Think back to how people searched online about ten years ago. If your child’s soccer jersey was torn in a game, you might have typed something like ‘regulation soccer jersey for 11-year-old, durable’ into the search bar. You’d then have to scroll through a bunch of website links before finding what you needed. That has been the story of web search for the last twenty years. Website owners created pages to get your clicks, Google put ads on anything that seemed like it might be for shopping, and everyone got used to search with short, specific phrases.

Now the same parent can just type the whole thing in: ‘My 11-year-old is playing in a little league, we’ve already had 2 cheap jerseys torn up, what should I buy?’ Google’s AI Overview and as well as ChatGPT, Gemini, Perplexity, and Claude can handle that straight away, often without any further clicks. That’s been the trend over the past 2 years. People are typing in much more detailed queries; they’re asking more follow-up questions, and they’re using search in a completely different way.

In June 2026, Lutz Finger from Forbes reported on an A/B test conducted with Cornell master’s students across several e-commerce stores.  The test found nearly nine times the conversion rate for AI-guided conversational site search compared with conventional keyword site search.

People stopped asking in keywords

Nielsen Norman Group refers to the old way as ‘keyword foraging,’ and that’s absolutely the right term. You would often open five search tabs, then put the answer together yourself. That is keyword guessing.

In a 2025 study by Kate Moran, Maria Rosala, and Josh Brown, they looked at people using both traditional search and the new digital assistants. The assistants helped people refine their queries, but nobody stopped using Google. In fact, they tended to alternate between the two, occasionally checking one against the other. It’s a very narrow study with just nine participants, so it’s really a study of behavioral change rather than broad trends.

According to Google, in September 2024, long searches (those over 5 words) were growing 1.5x faster than short searches. The new search features have accelerated that even more. In their May 2026 report, they noted that searches in these modes are typically 3x longer than standard searches, and both long and short searches are still growing. On the most recent earnings call, Alphabet also reported that 1 in 5 of these searches now involve voice, image, or document.

The language of the search queries has changed. “What”, “how” and “I” are all now some of the most common words to start a search query. “I” in particular is a massive one. You don’t normally see “I” at the front of a keyword search query; you see it all the time in conversational search.

Why Google can handle long questions now

Long-tail questions used to fall flat under the old retrieval model, which focused on term matching. Google’s patents show you what the new approach is.

The first patent to note is US11769017B1, “Generative summaries for search results,” filed in March 2023 and granted in September that year. It’s a system which reads the result pages, creates a summary and then links the points in that summary back to the relevant source(s). It will only display the summary if it’s confident in the result. Many people in the search world regard this as the fundamental underpinning of Google’s new summary capability.

The second patent, “Search with stateful chat” (US20240289407A1), was filed in August 2024. This system takes the previous questions in a search session and uses them to inform the next search. That’s precisely what Google now calls query fan-out. The official documentation states that summary features and conversational modes can run multiple related searches in the background to build a complete answer. Another patent, “Thematic search” (US12158907B1), focuses on grouping results by theme, with each group having its own summary.

A patent tells you what Google built and what it tried to protect, not necessarily what’s running today. Google has never referred to any of these as ‘the AI Mode patent’. Treat them as pointers of direction and then cross that against the actual Google docs and what you see in the live results.

The practical effect is easy to picture. That one ‘soccer goal’ query could trigger a dozen background searches on wind loading, sizes for an 11-year-old, and what materials will work. A page could be pulled in on one of those even if the parent had never searched for it at all. The importance of ranking for a single term is decreasing; the importance of covering the questions the model is going to ask is increasing.

Searching for things nobody used to search for

Search now covers a vastly wider range of topics. Michael White, global sector specialist at Schroders, noted in late 2025 that ChatGPT and similar tools were being used for all sorts of things that you’d never type into Google (i.e., generate an image, fix some code, improve an email). Google has seen the same thing. On its 2025 earnings call, Google said that when people discover the new search can do these more sophisticated things, they search more. In fact, they specifically said that the younger demographic is searching more as a result.

The dominant themes in these new search modes are creativity and content generation. Google points out that queries related to planning are growing faster than the overall trend, and searches starting with ‘where should I’ and ‘ideas for’ are also up.

The numbers confirm this. In its Q3 2026 report (issued 28th October), Alphabet noted that search query growth was the strongest it had seen in five years (across both commercial and non-commercial search) and attributed this to new technology investments. By the end of its first year, the new search mode had reached over a billion monthly users globally, and query volume was more than doubling quarter-on-quarter.

You’ll occasionally hear the claim that traditional search volume growth has gone from 19% in Q1 2026 to 12% in Q3. That’s true, but again it’s about revenue. Google Search and other revenue grew 19%, 17%, and 12% in Q1, Q2, and Q3, respectively. Philipp Schindler (Google’s CBO) told analysts that lower monetization in AI Mode was one reason for the dip, along with currency and some retail and media trends. People are searching more than ever, and Google still hasn’t figured out how to monetize this new kind of search. That gap is the commercial story.

Search turned into a conversation

Traditional search would have dealt with each query in isolation. If you got a bad result, you’d just rephrase and start again. Modern search is all about the conversation. Take the soccer goal example again. You could ask for a cheaper one, then ask about a small yard, and the system would have remembered your original query. Google’s own reports show that the number of follow-up questions has steadily increased since the feature launched, which is exactly the trend their conversational search patents describe.

