controversial part is llms.txt — and the 2025 edition of this guide oversold it. Here's the corrected take.
Schema markup is the single biggest 2026 lever.
- Organization — entity disambiguation
- LocalBusiness (Restaurant, MedicalBusiness, ProfessionalService)
- Person — for founders, key team members
- Service / Product
- FAQPage
- Review & AggregateRating
- HowTo / DefinedTerm — for named frameworks
- Article / NewsArticle
- BreadcrumbList
The llms.txt situation.
So should you implement it? If you have technical documentation, API, or developer-facing product — yes. If you run a content site, marketing site, or eCommerce store — it's optional.
# Top4 Technology — Expertise Index
## What we do
We build digital marketing systems for Australian SMBs and government clients,
with deep specialism in AI search visibility, Kentico CMS, Drupal, and
Google Premier Partner-level performance media.
## Core methodology
The Top4 Demand Engine — a SEO + AEO + Local Search + Reviews stack designed
for service-based businesses with $1M-$50M annual revenue.
## Documentation
- [Services overview](https://www.top4technology.ai/services)
- [Industries we serve](https://www.top4technology.ai/industries)
- [Case studies](https://www.top4technology.ai/work)
## Contact
Email: hello@top4.com.au
Web: https://www.top4technology.aiOther technical foundations: Bing Webmaster Tools, Google Business Profile, Bing Places, IndexNow, Wikidata entry, Site speed.
Page-level structure for AI citation.
Local AI visibility — the unfair advantage for SMBs.
If you run a local business — a clinic, a law firm, a tradie, a real estate agency, a restaurant — the AI search shift is the best thing to happen to your category in fifteen years.
Here's the asymmetry. Big national brands are losing AI visibility because they can't write geographic-specific content at scale... Local businesses can. And the AI engines reward exactly that specificity.
- Entity consistency across local search infrastructure — NAP audit, directory list (Google Business Profile, Bing Places, Apple Business Connect, industry-specific directories, Yelp/TripAdvisor/Trustpilot)
- LocalBusiness schema with full property completion — specific subtypes, areaServed, geo coordinates, openingHoursSpecification
- Geographic content depth — neighbourhood-level pages, suburb-specific guides
- Multi-platform review velocity — across at least three platforms
The 90-day local AI visibility plan.
Days 1–30 · Foundation
NAP audit, LocalBusiness schema, robots.txt audit, Bing Webmaster Tools, Wikidata entry
Days 31–60 · Content
Location pages, TL;DR blocks, FAQ page, question-shaped H2s, two case studies
Days 61–90 · Distribution & signals
Review request process, pitch podcasts/publications, Reddit/Quora participation, local press release, AI citation tracking
Three months in, you should be seeing first citations in Perplexity (the fastest engine to recognise new entities) and improved presence in Gemini local results.
Agentic commerce — when the agent does the buying.
If you run an eCommerce business, this is the chapter you've been waiting for. The shift to agent-driven shopping is happening faster than anyone planned, and the rules for being discoverable are completely different from traditional eCommerce SEO.
Quick definition: agentic commerce is shopping where an AI agent does the discovery, comparison, and increasingly the checkout, on behalf of a human. The human sets the intent ("I need running shoes for flat feet under $150 that ship to Sydney") and the agent handles the rest. This is not a 2030 prediction. It's a 2026 reality.
ACP — Agentic Commerce Protocol
Co-developed by OpenAI and Stripe. Powers checkout sessions inside ChatGPT. Now adopted by 25+ partners including Salesforce, Squarespace, and Adobe Commerce. Focuses on the purchase-stage transaction within the ChatGPT ecosystem.
UCP — Universal Commerce Protocol
Co-developed by Google and Shopify, launched at NRF January 2026. More comprehensive — covers the full journey from product discovery through post-purchase support. Partners include Visa, Mastercard, Stripe, Walmart, Target, Etsy, and Wayfair.
You don't need to implement either protocol directly. If you're on Shopify, Shopify handles both via Agentic Storefronts (free, automatic for eligible US merchants as of March 2026). If you're on a different platform, your obligations are about the data that feeds these protocols.
What AI shopping agents actually need from you.
Here's the inversion most ecommerce teams miss: AI shopping agents do not read your blog posts to decide whether to recommend you. They read your product feed, your structured product data, your pricing accuracy, your inventory status, and your reviews. Editorial content does not move agentic commerce visibility. Operational rigour does.
- Complete product feed — title (max 150 chars), description (max 5,000), price with ISO 4217 currency code, availability, images, GTIN where available, full attribute set including colour, size, material, intended use
- Real-time price and stock sync — agents cross-reference your storefront price against your feed; mismatches mean your product gets skipped or flagged
- Product schema (Product + Offer + AggregateRating + Review) across every product page
- Crawlable robots.txt — OAI-SearchBot and Googlebot must be allowed; this is the number-one reason brands are invisible in ChatGPT shopping results
- Live policy pages — return, shipping, warranty, support. Missing or gated policies reduce your trust score in agent selection
- Review velocity — average rating, total count, and recency all factor
The agentic commerce numbers.
