AI Visibility

GEO vs SEO for ecommerce: what changes, what doesn't, and what to stop doing

Generative engine optimisation is not SEO with a new name, and it is not a replacement for it. Here is precisely where the two diverge, which SEO habits become counterproductive, and how to run both without doing the work twice.

August 4, 2026 · 12 min read

SEO competes for a position in a list of ranked results where positions four through ten still earn clicks. GEO competes to be one of roughly three sources cited inside a single synthesised answer, where everyone else gets nothing. That winner-concentration difference is what changes the tactics: self-contained answerable passages beat comprehensive pages, and being structurally unreadable means not competing at all rather than ranking poorly.

Two bad takes dominate this topic. One says GEO is a rebrand and good SEO covers it. The other says search is over and SEO is dead. Both are wrong in ways that cost money — the first leaves you invisible on a growing surface, the second has you abandoning a channel that still drives the majority of ecommerce discovery.

The useful version is narrower: most of the technical foundation is shared, the content strategy genuinely diverges, and a handful of long-standing SEO habits are now actively counterproductive. This article separates those three groups.

The one structural difference that explains the rest

A search results page shows ten organic results. Position one takes a large share, but positions four through ten still receive meaningful traffic. Ranking is a gradient, and there is a long tail of partial wins.

An assistant's answer typically synthesises from a small handful of sources and names fewer. There is no position seven. You are cited or you are absent, and the gradient is replaced by a cliff.

	SEO	GEO
Unit of success	A ranked page	A cited passage
Winners per query	Ten, with a usable tail	Roughly three to five; the rest get nothing
Query model	Match a query string, plus intent	Answer a question, including phrasings never typed before
Ideal content shape	Comprehensive page that holds attention	Self-contained passage quotable without context
Feedback speed	Weeks to months	Days — retrieval happens at question time
Primary failure mode	Outranked by a stronger page	Structurally unreadable, so never considered
Measurement	Impressions, position, clicks from Search Console	Citation rate from your own probes; no platform reports it
Traffic consequence	A visit you can attribute	Often an answer with no click at all

Where the disciplines diverge

The final row deserves attention because it is the one that unsettles people. An assistant can answer a shopper's question using your page, credit you, and produce no visit. That is genuinely a loss relative to a click — and it is also how the surface works, and stores being named in answers still convert better than stores that are not mentioned at all.

What is shared — do this once

The overlap is larger than the GEO discourse admits, and it is the cheapest work available because it pays into both channels.

Crawlability. Both need to fetch your pages. The difference is the list of user agents — SEO cares about Googlebot and Bingbot, GEO adds GPTBot, ClaudeBot, PerplexityBot and Google-Extended.
Rendering. Content that only exists after client-side JavaScript is a risk for search crawlers and a bigger risk for AI fetchers. Server-rendered content is safer for both.
Structured data. Product schema drives rich results and gives assistants high-confidence facts. One implementation, two beneficiaries.
Information architecture. Clear categories with canonical URLs help both a crawler building an index and an assistant deciding which page authoritatively covers a topic.
Page speed. A slow page risks timing out for an assistant fetching under a latency budget as surely as it hurts search performance.
Genuine authority signals. Being mentioned, reviewed and linked by credible third parties matters to both — assistants weigh corroboration across sources heavily.

What genuinely diverges

Self-contained passages beat comprehensive pages

SEO rewarded the definitive 3,000-word guide, partly because comprehensiveness signalled quality and partly because a longer session was itself a positive signal. An assistant does not read your page for enjoyment. It extracts the passage that answers the question and discards the rest.

So the unit of optimisation shrinks from the page to the paragraph. Each section should state its conclusion in its own first sentences, without depending on the two sections above it. Practically that means writing more like a reference and less like an essay: a heading that is a question, an answer immediately below it, then the supporting detail.

Questions replace keywords

Keyword research assumed a finite set of strings people type. Conversational questions are longer, more constrained, and frequently unique — nobody has typed "waterproof backpack for commuting with a 16 inch laptop under 100 dollars" enough times for it to appear in a volume tool.

