Ask Google something today and the answer is usually waiting at the top of the page. It gives a summary, a couple of sources, and a follow-up box if you want to keep going. You read it, and you’re done. You didn’t visit anyone.

Now click through to a company’s website with a specific question, like whether their software works with the accounting system you already use. The answer is probably there, somewhere. But finding it is up to you: choose a menu, guess which page covers it, or try the search box and scan the results.

A year of the first experience is changing how people judge the second. This piece is about that gap, and why the part of your site that closes it isn’t the menu.

Search didn’t shrink. Clicking did.

In February 2024, Gartner predicted that traditional search engine volume would drop 25% by 2026, lost to AI chatbots and virtual agents. It’s 2026, and that isn’t what happened.

On Alphabet’s Q2 2026 earnings call in July, Sundar Pichai said AI features are “driving growth in queries,” that AI Mode has passed 1 billion monthly active users, and that it is “driving an incremental increase in Search queries overall.” People are searching as much as ever, and probably more.

What collapsed was the click:

Google still says it sends “billions of clicks to websites every week.” That’s true, and it still counts as a collapse. The company that built the web’s referral economy decided the best search result is one you don’t have to leave the page for. Everyone who uses Google is learning that lesson from it, including your visitors.

Who gets the answer, and who gets skipped

Pew’s data has a detail that matters more for site owners than the click-through numbers. AI summaries don’t show up evenly. They appeared on 8% of one- or two-word searches, but on 53% of searches of ten words or more, and 60% of searches phrased as a question.

In other words, Google now keeps the specific questions for itself. “ERP software” still sends you to a list of links. “Which cloud ERP works for a 40-person job shop that already runs QuickBooks?” gets answered on the results page.

So the questions that used to bring someone to your product page, your spec sheet or your comparison guide are the ones least likely to send anyone there now. Forrester has told B2B companies to expect 10–40% traffic declines as research moves into answer engines. They also advise that a website’s job is shifting “from a traffic generation and lead harvesting engine to an information syndication engine.”

That shift is happening. The next question is what it does to the visitors who still arrive.

The visitor who still shows up has been retrained

People don’t keep separate mental models for separate websites. They never have.

Back in 2014, Nielsen Norman Group put it bluntly: users’ “search mental model is corrupted by the big search engines and they expect search to work in the same way on every site.” That was about keyword search. Google taught everyone to expect a relevant list of links, and every site search that couldn’t produce one looked broken.

The same thing is happening again, only faster. As early as 2013, Jakob Nielsen wrote that the big search engines had “morphed into answer engines” for simple questions. AI has now extended that to complicated ones. A year of typing full sentences into Google, ChatGPT and Perplexity sets a new expectation: describe what you need in your own words, and get the answer. Not a list of places the answer might be.

NN/g’s 2025 research on how AI is changing search found that people’s habits are sticky. Most people still start with Google, and nobody in the study relied on AI alone. But once people saw an AI tool pull scattered information together and fit it to their situation, the researchers watched their habits change during the session.

That’s the visitor arriving at your site now:

What navigation was quietly doing, and can’t anymore

This is not an argument that navigation is dead. The same 2014 NN/g article makes the case for it well. Navigation “shows people what they can find on the site,” and it lets them recognise what they want instead of having to remember what to type. For orientation, browsing, trust and crawlability, a clear menu still earns its place.

But menus were also doing a second job that nobody wrote down: being the main way to find a specific answer. They were never good at it. A menu is a map of how the company organises its information: by product line, by department, by whatever made sense in the last redesign. The visitor doesn’t think in your org chart. They think in questions.

That’s why every redesign ends in a long argument about menu labels, and why none of those arguments ever quite settles anything. The label doesn’t matter to someone who never reads the menu. Jakob Nielsen noticed back in 1997 that more than half of users were search-dominant. They went straight for the search box because they wanted “to find specific information as fast as possible.” They were the majority then. Answer engines have now trained everyone else to behave the same way.

So the job moves. Navigation handles orientation. Findability handles the answer.

