The ultimate guide used to be the dominant SEO play. Pick a topic, write 4,000 words, cover every adjacent question, watch it rank, watch readers stay and watch the enquiries trickle in. That worked because humans were doing the searching and the scanning of your content.
Two pieces of research published in April 2026 say that playbook has shifted, and AI search is the reason. The rest of this piece walks through what changed, what the data shows, a Perth example we saw last month, and how to build an AI search content strategy that wins both citations and clicks.
Content Overview
The Funnel That Used to Work
Three years ago, a business owner wanting to learn about Google Ads would type “Google Ads guide” into Google, land on a long blog post, and either keep reading or bounce. If the guide was any good, they’d absorb the basics, start trusting the agency that wrote it, and either have a crack themselves or book a call.
The whole funnel rested on the assumption that the human reads the page. Long-form content was the rational move when every minute on page built trust and every internal link could nudge the reader toward a contact form.
A few of these older, broader posts on our site, like our piece on TikTok Marketing for Australian Businesses still pull steady readers. But they don’t cut the mustard on their own anymore.
People Don’t Browse Anymore. They Ask.
Someone who wants the overview increasingly opens ChatGPT, Claude, Perplexity, or Google AI Mode and asks the question conversationally. They get a synthesised answer in a couple of paragraphs with two or three cited links underneath. They don’t scroll a 4,000-word guide.
The job of a blog post in 2026 is partly to be read by a human and partly to be the thing an AI tool grabs and cites when it answers for them. Different jobs, different content shapes.
Behind the Curtain: How an AI Tool Researches the Web
When someone asks ChatGPT a broad question like “how do I set up Google Ads for my small business in Perth”, it doesn’t only summarise its training data. It runs its own internet searches in the background to refresh and verify before writing the answer.
Kevin Indig and AirOps mapped 16,851 of these queries across 353,799 pages in their Fan-Out Effect study. They found that 88.6% of the time, ChatGPT issues exactly two “fan-out” sub-queries before answering, with more complex comparative or review questions sometimes generating four or more.
Those sub-queries are not phrased the way humans search. They’re far more specific. A human types “Google Ads for beginners”. ChatGPT, given the same question, might silently fan out into “Google Ads conversion tracking GA4 setup steps” and “Google Ads bidding strategy for small budget B2B Australia”. It grabs the top results for each, parses the pages, and decides which to cite back to the human.
So the question is no longer “is my page the best answer to the broad query the human typed?” It’s “is my page the best answer to one of the very specific sub-queries the AI tool fanned out into?”
Google’s Winners Look Different Now
Cyrus Shepard’s Zyppy Signal analysis looked at 400-plus websites that won or lost Google traffic year-on-year. He found five characteristics that predicted whether a site was winning:
- Offering a product or service
- Letting the user complete a task on the page
- Owning proprietary assets
- Holding a tight topical focus
- Having a strong brand measured by traffic from branded searches.
Sites with zero of the five had a 13.5% win rate; sites with three sat at 30.7%; sites with four jumped to 68.1%; sites with all five hit 69.7%.
ChatGPT Citations Reward Focus, Not Breadth
The AirOps and Indig data points in the same direction from the AI side.
First, retrieval rank dominates. A page in position 1 of ChatGPT’s web search results has a 58.4% citation rate. By position 10, that drops to 14.2%. Even pages with strong heading-level relevance get cited only 21.5% of the time when they sit at rank 11 or worse.
Second, when you hold query match constant, pages covering 26 to 50% of ChatGPT’s fan-out subtopics outperform pages covering 100% of them. AirOps’ own headline from the data: “a page that nails one question outperforms a page that adequately addresses five.”
One bonus finding: domain authority and backlinks did not positively correlate with ChatGPT citation in this dataset, and were slightly inversely correlated. Page-level relevance now matters more than the domain-level signals SEO has historically optimised for (only for AI SEO).
A Real Example: The Specific Question That Converted
We saw this play out in our own analytics recently. A user typed a very specific Google Business Profile question into a chatbot. The chatbot fanned out into sub-queries and pulled in our Google Business Profile optimisation guide as a cited source. The user clicked straight through to the section that answered their question, and they converted.
One focused page nailing one specific sub-query, one citation, one click, one customer. Exactly the pattern AirOps’ data would predict.
A broad “how to set up Google Business Profile” overview wouldn’t have shown up there. The broad guide might rank for the broad query, but the moment the chatbot fans out into a specific sub-query, the page that wins the citation is the one whose headings closely match the sub-query, not the one mentioning the topic in passing inside a 5,000-word overview.
Building an AI Search Content Strategy
The lazy take is “stop writing ultimate guides,” and that’s wrong. The answer has four parts that run in parallel.
Keep Your Broad Guides
Broad guides still serve the human who wants the overview, the brand-search reader, and remarketing audiences. They pull broad-keyword traffic, and we have a few on our own site doing exactly that.
What we’d retire is the “ultimate guide” written purely as an SEO swing for the fences with no real point of view inside it. Those weren’t winning citations even before AI search took off, and they were already doing very little for the business once they ranked.
Write Focused, Single-Question Pages Alongside Them
This is the new bucket. One page covers one specific question with one direct answer. The AirOps data points to a sweet spot of 500 to 2,000 words, four to ten subheadings, college-level readability, and FAQPage or BreadcrumbList JSON-LD schema. FAQPage in particular hit a 45.6% citation rate in their dataset, well above average.
The headings should match how the question gets asked in the wild. If you sell physiotherapy in Subiaco, “Subiaco physio” is a service-page keyword. “Can a physio help with sciatica that flares up after running?” is a focused-page question. Both belong on your site, and they aren’t the same page.
Build the Brand So People Type Your Name
The strongest signal from Cyrus Shepard’s data was destination demand. People typing your business name into Google or asking ChatGPT for you by name is the most defensible position you can hold, because nobody can disintermediate you when you are the search itself.
For an Australian SME, that means consistent presence in your category, content that sounds like you and not like every other agency, and the slow, unsexy work of being the name people associate with what you do. More on this in our piece on why branding matters in generative search engines.
Ranking Still Matters More Than Anything
Rank position still matters more than anything else, in both Google and inside the AI tools’ own retrieval. The AirOps data showed a page at rank 1 gets cited four times more often than one at rank 10. A perfectly written page at rank 11+ gets cited 21.5% of the time, while a mediocre page at rank 1 gets cited 56%.
“Rank higher” was always the answer, and it still is. What’s changed is what you’re trying to rank for and who’s reading once you do. The broad guide ranks to be read by a human. The focused page ranks to be cited by an AI tool quoting you back to a human. Same work, different jobs.
For more on how SEO and generative search fit together, see our take on GEO vs SEO. Short version: you need both.
If you’re commissioning content right now, ordering one 4,000-word ultimate guide for the quarter is the wrong move. Splitting that brief into a shorter overview plus three or four focused, single-question pages gives you both the broad-keyword coverage you’ve always had and the focused pages that win AI citations.
If you’re not sure whether your current content is set up to win in this new shape of search, our SEO and generative engine optimisation team can audit what you’ve got and map where the focused pages need to sit.
Rob Sharif
Technical SEO Lead
