Generative search explained: how AI answers work and what they mean for your site

For most of the web’s history, searching meant typing a few words and choosing from a list of links. Generative search changes the output. You ask a question and get a written answer first, assembled from several sources, with the links underneath.

When we first wrote about this in 2024, Google’s version was an experiment called the Search Generative Experience. It has since become a normal part of Google, under the names AI Overviews and AI Mode, and similar answers sit at the heart of ChatGPT’s search, Perplexity and Microsoft Copilot. This article explains how they work and what follows for anyone who relies on search to be found.

How a generative search engine works

Behind the answer sits a simple sequence:

1. Understand the question. The system works out what you mean, including context and follow-up questions, beyond the individual words.
2. Retrieve sources. It searches an index of the web, or the live web, for pages likely to contain the answer.
3. Write the answer. A language model reads the retrieved passages and writes a summary in plain language.
4. Cite. Most systems show the sources they used, so that you can check them or read more.

This retrieve-then-write pattern is the same idea businesses use to make AI answer from their own documents. We describe it in our guide to building a knowledge base with RAG.

What users gain

• Speed. One synthesised answer replaces the work of opening five tabs and comparing them.
• Better handling of complex questions. Long, multi-part questions that were awkward as keywords now work as written.
• Follow-up. You can refine the question in conversation instead of starting again.
• Different formats. Many systems accept images and voice as well as text.

What publishers and businesses risk

The change is less comfortable for anyone who depends on visits. In Pew Research Center’s study of the browsing of 900 US adults in March 2025, people clicked a traditional result link on 8% of visits when an AI summary appeared, compared with 15% when it did not. Links inside the summaries were clicked on just 1% of visits.

Two other concerns are worth knowing about. Answers can be wrong, and a confident summary can hide that. And users often can’t tell how the engine chose its sources, which raises questions of fairness and transparency.

What is still changing

• More modes of input. Searching with a photo, a screen or a voice question is becoming ordinary.
• Personalisation. Systems increasingly use your own context, such as location and past questions, to tailor answers.
• Action, not only answers. Assistants are beginning to compare products, book appointments and complete purchases on a user’s behalf, which makes accurate, structured product and service data more valuable.
• Better checking. Providers are working on showing sources more clearly and flagging uncertain answers.

What to do about it

You can’t control how search engines present answers, but you can make your pages the kind they draw on:

• answer buyer questions directly near the top of the page, in plain words
• keep facts accurate and dated, and cite your sources
• structure pages with clear headings and add structured data
• make sure AI and search crawlers can reach the pages you want quoted
• build your reputation beyond your own website, since these systems read the whole web
• track how your brand appears when you ask the questions your customers ask

Our step-by-step guide to SEO for AI search covers each of these, and our piece on using generative AI in your SEO workflow looks at the production side. If you’d like to see how you appear in AI answers today, that is part of our AI-era growth work, and you can talk to us about a review.