AIO SEO case study Australia: 18 months, no ads, three AI models
AIO case study · 18 months · Australia

The store that AI learned to name.

They came to me eighteen months ago with a website barely a year old, almost no footprint in Google and none at all in the AI answers. No paid ads. One PR channel. A category owned by national brands they were never going to outspend. Today they are named, by name, in AI answers in the corner of the market they chose, and the sales are continuous rather than seasonal. Here is the whole eighteen months, including the parts that were constraints rather than clever ideas.

Start at month zero What this proves, and what it does not
Client
anonymous
The shape of it
01 · At the start
1 year old
A young site, thin footprint, flat architecture.
02 · Paid advertising
$0
None then, none now. PR was the only outside channel.
03 · Daily clicks now
About 3.5x
Against the first month, indexed. Volumes stay with the client.
18
Months of work, still running.
Engagement began early 2025
0
Dollars of paid advertising, at any point.
Organic and PR only
3
AI assistants that named the brand in an incognito pulse check.
Documented, table below
2
Anchor categories chosen, out of everything they sell.
The whole strategy sits on these
01 · Month zero

What does a site look like before any of this works?

A premium ecommerce brand, one year old, with real inventory and almost nothing to show for it in search. They were not being punished by Google; they were simply not being read. The architecture was flat, which is the usual state of a young store: everything hanging off one level, no page telling a search engine or a model what this business is actually expert in.

In AI answers they did not exist at all. That part is worth sitting with, because it is not the same problem as ranking badly. A model that has never been given a reason to associate a brand with a category will not name that brand, no matter how good the products are. It is not a ranking, it is a recognition problem.

The actual diagnosis

They had the inventory. What they did not have was a structure that let a model verify expertise in anything specific. Nothing on the site said "this is what we are the specialist in", in a way a machine could read and cross-check.

02 · The constraints

What do you do when you cannot outspend anybody?

This is the part most case studies leave out, and it is the part that decided the strategy. There were no paid ads, and there still aren't. The only outside channel was PR, which they were already doing well. And the category they sell into is held by national brands with budgets that make a direct fight pointless.

So the plan was never "compete everywhere". It was pick ground we can actually hold. Two questions did the choosing. What are they genuinely better at than the big brands? And which searches do the big brands answer badly, because answering them properly would take specialist knowledge they do not have?

01

Major cities first

Not a national land grab. We worked city by city, starting where their customers already were, so each new page had a real audience behind it rather than a keyword.

02

Their strengths, not the category's

We built on the two things they knew more about than anyone bidding against them. Depth in a narrow place beats breadth you cannot defend.

03

PR left alone, and used

Their press work carried on untouched. What changed is that the site finally gave those mentions somewhere solid to point, so the coverage started doing structural work as well as brand work.

We could not buy attention. So we had to be the best answer to the questions the big brands answer badly.

03 · The build

What is an Entity Mesh, in practice rather than in theory?

An Entity Mesh is a way of building a site so that the whole thing argues for the same expertise. Individual pages stop being separate bets and start being evidence for one claim. Four pieces of work, in the order we did them.

01. Two anchor nodes, chosen on margin and on strength.

Out of everything they sell, we picked two sub-categories: high margin, and the two places they genuinely knew more than the national brands. Both were rewritten, technically and in the words a customer reads, so that specialist status was unambiguous rather than implied. Everything else on the site was left alone at this stage. That restraint is the strategy, not a shortcut.

02. Depth, without a single new product.

A collection page that is too thin reads as a lack of authority, to Google and to a model. We audited every gap in the two anchor categories, then ran a product-tagging overhaul so that the depth already sitting in the warehouse actually appeared on the right pages. Nothing was invented. Not one new SKU was added. The inventory was always there; it just was not visible as depth.

03. The mesh itself.

City and regional node pages were interlinked with their parent categories, deliberately, so that each page had a clear place in a structure rather than floating. This is what turns a catalogue into something a model can map. A crawler arriving at any page can work out what the business is expert in, where it operates, and how the parts relate.

04. The question layer, running in parallel.

At the same time, we started answering the questions people really type into ChatGPT. Not keyword variants; questions with the hesitation still in them, the ones a buyer asks a friend before spending real money. Each one was answered on the page that deserved to own it, in language a model can lift cleanly. This is the piece that turned a well-structured site into a quotable one.

