AI Search Changes How You Find Facts and Get Clicks

You search for an answer, read the AI summary, and close the tab before visiting the page behind it.

The answer may be useful. The person who wrote the source gets no visit, and you never see what the summary may have left out.

You can use AI search without choosing to

You don't have to switch search engines to change how you search, because Google's AI summaries can answer a question before you open a link.

Pew's February 2026 survey found 49% of US adults used chatbots, up from 33% in 2024, although the question wording changed.

It also found 42% used chatbots to find information. If you've asked one to explain an error message, you've used it to do a search engine's job.

The AP-NORC survey from 2025 found AI information seeking too, but asked about AI more broadly, so its results are not directly comparable.

Both surveys concern US adults. They count people who report using AI, so 42% is not the share of searches handled by chatbots.

And an Overview doesn't greet you on every search. Google says it appears selectively, when its systems judge that it adds something to the results.

The links are still there, but fewer people open them

If the answer is sitting at the top of the page, why scroll down to a website you've never heard of?

Some people still do, as Pew's browsing study shows. But visits with a click on a standard result were less frequent when a summary appeared.

Links inside the summaries got a click in 1% of visits with a summary. Being named as a source can still leave you without a visitor.

Pew tracked 900 US adults, then reran their searches later to identify summaries. The comparison shows an association, not a randomly assigned test.

Ahrefs' February 2026 research also found lower click rates for top results when an Overview appeared, using its own search data.

An August 2026 experimental preprint tested removing AI Overviews and AI Mode from Google. Participants clicked through to publishers more often with those features removed.

Whether someone clicked because they trusted the site's name is a separate question, and these results don't tell us which unfamiliar sites readers avoided.

Those lost clicks need a closer look

We recalculated the gap between 15% and 8%. It is seven percentage points, or about 46.7% less than the rate without summaries.

Picture 1,000 search visits in each group, with those rates: 150 would include a standard result click without summaries, compared with 80 with summaries.

Those are clicks on any standard result, across the study's searches. The example doesn't predict how many visits your own page would lose.

If you run a website, your own figures depend on the questions it answers, where it ranks and what else appears around the link.

A traffic drop is worth investigating before blaming the answer box. Check whether the same queries still get impressions and whether your rankings have changed.

Compare the same page, query and device across matching periods. A drop in total traffic can hide steady visits to one article.

A source link is worth opening

An answer can sound convincing and have a link underneath it. Neither tells you whether the linked page supports what you just read.

NIST's AI risk profile describes confident false answers as a risk. Guessing whether the prose sounds human won't help you check a disputed number.

The Tow Center's 2025 tests found fabricated links and incorrect attributions when tools were asked to identify news sources, showing why the link itself needs checking.

The task and tools were specific to that study. Its findings are a reason to inspect citations, not a current error score for every model.

• Open the source and find the passage that supports the claim.
• Check when it was published and who or what was studied.
• Find another source with its own evidence, not a copy of the first.
• Recalculate the percentage or run the code you intend to use.
• Leave the claim unverified if the evidence doesn't hold up.

The habit to watch is skipping the source

You can use AI every day and still check its work. The question is what you do after the answer appears.

After you get an answer, do you open the study it cites? Or does the neat explanation feel like enough to move on?

A Microsoft Research study found that workers who had more confidence in AI reported less effort on critical thinking when they used it.

The researchers asked workers to describe their own effort. They didn't test whether those workers had lost the ability to think critically over time.

NIST also flags overreliance as a risk. Accept an answer without checking it, and you can carry its mistakes into your own work.

Our guide to chatbot training and hallucinations explains why a model can produce fluent text and still invent a fact, even when you ask a clear question.

Let the assistant find documents and explain unfamiliar terms. For a decision that matters, read the relevant passage yourself before passing the answer along.

Good work can still be hard to find

You finish an article, publish it and wait. A few visits trickle in, and you wonder whether writing it was worth the effort.

Start by checking whether it is indexed. Google describes crawling, indexing and serving results as separate stages, and publishing a page does not guarantee any of them.

Google also uses links to discover pages and judge relevance, so an article linked from related pages has a route to discovery beyond its own search listing.

Few visits mean few chances for readers to return, share your article or try your tools, which makes the effort harder to justify.

That is a business problem, but a low click count does not explain a low rank. Check the topic, competing pages and indexing status before deciding what is wrong.

An AI citation might introduce your site's name to someone. If they never visit, you have no on-site page view to show an ad against.

Google can cite a page below the first ten results

A page can rank below the first ten blue links and still appear as a source in an AI Overview, as the citation data shows.

In Ahrefs' March 2026 study, 37.1% of cited URLs were among the first ten organic blue links for the same query.

Another 26.2% ranked from 11 through 100, and 36.7% were outside the top 100, so citations clearly extended beyond the first ten results.

Every page in that count already had a citation. Pages that got none were left out, so it can't tell you a new site's chances.

Google's AI search guide explains that the system can search related subtopics, which helps explain how it can find sources beyond your initial results.

Your page needs to be indexed and eligible for snippets. Your site also needs to be included in Search AI features through Search Console.

Google's AI features documentation makes no promise that meeting the requirements will get your page into an answer:

Indexing and serving isn't guaranteed.

Finding your page to answer a question is also different from training a model on it, so a citation doesn't establish what went into training.

Make the visit worth more than the summary

If you write developer guides, show a bug the reader can reproduce, the input that caused it and what happened when you changed the code.

That gives someone a reason to visit and test your explanation. They can compare your result with their own, instead of trusting a summary.

Our JSON syntax checker offers that kind of test, with a clear limit: it checks syntax, while the facts inside the JSON still need verifying.

Google's AI search reports show AI appearances separately. Google says the reports reached websites worldwide by August 31, 2026, so check yours alongside your clicks.

• Check indexing and AI inclusion before rewriting a page that never appeared.
• Compare the same query, page, country and device when clicks change.
• Count AI appearances and website visits separately in your traffic notes.
• Publish your test inputs and results so readers can repeat the work.
• Check tool use and earnings to see whether those visits paid off.

Adding a useful example won't guarantee more traffic. If ads fund your site, compare the visits and earnings before and after the change.

Before using your next AI answer, open one cited source and check a number against the original table or calculation.