The AI Learning Hub
Published 1 November 2026

AI and Search: Why Fewer People Click, and What Still Gets Found

For twenty-five years, search worked the same way. You typed a few words, got a page of ten blue links, and clicked one. The search engine was a signpost. The answer lived somewhere else.

That arrangement is breaking. Google now puts an AI-written summary, called an AI Overview, at the top of many results, and offers a full conversational AI Mode. ChatGPT, Perplexity, Claude and others answer questions directly and search the web on your behalf. More and more often, the search engine is the destination.

This post looks at what changed, who it affects, and what still works. As always in this journal, it is dated: 1 November 2026.

What actually changed

The answer moved onto the results page. Instead of pointing you to pages that might answer your question, the engine reads those pages and writes the answer itself, usually with a few source links underneath.

Questions got longer. People type to an AI the way they talk: “what’s a good laptop for video editing under $1,500 that isn’t too heavy”. Search used to reward short keywords. It now handles whole situations.

Search became a conversation. You can ask a follow-up (“what about battery life?”) without starting over. The engine remembers the context.

Some searches never leave the chat. A growing share of what people once “googled” now starts in an AI assistant, which may search several sites, compare them and give one combined answer.

Fewer clicks: what the evidence says

The obvious question is what happens to the websites whose content the AI is summarising. The early evidence points one way, although the size of the effect depends heavily on how it is measured.

  • A randomised experiment posted in April 2026, with about a thousand desktop Chrome users, found that AI Overviews cut clicks to outside websites by roughly 40% and noticeably increased searches that ended without any click.
  • Ahrefs, analysing search data at scale, reported in early 2026 that when an AI Overview appears, the top-ranked result gets around 58% fewer clicks.
  • Publishers’ own industry bodies report losses ranging from small to severe, depending on the kind of content.

Google disputes some of these figures and has added more visible source links to its AI answers. But the direction is consistent across studies: when the answer is on the page, fewer people click through.

The losses are not spread evenly. Simple factual questions (“how long to boil an egg”, “capital of Australia”) are hit hardest, because the summary fully answers them. Content people want to experience for themselves, like reviews with original testing, detailed guides, tools, communities and news with real reporting, holds up better.

What this means if you are searching

AI answers are convenient, and for low-stakes questions they are usually fine. For anything that matters, the same rules apply as for any AI output:

  1. Open the sources. An AI summary can blend two sources into a claim neither of them made, or quote an outdated page. The links are there for a reason.
  2. Notice what is missing. A summary gives you one confident answer. A page of results showed you that people disagreed. For health, money, legal or safety questions, look for that disagreement yourself.
  3. Check the date. AI summaries don’t always make clear when their information is from. Prices, laws and product details go stale quickly.

Our post on how to check an AI answer covers a quick routine for this.

What this means if you run a website

The old playbook of picking a keyword, writing a page and climbing the rankings is losing value on its own. What is replacing it has picked up names like “generative engine optimisation” and “answer engine optimisation”, but most of it is good practice that matters more now:

  • Be the source, not the summary. AI systems summarise what already exists. Original data, first-hand testing, expert experience and clear opinions are what they cite, and what a summary cannot replace.
  • Answer clearly and early. Pages that state the answer plainly, then explain it, are easier for both people and AI systems to use and to cite.
  • Make your pages machine-readable. Clean headings, descriptive titles, and structured data (the schema.org markup that tells machines “this is a course”, “this is an article”, “this term means this”) help AI systems understand and attribute your content.
  • Build reasons to visit directly. Newsletters, tools, communities and a recognisable name all bring people to you without depending on a search engine sending them.
  • Measure differently. Being cited inside an AI answer builds awareness even when it doesn’t produce a click. Track brand searches and direct visits, not just organic traffic.

The bigger picture

The deal that held the web together was simple: websites let search engines read their pages, and search engines sent visitors back. AI answers weaken the second half of that deal, which is why publishers are negotiating licences, filing complaints and in some cases blocking AI crawlers. How that settles will shape what the open web looks like in five years.

For now, two things seem safe to say. People will keep wanting answers fast, and AI gives them that. And the content that keeps its value is the content that AI needs but cannot produce itself: something new, tested, or genuinely expert.

Where to go deeper

AI Essentials explains how these models find and summarise information, and why they sometimes get it wrong. Prompting shows how to ask AI tools better questions, and Does Your AI Actually Work? covers how to judge whether an AI answer can be trusted. All free.

Related free courses