GEO vs SEO: what actually changes for retail brands—and what stays the same

Google's own documentation now answers this question directly, and the answer is blunt: optimising for generative AI search is optimising for the search experience, and thus still SEO. That guidance, updated in July 2026, also says llms.txt files are ignored by Google Search, that you don't need to chunk content into fragments, and that structured data isn't required for generative AI features at all. Most of what circulates as GEO — generative engine optimisation, the practice of trying to get cited inside AI-generated answers — is, on Google's account, either already SEO or already useless.

Retail is the one vertical where that framing breaks down, and the reason is measurable. This piece walks three real page types — a product page, a category page and a store page — and shows side by side what each discipline rewards on each one, and where the retail evidence diverges from the general advice.

‍

What each discipline is optimising for—and why Google says they're the same thing

Two mechanisms do most of the work in AI search, and both are described in Google's guidance. The first is retrieval-augmented generation, or grounding — the model pulls live pages from the search index and generates an answer from them, rather than from memory. The second is query fan-out: the model silently generates a set of related sub-queries and fetches results for those too. Ask "best waterproof hiking boots under $200" and the system may also run "hiking boot waterproof ratings" and half a dozen others you never see and cannot track.

That second mechanism is why ranking for your target keyword is no longer the whole game. In Ahrefs' analysis of 863,000 keyword SERPs and roughly 4 million AI Overview URLs, only 37.9% of cited pages also appeared in the first 10 results for the same query. Seven months earlier, the same team measured that figure at 76.1% across 1.9 million citations.

The evidence here is mixed. A separate analysis by seoClarity across 362,000 keywords found 94% of AI Overviews cite at least one URL from the top 20 organic results, which points the other way. Both can be true: ranking gets a page into the candidate pool, and something else decides which candidates get quoted. Treat rank as necessary but not sufficient.

‍

The retail exception nobody puts in the GEO explainers

Here is the finding that should change how you read every generic GEO article. Across a 16-month tracking study of nine industries, e-commerce was the only vertical where citation-to-ranking overlap stayed flat — moving 0.6 percentage points while education moved 53.2 — and AI Overview coverage of e-commerce queries actually fell 7.6 points. This figure comes from BrightEdge, who sell SEO software, and the sample size behind the industry splits is not published, so hold it loosely. But the direction is consistent with what Google does elsewhere: transactional queries are being routed to shopping surfaces rather than answered with a synthesised paragraph and a list of citations.

The implication is uncomfortable for anyone selling a single GEO playbook. Your blog content and buying guides are competing in an AI-answer environment. Your commerce pages, in large part, are not — they are competing in a feed-driven product environment that runs on different inputs entirely.

‍

Example one: a product page for a single SKU

What SEO rewards here

The familiar list, and it still works: a title tag that matches how people search for the item, unique on-page copy rather than the manufacturer's boilerplate, review content, internal links from the parent category, clean canonicals across colour and size variants, and a page that loads fast on a phone. The goal is a top-10 position for "[brand] [product name]" and its close variants.

What GEO rewards here

Something almost entirely different: the feed. Google's guidance names Merchant Center feeds and Google Business Profiles as the route to product and local visibility in AI responses, and its structured data documentation is explicit that providing both on-page markup and a Merchant Center feed maximises eligibility and helps Google verify your data. Note the tension with the mythbusting section of the same guidance — structured data isn't required for generative AI search in general, but for products the feed is the mechanism, and on-page Product markup is what corroborates it.

So the levers are: attribute completeness, valid GTINs, availability and price that match between feed and page, and variant attributes populated properly. What matters is that the machine-readable record is complete and internally consistent, not that the prose is persuasive. And the arithmetic is unforgiving at scale — on a 40,000-SKU catalogue, 88% attribute completeness leaves roughly 4,800 products carrying gaps, and every one of those is a product an assistant can't confidently recommend when a shopper asks a question your data can't answer.

Where they meet

Both want the same underlying thing: accurate, complete, non-duplicated product information. SEO wants it rendered for a human; GEO wants it structured for a machine. The part that doesn't overlap is nearly all feed hygiene — which sits with your merchandising or data team, not your content team.

‍

Example two: a category page

What SEO rewards here

Ranking for the head term — "women's running shoes", "outdoor furniture" — plus faceted navigation that doesn't spawn thousands of thin indexable URLs, useful introductory copy, and internal linking that pushes authority down to the products that convert.

What GEO rewards here

Surviving fan-out. A category page's AI-search competitor isn't another retailer's category page; it's the model's own synthesis, assembled from a dozen sub-queries and whatever editorial, forum and video content answered them best. Comparison intent is where this bites hardest: Seer Interactive found that 95.4% of "X vs Y" informational queries in its study of 49,353 distinct queries returned an AI Overview. If your category and comparison pages are a traffic pillar, that traffic is almost entirely inside an AI-affected environment now.

