Are AI Overviews Killing My Website Traffic? What the Data Actually Shows

Clicks are lower. There will be an AI block above your result. An article was published last week that “organic traffic is dead.” So, you’re asking the question on the nose: are AI Overviews reducing website traffic? The reality is that it depends on your questions, and no one can tell you your number. This guide will assist you in knowing how to check it yourself in an hour, and what to do with the information you discover.

This guide explains:

• How to read your own impressions-against-clicks pattern
• Why the published percentages contradict each other
• The hour-long measurement method, step by step
• Which of your query types are really exposed
• What else could explain the drop before you blame the feature

Are AI Overviews killing my website traffic?

Yes, for some websites. Many say no because the reduction is mainly in definition and how-to search, and local and buying traffic is not that much affected.

The table below is a diagnosis chart for your Search Console chart. Identify a pattern in the column that matches the one you see and read across.

The first row indicates the direction of AI Overviews traffic impact that people are concerned about. Other changes are similar. Treat AI Overviews as a lead, not a verdict.

It’s better to be in the block than to gauge the damage. Once the data is received, you can determine this. The other option, which our guide on getting inside the block instead, is to know what makes a page qualify for an AI Overview, and how a passage is chosen, so you can make a decision on whether or not you want to bother.

Why nobody can give you a percentage

All of the published studies are based on query samples, sectors and dates, and none of them will have access to your queries, so a number from a different study doesn’t mean a thing about your site.

One study of 1000 of the keywords in one sector is a reasonable estimate of those keywords and not a fact about the entire web. Any figure for the click-through rate of AI Overviews is a number concerning someone’s sample. Articles on “zero click searches 2026” mix samples that were never meant to be added together.

Google’s documentation is the only source on the first party. It’s good to read because it does not say. Google’s documentation on AI features and your website says sites in AI features count in search traffic in Search Console under the Web search type in the Performance report. Google claims that Google has witnessed better click-throughs to pages that contain AI Overviews. Google doesn’t offer a percentage or separate filter.

For owners looking to show less, they aren’t referring to the AI Overviews switch, Google says. So, this is where we are. We never publish a percentage, as we can’t defend a percentage. No one can quote a percentage of you, either.

How do I measure the impact on my own site?

Compare query-level impressions. Clicks for the same 90 days, broken down by query type. Identify situations where CTR is diminishing. Impressions and position stay the same. First, know the limits. Google includes the number of times AI features appear in the Performance report. No filter to separate them. A separate AI report is being rolled out to some properties. It only displays impressions per page, not clicks. All the following methods are impressions and click-based estimates. Not a measure.

Follow these steps in order. None requires a paid tool.

1. Export query-level data for the past 90 days and the past 90 days one year ago.

2. Tag each query by type using the table below. Don’t worry about being perfect. Rough grouping is fine. The goal isn’t precision. It’s spotting patterns.

3. Calculate click-through rate per query type for both periods. That’s clicks divided by impressions.

4. Note the position per query type for both periods. If position changes, the comparison doesn’t hold.

5. Focus on groups where position stayed steady, and click-through rate went down. That’s your exposed traffic. The queries where AI answers are likely taking clicks.

6. Check conversion or inquiry trends for the period broken down by landing page.

Three rules help you interpret the result: Position must be steady. The comparison means nothing. One quarter’s data is a signal, not a conclusion. Also add annotations for dates of known algorithm updates before drawing any lines. Running this once is a diagnosis. Making it part of reporting is what a mature SEO program does. Our approach to search work that accounts for answer results builds this comparison into reporting. That way you see exposure by query type before it shows up in revenue or in your sales calls.

The only search method covered by this method is search only. Assistant visits, such as ChatGPT, are reported. You should also inspect those. Our companion guide addresses the referral side of the same question and how to locate the assistant-driven visits in your analytics, so you can use that data to compare with the search data.

Which of my queries are most exposed?

The most exposed queries are definition, explainer and simple how-to queries, while local, buying and navigation queries are least exposed.

The table below shows relative exposure of the query types only. There are no percentages used in it, as it is a comparative ranking.

A site with mostly definitions and other informational intent is exposed, while a site with mostly local service intent largely is not. That is why one industry figure is meaningless.

Talk of zero-click behavior is really talk about the top rows of this table.

Health content is the clearest real case of a site whose traffic is mostly explanation, so it is a good profile to compare yourself against. Our work with practices whose traffic is mostly informational starts by splitting queries into the types above. We then protect the ones that bring in appointments and calls, not just readers.

What else could explain the drop?

A core update, a seasonal cycle, a tracking change, a site move or a better competitor page all make the same shape in a chart, and at least one is usually in the same window.

• A core update: check the dates of known updates against the date your line bent.
• A seasonal cycle: compare the same weeks last year before you compare last month.
• A tracking or consent change: check whether your analytics tag or consent banner changed.
• A site migration: check redirects and whether the old URLs are still indexed.
• A competitor with better pages: check who now ranks above you for the lost queries.

A Google AI Overviews traffic drop can sit on top of any of these, so a drop that matches a rollout is not proof of cause.

Google warns that nobody can promise a top ranking, and it points to a separate guide on debugging traffic drops. You can read it in Google’s guidance on whether you need an SEO before anyone sells you a diagnosis. That page sets realistic expectations for what search work can and cannot do for a business like yours.

A business that blames the feature and stops looking will miss a cause it could fix. Our SEO audit helps you find out what actually changed by checking updates, tracking, migrations and competitors against your own data. Your next step then rests on evidence, not on a headline.

If the loss is real, what actually works?

Move effort toward the query types that still produce clicks, rebuild pages so you are cited and not replaced, and build demand that does not need a search result.

If your data points to an organic traffic decline that AI features explain, there are three strands.

The first strand is page structure, because a page that answers the question in a self-contained passage can be cited, while a page that only describes the company gets replaced. Our on-page work on pages rebuilt around the questions people ask puts the answer first under each heading. The page then earns a citation and still gives the reader a reason to click.

Google rates the content that competes for the remaining clicks against a people-first standard, so it helps to know that standard. You can read it in Google’s guidance on creating helpful content. It describes content made for people, not to game rankings, and it gives questions you can use to check your own pages before you publish.

The second strand is demand that does not depend on a result page, such as direct visits, referrals, and repeat readers. Our content work on building demand that does not depend on a click focuses on pages and series people return to by name. That is slower to build, but it does not rely on one search feature.

The third strand is the hedge. Paid traffic buys time, but it does not fix your exposed query mix. Our work on covering the gap while organic recovers uses paid search honestly, as a bridge with a budget and an end date, while the organic fixes above take effect.

What should I stop worrying about?

Traffic that never converted is not worth mourning, and a site whose inquiries are steady while sessions fall has lost readers, not customers.

This comfort is earned, not offered, because if the clicks that vanished never turned into calls or bookings, the number to protect is the one your business runs on. Our landing page work on making the clicks you still get count; we build each page around one action, so a smaller audience converts at least as well as a larger one did.

Search results have absorbed clicks before, through featured snippets and map results. Sites adapted by moving effort to the queries that still produced clicks. Nobody can say if this will get worse, and anyone who says so is guessing.

Some businesses did build their model on informational traffic. For them this is a real structural problem, not a measurement quirk.

What the data on AI Overviews and traffic actually shows

The short answer to “are AI Overviews reducing website traffic” is that it varies by query type, and only your own data can answer it.

1. It varies by query type.

2. Your own data is the only data that answers your question.

3. Position must be held steady before you draw any conclusion.

4. Inquiries matter more than sessions.

5. One quarter is not a trend.

Measure before you react, and compare against a baseline from the same period last year. Want to see how this works on your own data? If you would like a specialist to run the comparison with you, Pink Socials will walk you through your query mix, which groups look exposed, what to do first and how to check the result next quarter.