YoY growth in AI-referred traffic to US retail sites on Black Friday 2025.
Adobe via MetaRouter
expected AI platform retail spending in 2026 — nearly 4× the 2025 figure.
eMarketer · Dec 2025
higher purchase likelihood for shoppers arriving from AI services vs traditional channels.
eMarketer
McKinsey projects $3–5 trillion globally in agentic commerce by 2030. Morgan Stanley predicts nearly half of online shoppers will use AI shopping agents by 2030, accounting for ~25% of their spending. Even discounted heavily, the channel is going to be enormous and the decisions made about feed quality and protocol participation in 2026 will compound for years.
The eCommerce 2026 priority list.
If you sell online and you only do six things in the next 90 days, do these:
- Product data audit. Pull your full feed. Score every product on completeness against the 12-attribute baseline (title, description, price, availability, images, GTIN, brand, condition, colour, size, material, category). Anything below 90% completion is invisible to agents.
- Schema audit on top 100 SKUs. Product, Offer, AggregateRating, Review. Validate all four.
- Bot access audit. Confirm OAI-SearchBot, Googlebot, ChatGPT-User, PerplexityBot, ClaudeBot are not blocked. Check Cloudflare and WAF rules, not just robots.txt.
- Policy pages live and indexable. No login walls, no forms, no PDFs. Plain HTML, schema-marked where applicable.
- If on Shopify: enable Agentic Storefronts. If on another platform: implement UCP via your stack's supported integration.
- Track AI shopping referrals in GA4. Build custom segments for traffic from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com. Watch the trend line monthly.
What about content?
Content still matters in agentic commerce — but not the way it mattered in classical SEO. The content that earns AI shopping citations is structured comparison content. Agents read these to decide which products to surface in their comparison set.
- Comparison pages for every meaningful product cluster — your products vs competitors, ranked honestly
- "Best X for Y" listicles owned and authored by you
- Buying guides shaped as Q&A — "what should I look for in X," "how do I choose between Y and Z"
- Specification pages with schema — every spec, every variant, every edge case
Where this is going.
- Universal feeds. The current fragmentation — separate feeds for Google Merchant Center, Meta Catalog, Amazon, ChatGPT, etc. — will consolidate into 2–3 dominant feed standards. Shopify Catalog is already a leading candidate.
- Agent-to-agent negotiation. Forrester predicts 20% of B2B sellers will face agent-led quote negotiations by end of 2026. Your sales process and pricing pages need to be readable by buyer-side agents.
- Loyalty and identity layers. Google's UCP added Identity Linking in March 2026, allowing loyalty programmes to integrate. Brands with strong loyalty data will get preferential placement.
Custom GPTs & AI agents that earn their keep.
The 2025 edition pitched custom GPTs as essential for every business. We were right that the category matters. We were wrong that every business should have one.
Four jobs custom AI agents are actually good at:
- Internal expert assistance — captures institutional knowledge for junior staff
- Sales qualification and discovery — runs real qualification conversations, books meetings
- Customer support deflection — trained on product docs, FAQs, resolved support tickets — handles 60–80% of L1 questions
- Specialist content production — voice-trained to produce first drafts
Note: Notice what's not on this list: 'general assistant for our customers.' Those don't work. Build agents around problems, not personas.
The 2026 build stack: OpenAI Custom GPTs, Claude Projects, Microsoft Copilot Studio, Retell AI/Vapi/Bland AI, n8n + Claude/GPT API, GoHighLevel/HubSpot/Salesforce.
Five-step build that actually works:
- Pick one job
- Write the system prompt before you build
- Train on real artefacts, not summaries
- Test with your hardest cases
- Define escalation triggers explicitly
Measuring AI visibility — the new analytics stack.
If you can't measure it, you can't budget for it.
Three categories:
1. Citation tracking — what AI is saying about you:
- Profound — enterprise, $499/month+
- Otterly.AI — Lite $29/month, Standard $189/month, Pro $989/month
- AthenaHQ, Scrunch, AIclicks, Rocketito
2. Bot analytics — what AI is reading from you:
- Ahrefs Bot Analytics (free in beta)
- Cloudflare AI Audit
- Server log analysis
3. Referral tracking — what AI is sending to you:
- GA4 segments for: chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai
Visibility Metrics
- Citation rate
- Share of voice
- Sentiment
Performance Metrics
- AI referral traffic
- AI referral conversion rate
- Branded search lift
The prompt set that drives the dashboard.
Four prompt categories: Branded (5–10), Category (10–20), Comparison (5–10), Problem (10–20). "Run these monthly."
Action plan by vertical.
The principles apply to every category. The tactics don't.
1. Legal Services
"Solo and small firms. The most underserved AI visibility opportunity right now."
30-day: Attorney schema, GBP, Bing Places, FAQ pages. 90-day: Wikidata, named methodology, case study library. Watch: Avvo, Justia, FindLaw.