There is one finding that tells you where to aim first. The peer-reviewed Marketing Science analysis of 973 e-commerce websites found that LLM referral performance was stronger in complex product categories. Considered purchases — appliances, technical apparel, anything with specifications a shopper needs explained — are where this effort pays back. Impulse and replenishment lines are not, and rebuilding those category pages for AI retrieval is very likely a poor use of your quarter.

‍

Example three: a store page

What SEO rewards here

Local pack presence: a complete Google Business Profile, consistent name, address and phone details across directories, LocalBusiness markup with accurate geo data, genuine store-specific content rather than a template with the suburb swapped out, and current trading hours.

What GEO rewards here

Almost exactly the same things, with more weight on the profile than the page. This is the page type where GEO and SEO have converged rather than diverged, because the structured local record Google already holds is the raw material for the generated answer. Google's guidance points to managing your business details on Google Search and flags Business Agent, a conversational surface that lets customers chat with your brand directly in results.

What has changed is exposure, not technique. Seer's data shows 76.9% of informational "near me" queries returning an AI Overview — Google is layering generated answers on top of local results, not replacing them. Early results suggest this compresses the click, but it has not been widely replicated for Australian retail specifically.

‍

What stays the same on all three pages

It is well established, and stated plainly in Google's documentation, that a page must be indexed and eligible to appear with a snippet before it can show in a generative AI feature — and the site must additionally be included in Search generative AI features in Search Console. Every technical prerequisite you already had is still a hard gate. There is no GEO route around a crawl problem.

Thin content still loses, too. Google's guidance singles out commodity content — the generic listicle anyone could have written — as least likely to be surfaced, and warns that spinning up a page per fan-out variation breaches its scaled content abuse policy. The first-party operational data your competitors don't have is worth more than another optimised paragraph.

And this is still a small channel in absolute terms. The Marketing Science study found ChatGPT accounted for under 0.2% of e-commerce traffic. On 500,000 monthly sessions that's around 1,000 visits — at a 2% conversion rate, roughly 20 orders a month. Australian behaviour is moving faster than that share suggests: the AI Brandscape 2026 report found 38% of Australians already use AI tools to complement or replace search, and 39% use AI to make buying decisions. The published summary doesn't state its sample size, so treat those figures as indicative.

‍

Where this evidence is weak

• The citation-overlap figures are genuinely disputed. Top-10 overlap estimates range from roughly 17% to 38% depending on whose parser and dataset you use, and the studies are not directly comparable.
• Conversion quality is contested. The peer-reviewed Marketing Science work found LLM referral conversion below every traditional channel except paid social; Adobe, which sells analytics software, has reported AI-referred retail traffic converting substantially better. The attribution methods differ, and both may be describing their own data accurately.
• We could not find reliable Australian data on which retail page types are being retrieved by AI engines. Nearly every study cited here is US-measured, and Australian catalogue sizes, market concentration and inverted seasonality differ enough that transfer is not safe.
• Attribution is unsolved. GA4 has no native AI channel, so AI-referred sessions land in general referral traffic unless you build the segmentation yourself.
• The commerce protocol layer is unsettled, with competing standards from Google and OpenAI. Anything built against a single protocol today carries rework risk.

‍

So where should you actually spend?

Rank the work by page type, not by discipline. Start with feed and catalogue data. It is the highest-certainty investment here, because it pays into paid shopping, organic shopping surfaces and AI product discovery at once, and because Google's own documentation names it as the mechanism. Second, rebuild comparison and buying-guide content for considered-purchase categories only — that is where the peer-reviewed evidence locates the upside. Third, tighten local records, which is cheap and mostly work you owe your stores anyway. Ignore, for now, llms.txt files, content chunking and mention-seeking campaigns: Google states these do nothing for its surfaces.

The constraint is rarely knowing what to do. In sponsored research published by Incisiv with AWS, 89% of retailers reported failing to scale innovations across the organisation — a vendor-published figure, but one worth remembering before you commission a fourth pilot instead of fixing the catalogue.

‍

The bottom line

GEO and SEO are not competing disciplines for retail brands. They are the same discipline applied to two different assets. For content pages the work has genuinely changed — fan-out means topical depth now beats single-keyword ranking. For commerce pages the work has moved rather than changed: out of the copy deck and into the feed. If you do one thing this quarter, audit attribute completeness across your product catalogue and check that your feed and your on-page markup agree with each other.

Which of your product, category and store pages are actually being retrieved by AI engines today is a measurable question — and one you should answer with your own data rather than US benchmarks. Talk to us about auditing your catalogue data and setting an AI retrieval baseline.