Why the two halves had to run together

The structure gives a model something to trust. The questions give it something to quote. Structure with no answers gets crawled and ignored; answers with no structure get crawled and attributed to somebody else. Neither half works alone, which is why I do not sell them as separate services.

04 · What happened in Google

Did the organic side actually move?

Yes, and slowly at first, which is the part nobody puts in a case study. The first months are structural work with very little to show. The curve turns once the structure is in place and the question pages start being indexed against real searches.

Daily clicks from organic search

Indexed. First month = 100
First month of work100
Around month sixAbout 200
Around month tenAbout 250
Around month fourteenAbout 300
Month eighteen, August 2026About 350
Clicks about 3.5x the starting month · impressions about 2x
Read off the daily curves in the client's Search Console on 31 August 2026, and indexed. The absolute volumes stay with the client. Approximate by construction: these are readings from a chart, not exported totals.

The middle of that curve is where the compounding shows. Every new city node made the anchor categories stronger, and every strengthened category made the next city node land faster. That is the difference between an entity strategy and publishing more articles; the second one starts again each time.

The four-week milestones, from the earlier phase.

Published on this page since the first version, and left here because they date the turn: rolling 28-day clicks passed 700 on 14 October 2025, 800 on 7 November, 900 on 20 November, and 1,000 on 2 December 2025. Roughly a hundred more clicks every fortnight, through the quarter when the mesh went live.

05 · What happened in the AI answers

Do the models name them now?

After deployment I ran a pulse check: the same high-intent buying questions, incognito, standardised, across four assistants. Three of them named the brand without being prompted with it. Here is what each one said, in my own words rather than theirs.

Incognito pulse check, high-intent buying prompts
What each assistant returned about the brand
AssistantWhat it did with the brand
Gemini (Google) Cited the brand explicitly as a specialist curator for the target sub-regions.
GPT-5 (OpenAI) Named the premium range and picked up specific value propositions, including shipping and discounts.
Grok 4 (xAI) Mapped the inventory diversity accurately across the newly optimised regions.
Standardised prompts, incognito sessions, run by me after deployment. Model answers vary between runs; this is a snapshot, not a ranking.

The detail I care about most is not that they mentioned the brand. It is what they said. Specialist curator. Regional coverage. Inventory diversity. Those are the exact claims the mesh was built to make. The models did not invent a flattering description; they repeated the structure back to me.

And then the AI surface became its own line in Search Console.

From the middle of May 2026, AI-driven impressions stopped being an occasional blip and became a daily line with a floor under it. Late May produced a spike near three times the baseline, which settled rather than collapsed; through June, July and August the line held roughly double where it started, with no paid support underneath it.

Daily impressions from the AI surface

Indexed. First days of the window = 100
18 May 2026, the first days100
Late May, the spikeAbout 450
June to July, the new floorAbout 200
26 August 2026About 190
Read off the daily AI impressions curve in the client's Search Console, window 18 May to 26 August 2026, and indexed. Absolute volumes stay with the client.

A spike is a moment. A floor is a position. The reason the floor exists is that the answer being quoted sits on a page that was built to be quoted, so the next model that looks finds the same thing.

06 · Being straight about the evidence

What this proves, and what it does not.

What it shows: an account with no paid support, working a narrow category, grew organic clicks to roughly three and a half times its starting month over eighteen months, and is now named by three assistants for the exact expertise the structure was built to claim.

What it does not show: that every gain came from this work. Their PR carried on the whole time, the market moved, and Google changed several things along the way. I cannot hand you a clean split between those, and anybody who offers you one is guessing. What I can say is that the claims the models repeat are the claims the mesh was designed to make, which is a tighter piece of evidence than a traffic line on its own.

Talk about your site
07 · If this sounds like you

Would the same approach work on your site?

The conditions that made this work are specific, and they are worth checking against your own situation before you spend anything.

1

You already hold the inventory or the expertise, and it is not visible.

This is a structure problem, and structure problems are the cheap kind. Nothing here required new stock.
2

There is a corner of your category where you know more than the big brands.

If there is not one yet, that is the first piece of work, and it is a business question before it is an SEO one.
3

You can wait through a slow first quarter.

The curve above is flat before it turns. If the business needs revenue inside eight weeks, this is not the right first move, and I will tell you so.
4

Somebody can answer the real questions properly.

The question layer only works when the answers are specific. Generic answers get crawled and skipped, by Google and by the models.
08 · Questions about this case study

The things people ask me about it.

What is an Entity Mesh in SEO?

