AI Overviews and the New Math of Search Clicks

For most of the last two decades, the deal with search was simple. You ranked for a query, you earned a click, and the click had a roughly predictable value. You could estimate traffic from a ranking position, estimate conversions from traffic, and estimate revenue from conversions. That chain held well enough to plan a content budget around it.

AI Overviews broke a link in that chain. When Google answers the question directly at the top of the page, the ranking position still exists, but the click that used to come with it does not always follow. The economics you were planning around have quietly changed underneath you. This piece is about what actually changed, which pages are exposed, and how to think about content spend when a portion of your rankings no longer sends traffic.

What changed in the click chain

The old model treated a top ranking as a proxy for traffic. Position one meant a large share of clicks, position five meant a smaller share, and so on down the page. Those share estimates were stable enough to forecast against.

AI Overviews insert an answer layer above the traditional results. For a query like "how often should you replace an HVAC filter," the engine now often composes an answer on the page itself, pulling from several sources, and shows the ten blue links below it. The searcher reads the answer, gets what they came for, and never scrolls. Your page may still be cited inside that answer, but a citation is not a click, and even when it is a click, it arrives after the person already has their answer.

The practical result is a split. Some queries now leak most of their clicks to the answer box. Others are barely affected. The difference is not random, and understanding the pattern is how you decide where to keep investing.

Which queries lose clicks and which keep them

Queries lose clicks when the answer is short, factual, and complete enough to satisfy the searcher in a sentence or two. These are the questions where the whole intent is "just tell me." Definitions, unit conversions, simple how-often and how-much questions, business hours, quick specs. If an AI Overview can fully resolve the query in three lines, the click is gone whether you rank or not.

Queries keep their clicks when resolving them requires something the answer box cannot deliver: a decision, a comparison specific to the searcher's situation, a purchase, a form, a phone call, or trust in a named provider. "Is a cracked heat exchanger dangerous" might get summarized. "Emergency HVAC repair near me" will not, because the intent is to hire someone, not to read a paragraph.

A rough way to sort your own keywords

• Fully answerable in text — high leak risk. Informational, factual, no decision attached.
• Answerable but leads to a decision — partial leak. The overview gives context, but the searcher still needs to choose, compare, or verify.
• Not answerable in text — low leak risk. Transactional, local, "near me," quote requests, anything tied to hiring or buying.

Run your top traffic-driving pages through this sort. The pages sitting in the first bucket are the ones whose economics changed the most. The pages in the third bucket are largely untouched, and in some cases better off, because competing thin content is getting absorbed while your conversion-intent pages still get the click.

How do you tell if AI Overviews are actually costing you traffic?

You confirm it by pairing Search Console impression and click data with a manual check of whether an Overview appears for the query. Pull your top queries by impressions in Google Search Console over the last six to twelve months. Sort for pages where impressions held steady or rose while clicks fell — that gap is the signature of an answer layer capturing attention before the click. Then take the ten or fifteen queries showing the widest gap and search them yourself, ideally logged out and in an incognito window, to see whether an AI Overview renders. If the queries with falling click-through are also the ones showing an Overview, you have your answer. If clicks fell without any Overview present, the cause is something else — a ranking drop, seasonality, or a SERP feature like a featured snippet or a local pack.

Do this per page, not in aggregate. Site-wide click-through can look flat while a specific cluster of informational pages quietly bleeds out and a set of transactional pages holds firm.

A worked example of the new math

Suppose a plumbing company in Tucson has a blog page ranking well for "how to tell if you have a slab leak." Before, that page brought in a steady stream of visitors, and a small fraction of them eventually booked an inspection.

Here is how the same page might perform after an Overview starts answering the query directly:

Impressions actually rose, so by a ranking report this page looks healthy. But clicks dropped by more than half because the answer box handled the informational part of the query. These numbers are illustrative, not a measured result — the point is the shape of the change, not the exact figures.

Now compare the company's "slab leak repair Tucson" page. That query carries hiring intent, no Overview reliably satisfies it, and its click-through barely moves. Same site, two pages, two completely different exposures. Budgeting as if both pages behave the same is the mistake.

What should you do about content that's losing clicks to AI Overviews?

Shift your investment toward content the answer box cannot replace, and rework your exposed informational content so it either earns the citation or feeds a decision. Concretely:

1. Protect and expand decision-and-hire content. Comparison pages, pricing explainers with real ranges, location pages, service pages, and case-style write-ups tied to a specific problem and outcome. These convert and they don't summarize away.
2. Re-scope thin informational posts. A page that only answers "how often to service a furnace" in 300 words has little defense. Rebuild it to cover the follow-up decisions — what a service visit includes, what neglect actually costs, when to repair versus replace — so a reader who wants more than the one-line answer has a reason to click through.
3. Format for citation. If a query will get an Overview no matter what, being the cited source is the consolation prize, and it still carries some traffic and authority. Clear direct answers, structured lists, and specific numbers get pulled more readily than hedged prose.
4. Measure by contribution, not clicks alone. Track which pages sit upstream of form fills and calls, not just which pages get the most visits. A page with falling clicks that still assists conversions is worth keeping; a high-traffic page that never assists anything was never earning its keep.

This is roughly how we plan editorial programs at ClearPath Content — map the question-space, then weight production toward the questions that still send a click and still lead to a booking, rather than chasing informational volume that the answer layer now eats.

The takeaway

Rankings and traffic have partly decoupled, and the split runs along intent. Informational queries that resolve in a sentence are leaking clicks; decision, local, and hiring queries mostly are not. Audit your top pages against that line, move budget toward content the answer box can't replace, and measure pages by whether they assist real conversions instead of raw visits. The businesses that adjust their content mix to this reality keep growing while the ones still counting impressions wonder why the numbers stopped adding up.