Most sites think they’ve covered findability because there’s a magnifying glass in the header. Findability is really whether a question, asked the visitor’s way, reaches the right paragraph on your site. That comes down to three things:

  1. Retrievable. Can a specific answer actually be pulled out? If it’s on page 11 of a scanned PDF, in an image of a table, or split across four near-duplicate pages, it’s on your site but nobody can find it. Nobody is short of content. They’re short of a way to get one paragraph out of eleven years of it.
  2. Answerable. Can the site respond to what was asked, in the words it was asked in? Matching keywords returns documents. Visitors who’ve been using answer engines expect an answer, and they judge a list of ten links the way they now judge ten blue links: as extra work.
  3. Verifiable. Does the answer show where it came from? This is the part most sites and most chatbots get wrong. People go back to Google’s links, and to a human, when they don’t trust a bare answer. An answer that names the document, the section and the page is one they can check in a click, and one they’ll act on.

The third point is where a site can beat the answer engine it’s competing with. Google’s summary is built from whatever it could crawl, and it’s generic. Your site holds the actual spec sheet, the actual policy and the actual vendor profile. An answer grounded in those and cited back to them is more specific and more trustworthy than what the visitor just got from a chatbot. That’s the reason to answer on your own site instead of leaving the job to Google.

There’s a bonus. The work that makes you findable on your own site is most of the work that gets you cited off it. Content that’s clearly structured, specific, written in plain answer-first language and easy to pull out in pieces is exactly what AI Overviews, ChatGPT and Perplexity quote. On-site findability and “AI search visibility” are largely the same project.

The question log is the analytics you’ve been missing

Navigation analytics tell you where people clicked. They can’t tell you what people wanted and didn’t find, because a visitor who can’t find something leaves without a trace.

Questions are different. Every question typed into your site is the visitor telling you, in their own words, what they came for. Every question your site couldn’t answer is a content-gap report written by a customer.

Top10ERP.org, a public ERP comparison directory, added an assistant that answers visitor questions from its own vendor and product content, cites the page each answer came from, and logs what it couldn’t answer. Their president, Christina Morrison, put the second benefit plainly: the dashboard gives them “real insight into what visitors are asking so we know exactly where to expand our content next.”

Most sites currently throw that signal away. Search queries, if they’re logged at all, sit in an analytics report nobody opens. In an ask-first world, that log is the most honest research your content team will ever get.

What to do about it

None of this needs a redesign. Most of it is a few afternoons of looking honestly at what you already have.

  1. Run the ten-question test. Ask sales and support for the ten questions they hear most often. Type each one into your site search, phrased the way a customer says it. Screenshot the results. For most sites, those screenshots make the whole argument.
  2. Look at entry pages, not the homepage. Check what share of sessions start somewhere other than /. Those pages are your front door now. Each one should answer its own question and point clearly to the next one.
  3. Write answer-first content. Use headings a person would actually ask. Answer in the first sentence, not the fourth paragraph. Use numbers, limits and compatibility, not adjectives.
  4. Get answers out of PDFs. Put them into structured pages, or at the very least make sure the PDFs are indexed properly and can be quoted section by section.
  5. Promote search from an icon to an entry point. If most visitors are search-dominant, the box that serves them shouldn’t be the smallest thing in the header.
  6. Log every question and read the log monthly. The unanswered questions are your content plan.
  7. Keep the menu, but give it a smaller job. Orientation, trust and structure, not retrieval.

Where the numbers stop and the argument begins

Being straight about what this evidence does and doesn’t show:

The short version

Google has learned that people would rather get an answer than go looking for one. It rebuilt the world’s most-used product around that, and now more than a billion people a month use it that way. Nobody who uses Google every day comes to your site expecting to dig through a menu anymore.

Your homepage used to be the front door. Now the front door is whatever question someone just asked, on whatever page they happened to land on. The sites that do well from here won’t be the ones with the best menu. They’ll be the ones that can answer that question in the visitor’s own words, with the source attached.

Sources