Advertising is now being designed to work in these new search experiences. On the earnings call last week, Google announced it’s starting to serve ads in the summary and conversational search experiences for more users, more queries, and more countries. In the US, they’ve already begun testing multi-turn ads which can appear in the middle of a conversation rather than just at the start. In January, they ran a pilot of Direct Offers, a new search format that shows discounts to shoppers likely to buy.

How many searches actually go to waste with no click?

Ask three people what the zero-click rate is, and you’ll get three different answers. It’s all down to who’s doing the measurement and what they define as a click. You have to check the source.

The Schroders analysis is based on Semrush data on U.S. desktop searches (27% of searches result in no click, 56% in a click). The 56% figure includes searches that go to Google properties (Images, YouTube, Shopping, etc.), so it overstates open web traffic. The % of searches that result in no click has risen from March 2024 to March 2025 as more searches are answered directly.

SparkToro’s 2024 study used Datos clickstream data and a broader definition, and it came in at 58.5%. Of every 1,000 US Google searches, only 374 resulted in a click. Most of the difference between the two studies comes down to different panels and definitions.

For the summary features, Pew Research looked at March 2025 browsing data from 900 U.S. adults. When a summary appeared at the top of the results, users clicked the standard link on just 8% of visits, compared with 15% when there was no summary. Only 1% of the visits resulted in a click on a link within the summary. Google questioned Pew’s methodology and the set of search queries it used.

Pew’s data also shows a correlation between zero-click behavior and the way people phrase their searches. Just 8% of the one or two words searches resulted in a summary. That rises to 53% for searches of 10+ words and 60% for those that start with a question word (who, why, etc.). The more natural the question, the more likely it is to be answered with a summary rather than a list of links.

Google maintains that the clicks it delivers are higher quality. Search Central has suggested that the users coming in via the summary are more engaged and spend more time on the site. Site owners can’t really verify that, though, because Search Console lumps all traffic from the new features into overall search traffic and doesn’t report it as a separate category.

Agentic commerce hit a wall at checkout

The most ambitious idea is that rather than asking the AI questions, we’d just have to assign it a task or series of tasks. The first real attempt at that sort of thing collapsed.

Instant Checkout came to ChatGPT on 29th September 2025 with Etsy as a partner. It was an open Agentic Commerce Protocol on top of Stripe and Shopify, with Walmart also announced as a partner. By March 2026, only about 30 Shopify merchants had gone live (a long way short of the million-plus that had been promised). Daniel Danker from Walmart was at an investor conference that month and noted that the conversion rate on purchases made within ChatGPT was about a third of what Walmart saw when it sent customers directly to its own site. OpenAI then went into discovery: you find the product in ChatGPT, and then you check out on the retailer’s site or in the retailer’s own ChatGPT app.

Google took a different approach. At NRF in January 2026, they announced the Universal Commerce Protocol (UCP), which is an open standard developed jointly with Shopify and backed by a number of retailers (Etsy, Wayfair, Target, Walmart). UCP handles checkout on eligible product listings in AI Mode and in the Gemini app for US retailers (initially with Google Pay), with the retailer as the seller of record. Google also launched Business Agent, a branded chat agent which answers product questions directly in Search.

Finding products via digital assistants is increasingly the norm. Getting to the point where you can complete a purchase entirely through the assistant is another matter. Some people talk about the day when agents can book you a holiday or make a purchase on your behalf automatically, but again, neither is the current reality.

What this means for publishers and SEOs

If you follow Google’s main recommendations, you just need to do regular SEO. Your page must show up in Google’s index and be able to appear as a summary. The process of optimization has not changed. There are no secret tricks or special moves needed to get your content into summaries or conversation-style answers, according to the May 2026 Search Central guide. Just make sure your page can be found by Google, link to important pages, focus on making your site easy to use, and write your main content in plain text. You don’t need any extra code or fancy formatting.

Cover the question behind the question. Query fan-out is the process of taking one long question and breaking it into several smaller ones. A page about soccer goals should also cover wind, sizing, and durability. Each follow-up answer is another opportunity to be found and cited.

Publish content that is genuinely unique and not available anywhere else. Straight answers will just repeat facts that are already out there. Google’s helpful content guidelines also encourage original research, first-hand experience, unique data, and expert interviews (all of which a summary would be expected to cite).

Make sure your content appears on different websites and apps. In the NN/g study, people often switched between Google and chatbots, using both to double-check information. To stay visible, you need to show up and be mentioned on many types of search tools.

Adjust your metrics. Search Console will combine all traffic, and long question type searches tend to lose more clicks than short ones, so raw click numbers are pretty much useless. Look at impressions, look at citations, and then track what your visitors are doing once they’ve got to your site.

Search engines are becoming answer engines and not just a starting point. IN the last year, more and more, people are asking longer and more personal questions, and Google is giving them answers right away, so they don’t always have to go to other websites. But those answers still come from different sites. Most people still finish buying things on the store’s own website, and if something is important, they still visit the site to make sure it’s right. Your website is the source material for these answers, not just landing pages..

 

Sources: