Why Google Needed AI Overviews, and What It Owes the Web

This is an opinion piece. It reflects the personal views of Tahir Dinç.

I've worked in search since 1999, and for years I sorted websites by hand as a volunteer directory editor. So here's a view that may sound odd coming from an SEO: I don't think Google ever wanted AI Overviews. I still think they're worth backing, as long as Google keeps its side of a very old deal.

Most of the debate is about lost clicks. That matters, and I'll get to the numbers. But the more useful question is what Google and the open web owe each other now. If you're after the practical side, how AI Overviews work and how to get cited in them, start with our AI Overviews guide.

What Was the Original Deal Between Google and the Web?

For more than two decades, Google's relationship with websites ran on a plain exchange: access for visitors. I saw that exchange up close before most people knew it existed.

In 2004 I became an editor at the Open Directory Project, the volunteer directory everyone called DMOZ. We reviewed sites one at a time, kept the good ones and threw out the spam. Nobody was paid. We did it because we wanted a better web.

That volunteer work flowed straight into Google. Its own directory, Google Directory, ran on DMOZ data until Google shut it down on July 20, 2011, and DMOZ itself closed in March 2017. We sorted the web by hand. Google learned to do it at scale. Its mission statement describes the same job.

The business side of the deal was just as clear. On June 26, 2000, Google became Yahoo's default search provider, so Yahoo users were quietly reading Google results. In 2003 AdSense arrived: Google placed ads on other people's sites and split the revenue with them. Google got paid twice: when you searched, and again when you landed on someone else's site. When Yahoo moved to its own engine in February 2004, Google was already the name everyone knew, and it went public that August.

For two decades, that arrangement paid everyone. The more people clicked through to websites, the better Google, publishers and advertisers did. You can read how it started in our piece on BackRub, Google's first name, or follow the whole story in our Google timeline.

What Changed When AI Overviews Arrived?

AI Overviews put the old deal under strain, because they answer the question on Google's own page instead of sending the reader out. Google launched them in the US on May 14, 2024, with a promise that links inside the summaries would get more clicks than a regular listing. A search that ends at the summary is a zero-click search, and we've looked at what zero-click search is doing to the web.

I don't believe Google wanted to go this way. Its whole business grew by sending people outward. Keeping them on the results page cuts against the thing that made Google Google. So why do it? Because three pressures had been building for a while, and what happened after the launch showed how real they were.

• Chat assistants took off. ChatGPT arrived in late 2022, and people liked getting one clean answer instead of a page of links. By February 2026, OpenAI was reporting 900 million weekly users. Even Apple saw the effect: in May 2025, Eddy Cue told a federal court that Google searches in Safari had dropped the month before, the first decline in 22 years. Google replied the same day that its total queries were still growing. When Judge Amit Mehta ruled on remedies in the search antitrust case that September, he wrote that the emergence of generative AI had changed the course of the case.
• Result quality had slipped. Well before the launch, researchers in Germany tracked 7,392 product review queries for a year and found every major search engine losing ground to pages built for search engines rather than readers. Their paper was titled "Is Google Getting Worse?" Google's March 2024 update, by its own count, cut low-quality, unoriginal content in results by 45%.
• People wanted direct answers. Chat had already shown the appetite, and Google's own numbers after the launch backed it up: the kinds of queries that trigger AI Overviews grew more than 10% in the US and India, and Google now reports more than 2.5 billion monthly users.

The launch was rough. The summary that suggested glue for pizza made the rounds, and Google admitted the mistakes within weeks. It pressed ahead anyway. Standing still meant watching users drift to tools that send no traffic back at all.

Is Google Losing Money on AI Overviews?

Google isn't losing much money on AI Overviews in the short run, because most of its ad revenue is earned on its own results pages. I've managed Google Ads campaigns for clients for most of my career, so I know where the money is made: on the results page itself. The 2025 numbers make the same point.

Google also runs ads above and below AI Overviews. The money that depends on people clicking out (AdSense and the rest of the network) is the smaller slice, and it's shrinking.

What worries me is further out. Every AI Overview is built from pages somebody else wrote. If writing those pages stops paying, fewer get written, and next year's summaries have thinner material to work with. In other words, when Google sends traffic to websites, it's protecting its own supply chain.

What Do Google and Independent Data Say About Clicks?

Google says its AI features still send a lot of traffic to websites, while independent data shows fewer clicks when a summary appears. I put the claims side by side so you can judge for yourself.

Question	What Google says	What independent data and records show
Do sites still get traffic?	AI features in Search send billions of clicks to websites every week (July 2026).	Google doesn't publish click data. The Search Console report for AI features counts impressions only.
Do summaries lower clicks?	Links in AI Overviews get more clicks than a regular listing (May 2024).	Pew Research Center: a regular result was clicked on 8% of visits with a summary and 15% without one. Only 1% clicked a link inside the summary (July 2025).
Is search growing?	Query types that show AI Overviews grew more than 10% in the US and India (May 2025).	In the antitrust case, Google's head of Search, Liz Reid, testified that overall US query growth since the launch was about 1.5% to 2%.
What about news sites?	No separate figure.	Chartbeat data in the Reuters Institute's Digital News Report 2026: Google organic traffic to US news sites fell 38% in a year.
How healthy is the open web?	Google says it keeps sending billions of clicks.	In a September 2025 court filing in its ad tech case, Google wrote that "the open web is already in rapid decline." It later said it meant open-web display advertising.

My read of the table: Google sends less traffic out than it used to, and it quotes its best numbers. But the traffic hasn't stopped, and everything else in this piece depends on that.

What Does Google Owe the Open Web?

Google owes the open web four things in the AI era: traffic, credit, a real choice and a share of the value. Here's how I'd rate it on each in 2026.

1. Traffic. Keep sending readers out: through the source links in each summary, the regular results and the ad network. Google also added Preferred Sources, which lets readers flag outlets they trust so those sites show up more in AI Overviews and AI Mode (Google's conversational search). It rolled out globally in English in December 2025, and people picked around 90,000 sources. My rating: working on it. Traffic still flows, at lower rates, and only Google's own count is public.
2. Credit. Show clearly whose work an answer is built on, with a visible link to the source. My rating: working on it, with a push from regulators. In June 2026, the UK's competition regulator ordered Google to attribute publishers with clear links in its AI results.
3. A real choice. Let site owners decide whether their work feeds the summaries. For two years there was no clean way to do it; Judge Mehta wrote that publishers were "caught between a rock and a hard place." Since August 31, 2026, the "Search generative AI" control in Search Console lets any site opt out. It came out of the same June 2026 UK order. My rating: delivered, also with a push from regulators, and with a cost: opted-out sites get no traffic from those features.
4. A share of the value. Pay the people whose work makes the answers good. In December 2025, Google started a paid AI pilot with a small group of publishers, including The Guardian, The Washington Post and Der Spiegel. My rating: barely started. A handful of big publishers is not the open web.

That isn't a perfect score. But every item on the list points readers back toward the web, and that's the test I care about.

Do AI Assistants Send People Back to Websites?

In my experience, most AI assistants make it hard to reach the websites behind their answers. Try it yourself. Ask one for a list of companies that do what you need. You'll get a tidy list. Now try to visit those companies. Often you have to go looking for the sites yourself, because the assistant would rather keep you in the chat.

This isn't about one company. It's a pattern across most of the tools I try, though several have gotten better this year. The rule I'd hold all of them to is plain: if an answer is built on someone's work, the path to that work should be one click away. Google's results still send you to the source, and that's the bar.

Why Should Anyone Keep Publishing?

The open web only keeps working if people keep publishing, and AI depends on that more than anyone. Here's why.

I think of the web as a four-step loop: people publish, Google organizes, readers click through, and the work pays off, which gives people a reason to publish again. The diagram shows where it can break.

If you run a website, your job is to protect step three. Publish what a summary can't give: first-hand experience, original data, working tools and fresh local detail. Those are the pages most likely to earn the click and get cited in AI Overviews. Our GEO page (generative engine optimization) explains how we work on that, and our AIO page lays out the wider framework.

My Bottom Line

As long as Google keeps sending people out to the web, AI Overviews are the version of AI search worth backing. A closed chat window keeps you inside. Google's summary still points at its sources, and when you ask a follow-up in AI Mode, the trail can still lead back to the original pages. That's the old deal in a new form, and it's worth protecting. If Google stops keeping its side of it, my view changes.

Frequently Asked Questions

What Is Google's Mission?

Google's mission is to organize the world's information and make it universally accessible and useful. In 2023, Google said AI had become more critical than ever to that mission.

Why Did Google Create AI Overviews?

Three pressures pushed Google there: chat assistants were drawing users away, low-quality pages had crowded search results, and people clearly wanted direct answers. AI Overviews launched in the US on May 14, 2024.

Are AI Overviews Powered by Gemini?

Yes. AI Overviews run on Gemini models customized for Google Search. They're a Search feature, not the Gemini app; our Gemini guide covers the difference.

Can I Turn Off AI Overviews?

If you're searching, no. Google's help page puts it bluntly: "Features cannot be turned off." The closest workaround is the "Web" filter, which you can pick after a search to get plain links. If you own a website, you have a real switch in Search Console, and the trade-off is losing every impression and visit those AI features would have sent.

Do AI Overviews Reduce Website Traffic?

For some pages, clearly yes. In Pew's data, people clicked a regular result on 15% of visits without a summary and 8% with one. Quick-answer pages take the biggest hit. Pages built on experience, original data or tools hold up better, though a citation by itself doesn't guarantee a visit.

Official and Academic Sources