Search is changing: from keywords to AI recommendations

For twenty years, being found online meant the same thing: pick the right keywords, earn the right links, climb the rankings. That model is now unwinding. The interface between a customer and the businesses that serve them is shifting from a list of links to a single, reasoned recommendation, delivered by an AI assistant that never clicks on anything at all.

This shift did not arrive overnight. It has moved through four distinct eras, each redefining what "visibility" means for the businesses competing inside it. Understanding those eras, and where we are now, is the difference between preparing for the change and being quietly removed from consideration.

Era one: ten blue links

The original search bargain was straightforward. Google indexed the web, users typed keywords, and the engine returned ten organic results per page. Visibility meant ranking on page one. The discipline that grew around it, search engine optimisation, was, at its core, a keyword game: research what people search for, put those terms in your titles, headings and body copy, earn backlinks from other sites, and wait for the crawler to notice.

Businesses that understood this early built enormous organic traffic. Those that didn't relied on paid ads or word of mouth. The important structural point is that every result was a link. The search engine's job ended at the click. What happened next, the browsing, the comparing, the buying, took place on the merchant's own site, on the merchant's terms.

For nearly fifteen years, this model held. It rewarded content volume, link acquisition and technical SEO. It also rewarded patience: rankings took months to build and, once earned, tended to compound.

Era two: the featured snippet

Around 2014, Google began answering questions directly on the results page. Featured snippets, knowledge panels, "People also ask" boxes, collectively known as "position zero", extracted content from websites and displayed it above the organic results.

This changed the economics. A site could rank first organically and still lose the click, because the answer was already visible without one. Studies at the time showed that featured snippets captured a meaningful share of attention that would previously have flowed to the top organic result.

Visibility in this era meant structuring your content so Google could extract it cleanly: clear question-and-answer formatting, concise definitions, well-marked lists and tables. The businesses that adapted their content strategy thrived. Those that kept optimising purely for traditional rankings found their click-through rates quietly eroding.

The deeper lesson was subtler. Google had begun to value the answer over the source. The link was still there, but the incentive to click it was weakening.

Era three: AI overviews

In 2024 and 2025, the major search engines folded generative AI directly into the results page. Google's AI Overviews (and later AI Mode), Bing's Copilot integration and Perplexity's answer engine all do the same thing: synthesise information from multiple sources into a single, conversational response presented above, or instead of, the traditional results.

The shift here is qualitative, not just positional. In the snippet era, Google extracted a passage from one source. In the AI overview era, the engine reads dozens of sources, reasons across them and writes its own answer. Citations appear as small links beneath the summary, but research consistently shows that most users read the overview and never scroll further.

For businesses, this creates a new problem. Being cited in an AI overview is valuable, but being cited is not the same as being clicked. The search engine has become the reader, and the user trusts its summary. Visibility now means being one of the sources the model draws on, which requires authority, specificity and structured content that the model can parse and verify.

Keyword rankings still exist, but their relationship to commercial outcomes has loosened. A page can rank well, be cited in an overview and still see declining traffic, because the overview gave the user enough to act without visiting the site.

Era four: the AI agent

This is where the shift becomes structural. AI assistants, ChatGPT, Gemini, Claude, Copilot and their successors, are not search engines with a chatbot wrapper. They are the engines of agentic commerce, autonomous systems that interpret a user's intent, gather information from structured data sources, reason over it and, increasingly, take action on the user's behalf.

When a customer asks an AI assistant to find the best cordless drill for occasional home use under 100 pounds, the agent does not return a list of links. It evaluates products across multiple sources, weighs attributes against the user's constraints, considers trust signals such as reviews and seller verification, and presents one or two recommendations with reasoning. If the agent can complete the purchase, through protocols like Stripe's Agentic Commerce Protocol or Google's Agents to Payments, it offers to do so inside the conversation.

The customer never visits a search results page. They may never visit the merchant's website. The entire journey from intent to transaction happens within the assistant. Not all agents work the same way, our comparison of AI shopping agents shows how ChatGPT, Google, Perplexity and others differ in their evaluation.

What visibility means now

Each era redefined visibility, and each redefinition removed a layer of control from the merchant:

• Ten blue links: visibility meant ranking. You earned clicks and controlled the experience from there.
• Featured snippets: visibility meant being extractable. You might lose the click, but your brand still appeared prominently.
• AI overviews: visibility meant being citeable. Your content informed the answer, but the search engine presented it.
• AI agents: visibility means being recommendable. The agent must be able to find your structured data, verify your trustworthiness and, ideally, transact with you directly.

The shift at each stage is the same: from the user pulling information to the machine pushing a decision. In the agentic commerce era, that process reaches its conclusion. The machine does not just answer, it chooses.

How businesses win in the agentic commerce era

The old playbook, keywords, backlinks, meta descriptions, is not wrong. It is insufficient. Agents reason over structured, machine-readable data. The signals that determine whether a business gets recommended look different from those that determined whether it ranked:

Structured data over keyword density. Agents consume product feeds, schema markup and API-exposed catalogues. Attributes must be complete, consistent and accurate. A product with missing sizes or vague specifications is not ranked lower, it is not considered.

Trust over authority. Backlinks measured authority in the SEO era. Agents measure trust: verified seller status, authentic reviews, clear returns policies, accurate stock data. The agent's job is to protect its user from bad outcomes, and it evaluates merchants accordingly.

Transactability over traffic. A product the agent can purchase through a supported checkout protocol beats an identical product it can only link to. Being transactable inside the agent's workflow is a competitive advantage that did not exist two years ago.

Third-party corroboration over self-promotion. Agents look for external evidence, buying guides, editorial reviews, comparison content, to validate a merchant's claims. Content marketing is not dead; it has shifted from attracting clicks to building the evidence base that agents cite.

The uncomfortable reality is that your competitors do not need to be better. They need only be easier for AI to understand. In a system where the machine chooses, in agentic commerce, legibility is the new ranking factor.

What to do about it

The strategic response is not to abandon SEO but to build on top of it. Structured data, trust signals and agent-compatible checkout are additive, they improve traditional search performance while preparing for the channels that are replacing it.

For most businesses, this means auditing product data for machine readability, investing in trust infrastructure (reviews, policies, verification) and joining platforms or marketplaces that handle the technical layer of agent distribution. The economics favour participation over solo construction: the protocols are new, the standards are moving, and the platforms that aggregate structured catalogues, Vendoora included, exist precisely to absorb that complexity on the merchant's behalf.

Search is not disappearing. It is being absorbed into something broader: AI-mediated discovery, where the question is not "how do I rank?" but "will the agent recommend me?" The businesses that answer that question now will be the ones customers find, even if those customers never type a keyword again.

Read next