An Entity Mesh is a site architecture strategy that connects related content nodes, meaning categories, sub-categories and regional pages, through deliberate internal linking and consistent entity signals. Rather than optimising individual pages in isolation, it builds a structured web of authority that search engines and AI models can map and trust.

The practical test is simple. Land a crawler on any page and ask whether it could work out what this business is expert in, where it operates, and how the pieces relate. If it could not, there is no mesh yet.

How is AIO different from traditional SEO?

Traditional SEO optimises for keyword rankings in blue-link results. AIO optimises for citation by AI models such as Gemini, ChatGPT, Grok and Perplexity, which generate a direct answer instead of a list of links. The goal shifts from ranking to being recommended.

The work overlaps heavily, which is the point of this case study. The same structural signals did both jobs.

Can a site rank in Google and be cited by AI at the same time?

Yes, and this account is the evidence. The structural signals that earn AI citations, meaning topical depth, entity clarity and internal authority architecture, are the same ones that strengthen organic rankings. A well-built entity strategy does not force a choice between the two.

Can this work without a paid advertising budget?

This client has never run paid ads, at any point in the eighteen months. The only outside channel was PR, which they were already doing. Working without ads changes the strategy rather than ruling it out: you cannot buy your way into a broad category, so you pick a narrow one you can actually own and you build depth there first.

Does PR help with AI visibility?

It helped here, and the reason is worth being precise about. The press coverage existed before the site work and had not moved the needle in search. What changed is that the coverage finally had somewhere solid to point, so the mentions started reinforcing a claim the site was already making about itself. PR on its own is brand work; PR pointing at a structured entity does structural work as well.

How long does it take to see results?

Measurable click growth appeared within six to eight weeks of implementation on this account, with model citations confirmed after that. The bigger picture is slower and more useful: the curve on this page is flat for the first stretch, turns around month six, and keeps compounding through month eighteen. Timelines vary with site size, crawl frequency and existing authority.

How many pages did this take?

Fewer than most people expect. Two anchor categories, their collection pages rebuilt, a set of city and regional nodes added over time, and a question layer answered on existing pages wherever one already deserved to own the answer. The gains came from the relationships between pages, not from volume.

Does this only work for ecommerce?

The inventory depth step is specific to product businesses. Everything else, meaning anchor nodes, structure, regional coverage and the question layer, applies to a service business as well. The equivalent of inventory depth there is proof of work: named projects, locations served, and specifics rather than adjectives.

Do you share client names or niches publicly?

No. Client confidentiality is not negotiable, and in a narrow niche the niche itself identifies the client, which is why the categories are described rather than named and why no absolute traffic volumes appear on this page. Frameworks and the underlying receipts are shared in an initial conversation or at onboarding.

09 · Who ran this

I take a small number of these at a time.

I'm Roxane Pinault, an AIO and SEO specialist working with Australian businesses from the Central Coast of New South Wales. My work is entity architecture and modern SEO that compounds across Google and the AI assistants, which in practice means turning what a business already owns into something a model can verify and quote. I hold to a maximum of five clients at a time, because this kind of work needs attention rather than volume.

What you get

The same method as this account: anchor nodes, structure, and the questions your buyers actually ask, in an order that suits your business.

What you will not get

A guarantee, a client list, or a clean causal story where the data does not support one.

Roxane Pinault · AIO and SEO, Australia
Method and disclosure
Client anonymous by agreement; category described rather than named · Traffic figures: read off the daily curves in the client's Google Search Console on 31 August 2026 and indexed to the first month of the window, so no absolute volumes are published. Readings from a chart are approximate by construction · Four-week milestones (700 clicks on 14 October 2025, 800 on 7 November, 900 on 20 November, 1,000 on 2 December 2025) were published in the first version of this page and are retained · AI surface window: 18 May to 26 August 2026 · Model pulse check: standardised high-intent prompts, incognito sessions, run by me after deployment across four assistants; three named the brand. Model output varies between runs · No paid advertising ran on this account at any point.
Written by Roxane Pinault, AIO and SEO specialist, Australia. Get in touch

Ready to become the brand the models name?

If your site has the inventory but not the visibility, the gap is almost always structural rather than a content-volume problem. Send me the one question your customers would ask before buying, and I'll run it across four assistants and tell you who is being named instead of you. No charge for that, and no pitch attached. Where you go after that is up to you.

Let's talk about your site Connect on LinkedIn