2. Medical & Cosmetic Clinics
"Healthcare AI Overviews appear on 88% of relevant queries."
30-day: MedicalBusiness schema, Healthgrades/RealSelf. 90-day: named patient-experience framework. Watch: AHPRA compliance.
3. Real Estate
"Agents who appear as named expert for a specific suburb dominate consideration."
30-day: RealEstateAgent schema, suburb-level guide pages, REA and Domain profiles. 90-day: named appraisal methodology, monthly market commentary. Watch: suburb-specific depth.
4. Trades & Home Services
"The Local Service Ads layer plus AI search creates dual-engine opportunity."
30-day: specific business-type schema, Google Local Service Ads, photo library. 90-day: named service framework, suburb-level service pages, pricing pages. Watch: review velocity across Hipages, Oneflare, Houzz.
5. eCommerce
"The agentic commerce shift is the single biggest 2026 change."
30-day: product feed audit, schema on every PDP, robots.txt audit. 90-day: comparison page library, buying-guide content, review velocity. Watch: price and inventory accuracy.
6. B2B SaaS & Professional Services
"B2B Tech AI Overviews jumped from 36% to 82% of queries in twelve months."
30-day: Organization + Person schema, G2/Capterra/TrustRadius profiles, comparison pages. 90-day: named methodology, original research report, podcast appearances. Watch: LinkedIn is most-cited domain for professional queries.
The cross-vertical truths.
Five priorities: technical foundation, sharpen positioning, name and document methodology, build distribution, measure consistently.
"The 2026 winners aren't the businesses with the cleverest AI strategies. They're the businesses with the most disciplined application of unsexy fundamentals."
The next training cycle is always coming.
If you read the 2025 edition, you might remember the closing line: 'the next training cycle is coming. Your future AI visibility depends on the authority you build today.' That's still true. It's just more urgent.
The pace of change has accelerated, not slowed. AI Mode shipped publicly. Agentic commerce protocols launched. Citation tracking became a real category. Local AI visibility became a viable channel for businesses that couldn't afford traditional national SEO. And every model — GPT-5, Claude Opus 4.7, Gemini 2.5, the open-weight models gaining ground — gets retrained on a regular cadence that bakes the current state of the web into the next generation's view of who matters.
If you are not visible across the corroborating sources during a training window, you are not in that model's worldview when it ships. There is a real cost to being late, and it compounds. The window to establish your authority for the 2027 models is open right now.
The guide has eleven chapters. The core discipline is one idea: be the entity AI models can find, recognise, and trust. Everything else — the schemas, the frameworks, the robots.txt configurations, the prompt sets, the distribution plays — is evidence. Evidence that you're real. Evidence that you're relevant. Evidence that you're the consensus in your category.
Building that evidence layer doesn't require a large team, a large budget, or cutting-edge tooling. It requires clarity about what you do and who you do it for, consistency in how you express that across every channel the model reads, and patience for the compounding to work.
Three months of consistent execution typically produces first AI citations. Six months produces measurable referral traffic. Twelve months produces category leadership for brands that commit.
Six things to do before the end of the quarter.
- Audit your robots.txt for AI retrieval bot access — single biggest 5-minute win
- Validate your schema across the top ten pages on your site
- Sharpen your positioning statement to a specific audience, outcome, and methodology
- Stand up a citation tracking tool — Otterly, Profound, or equivalent — and run your first prompt set
- Confirm your Bing Webmaster Tools and Bing Places listings are claimed and clean
- Schedule the 90-day plan with someone owning each piece — not collectively, individually
None of this is novel. None of it is hard. The hard part is doing it consistently while every other quarter brings a new shiny tactic to chase. The advice that ages well in this market is unfashionable advice: get the fundamentals right, narrow your positioning, build a distinctive methodology, distribute it widely, measure it honestly, refresh it consistently.
Do that and the AI visibility takes care of itself.
Don't, and the next training cycle ships without you.
AEO questions, answered.
Everything business owners ask us when they first encounter AI search.
About the author
Michael started in digital in 1999 and founded his first agency in 2001, in the early days of the Australian internet industry. Top4 Technology grew out of that foundation into a digital marketing and technology platform now managing 200,000+ business profiles for 65,000+ paying subscribers across Australia, Indonesia, and Asia-Pacific.
Top4 Technology is a wholly-owned subsidiary of White Pearl Technology Group AB, a Swedish-listed digital transformation group on Nasdaq First North (ticker: WPTG). The Top4 team operates across LLM-led SEO, multi-location and government digital, enterprise CMS delivery (Kentico), AI automation, and CRM & marketing automation. The group is a Google Premier Partner, Meta Business Partner, Kentico Solution Partner, and SiteMinder Reseller.
He writes and speaks on AI search visibility, Generative Engine Optimisation, and the practical application of emerging marketing technology for businesses of every size. This guide is the second edition of his flagship annual AI visibility report. He is based in Sydney.