GEO vs SEO for ecommerce: get your products found when shoppers ask AI

Search engine optimisation improves your visibility in search through accessible websites, useful content, relevant product information and the other signals that help search engines interpret and present your pages. Generative engine optimisation describes work intended to improve how a brand and its products appear in AI-generated answers and recommendations. GEO here means generative engine optimisation, not geographic targeting; the GEO glossary entry has the short definition.

For an ecommerce team, the useful distinction is the task the shopper is completing. A category search might produce a list of pages to explore. A conversational request might ask a system to compare products against several requirements and explain a recommendation. Both need accurate, accessible information.

Google says its established SEO practices remain relevant to its AI features, with no special technical requirements or special schema needed to appear, and that meeting the requirements does not guarantee inclusion. Google's guide to optimising for its generative AI features is a sensible place to start before buying a separate optimisation package.

Take an illustrative example. A retailer targets “black daypack”. A shopper then asks: “Find me a black backpack under £80, around 24 litres, with dimensions I can compare against my airline's personal-item allowance.”

A page headed “The ultimate everyday companion” may communicate a style, but it does not establish capacity, measurements or price, and more persuasive adjectives will not supply those facts. A useful product record states the verified capacity, external dimensions, colour and current price, so the shopper can compare the dimensions against their own airline's rules. It should not turn the measurements into an unsupported “approved for every airline” claim.

Some of those answers belong in structured attributes and some need clear explanatory copy. Neither means abandoning SEO.

These are different outcomes, and telling them apart stops every missing recommendation being treated as a content failure.

• Being available to a system means it has a route to relevant information: a website, a supported product integration or another source. Check the actual platform rather than assuming one submission reaches every assistant.
• Being understood means the available information accurately describes the product, its variants and its offer; a product can be discoverable while its size, compatibility or availability stays unclear.
• Being chosen means it is included for a particular request. A complete record may establish that an item is unsuitable, which is still useful, so completeness is no guarantee of a recommendation.

OpenAI describes product selection in ChatGPT in terms of the user's request, context and product information, and treats the merchant separately, where availability, price and merchant quality can count; the ChatGPT Shopping ranking factors guide sets out both decisions. A citation is different again: an assistant might cite your buying guide without recommending a product, or show a product without sending a measurable visit.

If the product cannot be reached, investigate access

Check the live product URL, indexing and crawl settings, integration status and destination diagnostics. Establish whether the product or variant is actually available through the route you are evaluating. New articles will not fix a broken product URL or a rejected feed record; for ChatGPT, the troubleshooting guide runs the access and eligibility checks in order.

Access was part of the work in Sighted's pilot with Onboard Sportswear: duplicate URLs were consolidated, the site's first sitemap submitted and indexing gaps addressed, alongside enriching the products. Across the pilot, AI citation coverage rose from 0% to 33% of tested responses, organic clicks by 59% and Google AI-feature impressions by 1,140%. A crawler that cannot settle on which of two URLs is the page has less to cite.

If basic facts are missing, improve the source record

Look for absent dimensions, material composition, compatibility, model identifiers or other decision-making details for the category. Ask which information is verified and where it lives, and start with commercially important, available products whose gaps you can resolve. The product data optimisation guide shows a source-to-approval process that does this without inventing attributes.

If the facts disagree, resolve the contradiction

Check the exact variant across the visible page, structured data and feeds. A parent product page and one size-specific offer are not interchangeable. If one destination shows an old price or the wrong image, find the stale mapping or update process before producing more copy.

If suitability is unclear, answer the purchase question

A specification can be accurate but hard to apply. Explain verified use cases and limitations in language a shopper understands: for a backpack, “45 × 30 × 20 cm externally” is more useful than “perfect travel size”; for an accessory, a named compatible model beats “works with leading devices”. A product photograph can show appearance, but it cannot prove waterproofing, certification or electrical compatibility.

If the offer is uncompetitive, look beyond content

Price, delivery, availability and returns affect a shopper's choice, and better descriptions cannot restock a variant or shorten a delivery promise. Record these factors when comparing observations, so an operational issue does not become an endless copywriting project.

If nobody agrees what visibility means, define it

Decide whether the goal is a brand mention, a cited page, a product recommendation, a qualified visit or a sale. Each needs its own record, or a report can look impressive while answering the wrong commercial question.

Imagine a retailer with 200 backpack listings. Its priority range is in stock, its category pages attract relevant search traffic, and the team wants more visibility for detailed travel-shopping requests. This is a demonstration, not a client result or a test of any assistant.

A first review finds three issues:

• ten priority products lack external dimensions;
• several colour variants share an outdated image mapping;
• a buying guide claims universal airline suitability with nothing to support it.

The team takes measurements from the approved supplier record, corrects the variant mappings and checks the published output, then replaces the airline claim with accurate dimensions and an instruction to check the carrier's current allowance.

Only then does it decide whether another guide would add value. If customers still ask how to choose between capacities, a guide that answers that and links to the right products earns its place. Because the changes are logged, later observations are easier to read.

Choose a manageable group of priority products and a fixed set of realistic shopping requests, covering different needs and budgets rather than the one question most favourable to your range. For each observation, record the platform, date, request, location or settings, recommended products and cited sources, and save the answer or a screenshot. Repeated observations expose variability; one favourable answer is not a visibility percentage.

Keep the change log beside the observations. If price, availability, advertising and product data all changed, a before-and-after cannot isolate the effect of enrichment, so compare similar products over the same period where you can.

Then connect visibility to commercial measures: relevant visits, product engagement and purchases where attribution allows. More mentions justify a closer look, but they are not automatically more revenue. The AI shopping measurement guide covers how to connect these layers.

Does GEO replace ecommerce SEO?

No. Accessible pages, useful content and accurate product information remain important. GEO adds questions about how products and brands are represented in AI answers and shopping experiences, so plan the work together and measure the outcomes separately.

Can product data guarantee AI recommendations?

No. Better information makes a product easier to assess, and it may also show that the product is unsuitable for a request. The system, competing offers and the shopper's requirements all matter.

Should you write more blog posts or fix product data first?

Find the gap first. Missing specifications, contradictory variants and broken publication routes need fixing at source. Editorial content can then answer broader buying questions and point shoppers to relevant products.

Do you need a separate strategy for every AI platform?

Use a shared foundation of verified product information, then check each platform's supported routes and requirements. Do not assume a feed, schema change or content tactic works the same way everywhere.

1. Decide what visibility means for you: a brand mention, a cited page, a product recommendation, a qualified visit or a sale, each with its own record.
2. Check access for your priority products: the live product URL, indexing and crawl settings, integration status and destination diagnostics on the route you are evaluating.
3. Fill the missing facts from verified sources, starting with commercially important, available products whose gaps you can resolve.
4. Check the exact variant across the visible page, structured data and feeds, and fix any stale mapping before producing more copy.
5. Explain verified use cases and limitations in language a shopper understands, and record price, delivery, availability and returns when comparing observations.
6. Benchmark a fixed set of realistic shopping requests against a group of priority products, log each observation beside the change log, and connect the result to visits and purchases.

Sighted scores every product on the data an assistant reads, writes the missing fields from your own descriptions, options and photos, publishes what you approve (or what you choose to publish automatically), and checks on a schedule whether the assistants name you. The free AI visibility audit is the first of those steps on up to 50 of your product pages, and on Shopify the approved fields go back into the store.

The enrichment is the technology behind SO-ME, the consumer shopping platform, where the detailed requests in this guide arrive from the shopper's side. Deciding which facts are trustworthy, and whether a change helps a shopper decide, stays with your team.