The replacement is constraint coverage. Rather than targeting a phrase, ensure your pages explicitly state the attributes that constrained questions filter on. You cannot enumerate the questions; you can make sure the facts they depend on are present.

Corroboration matters more than links alone

Assistants synthesise across sources and are noticeably more willing to name a store that multiple independent pages describe consistently. A backlink from a directory that says nothing about you contributes little. A review on a credible site, a mention in a category roundup, and consistent facts across your own site and third parties contribute a lot.

This shifts effort from link volume toward being genuinely, consistently described in places other than your own domain — which is closer to plain public relations than to technical SEO.

SEO habits that are now counterproductive

Habit	Why it worked	Why it now hurts
Keyword-density writing	String matching rewarded repetition	Repetition adds no facts; models weigh clarity, and stuffed prose reads as low quality
A thin page per keyword variant	Captured long-tail query strings	Fragments your authority across pages that each answer a question incompletely
Burying the answer to increase time on page	Session length was a positive signal	An assistant extracting an answer never reaches paragraph nine; the buried answer is invisible
Vague superlatives	Read as marketing polish	"Premium quality" satisfies no constraint; "full-grain leather" does
Blocking AI crawlers to protect content	Sensible for publishers protecting training data	For a store, it removes you from the recommendations you are trying to win

Tactics to retire

Running both without doing the work twice

1
Fix the shared foundation first
Crawler permissions for both search and AI user agents, server-rendered content, complete Product schema, current sitemap. Everything downstream depends on this and it serves both channels.
2
Rewrite for constraints, not keywords
For your top products and category pages, ensure the attributes buyers filter on are stated explicitly with numbers and units. This improves search relevance and AI retrievability with one pass.
3
Restructure pages so each section answers alone
Question-shaped headings, conclusion first, detail after. Human readers skim this way too, so it is not a trade-off against the SEO audience.
4
Measure separately, because the instruments differ
Search Console for impressions, position and clicks. Your own probe panel for citation rate, because no assistant publishes one. Do not try to force both into a single number.
5
Re-check after every platform or app change
GEO's feedback loop is days rather than months, so changes surface quickly — in both directions. An app install that duplicates schema or a theme update that alters robots.txt can undo weeks of work quietly.

Frequently asked questions

Is SEO dead?

No. Search still drives the majority of ecommerce discovery, and Google's own AI surfaces draw heavily on the same signals that traditional ranking uses. The change is that search is no longer the only surface worth optimising for, not that it stopped mattering.

Do I need separate content for GEO?

Rarely. The same page can serve both if it is structured so each section answers its own question and states facts with specifics. Maintaining a duplicate content set is expensive and produces two half-maintained versions.

Which should I invest in first?

The shared foundation, because it is cheap and serves both. After that it depends on your category: if your customers are already asking assistants for recommendations, GEO's short feedback loop makes it the faster experiment. Measure your citation rate before deciding.

Does ranking on Google help me get cited by AI?

Indirectly and inconsistently. Some assistants ground answers in search results, so ranking helps there. Others crawl independently, and a page can be cited by an assistant while ranking poorly, or rank well and never be cited. Treat them as correlated, not equivalent.

How do I measure GEO when there's no Search Console for it?

You build the instrument yourself: a fixed panel of buying questions asked of each assistant on a schedule, recording whether your domain appears. It is coarse but it is a real observation, and it is the only one available.

What about zero-click answers taking my traffic?

It is a genuine cost and largely outside your control. The available response is to be one of the cited sources rather than one of the absent ones, since being named still drives consideration and direct visits — and to accept that some share of AI-influenced revenue will never be traceable to a click.

The verdict

GEO is not a replacement discipline and it is not a rebrand. It is a second surface with a harsher winner distribution, a much shorter feedback loop, and no platform-provided analytics. Most of the technical groundwork is shared with SEO, which means the incremental cost of competing on it is smaller than the discourse implies.

The practical move: fix the shared foundation, retire the habits that assumed string matching and session length, restructure so every section answers its own question, and start measuring citation rate now so you have a baseline before your competitors do.