Google AI Search in 2026: SEO, GEO and What to Do Next

A buyer searching for a particular fixing needs more than a fluent answer. They need the right size, finish, quantity and application. That is the starting point I use when thinking about AI search: what would make the answer useful enough for someone to take the next step?

My work includes SEO across more than 2,000 product SKUs in a B2B industrial catalogue. It has made me cautious about advice that promises visibility through a new acronym alone. A product page with conflicting specifications does not become useful because someone calls it GEO.

What has changed in Google AI search?

There are two recent developments worth separating. In May 2026, Google published a new resource covering generative AI optimisation, including original content and common AEO/GEO misconceptions. In June, it announced dedicated Search Generative AI performance reports. An update to that announcement says the reports had reached websites worldwide by 31 August 2026. Google's May announcement and reporting announcement explain the changes.

The reporting update matters because it adds a more specific view of AI visibility. Google describes impressions, pages, countries, dates and, for Search results, devices. It is still visibility data. An appearance in an answer is not an order, a qualified enquiry or proof that the visitor read your page.

SEO, AEO and GEO: what do the terms mean?

SEO is the established work of helping search engines discover and understand useful pages. AEO, or answer engine optimisation, usually describes making content suitable for direct answers. GEO, or generative engine optimisation, is used for visibility within AI-generated responses. The boundaries are loose, so I would ask anyone proposing this work to name the pages, changes and measures involved.

Google's AI-features guidance says that established SEO practices remain relevant. Pages need to be indexed and eligible for a search snippet; there is no special AI schema or extra AI text file required for inclusion. That is Google's position on its own search features, not a universal specification for every AI service. Read Google's technical guidance.

Make the page answer a buying question

Consider a hypothetical page selling threaded rod. A broad paragraph about strength tells the buyer very little. A useful page distinguishes diameter, length, material, finish, pack quantity and the conditions attached to any suitability claim. A comparison guide should explain which differences affect the choice and link to the relevant products.

I would organise that information around the decision. Put the short answer near the top, then specifications, important limitations and the next step. If a finish needs compatible components, make that relationship explicit. If the product is supplied without fixings, say so beside the product information. Do not force readers to assemble the answer from an image, a download and three separate tabs.

This is where genuine subject knowledge can add something. A supplier can explain the question customers repeatedly ask, show a real installation detail or clarify two products that are often confused. Rewriting a general definition for the twentieth time adds much less.

A practical AI search audit for an online shop

1. Choose a small product family. Start with a commercially important category where you understand the products and customer questions.
2. List the decisions. Include compatibility, dimensions, ordering quantity, delivery and common alternatives.
3. Check the evidence. Compare page copy with the product data, supplier information and photographs. Resolve disagreements before adding more text.
4. Check discovery. Follow the route from category to product without using the site's search box. Important pages need understandable links.
5. Record a baseline. Note relevant pages, search visibility and commercial outcomes before changing anything.

My WooCommerce product SEO checklist goes further into catalogue consistency. That foundation is useful whether a customer arrives from a conventional search result, an AI answer or a paid campaign.

Measure visibility and commercial value separately

I would keep three columns in a monthly review: discovery, engagement and outcome. Discovery records where the page appeared. Engagement records what visitors did after arriving. Outcome records purchases or meaningful enquiries. Mixing them makes a report look impressive while hiding the actual question.

For example, growing AI impressions alongside flat revenue might justify improving the next step on the page. It might also mean the page is answering an informational question that rarely leads to an immediate purchase. Neither interpretation is established by impressions alone. Compare similar periods, note catalogue changes and consider the delay between research and an order.

A small collection of manually checked prompts can help you see how a brand or product is described. Treat those checks as observations. Responses can vary, and a screenshot of one answer is not a stable market-share measure.

What I would prioritise this month

Choose ten useful pages and make them easier to trust. Correct the specifications, improve the links, add the missing buying information and make sure the relevant products can actually be ordered. Then review what changed. That is a task with a clear owner and a measurable scope.

I would be much less interested in publishing hundreds of nearly identical answer pages. The advantage an experienced e-commerce team has is knowing the products and the awkward questions. The content should show that knowledge.

Common questions

Does GEO replace SEO?

I treat GEO as a description of an additional discovery context. It does not remove the need for accurate, accessible and useful pages.

Will structured data guarantee an AI citation?

No. Markup can help describe content, but it cannot guarantee that a search system chooses it. Keep it consistent with the information readers can see.

What should a smaller store do first?

Improve a narrow category you can validate properly. A reliable set of product and comparison pages is a better starting point than a large publishing target.

Back to the journal