AI Overviews and the Disappearing Click

There is a version of this article that says AI has killed SEO. It gets a lot of traffic and it is wrong in a specific, expensive way.

What is actually happening is narrower and more interesting: the relationship between ranking and traffic is decoupling. You can hold your position and lose your clicks. That is a different problem from being outranked, and it needs a different response.

What the data shows

Start with the click-through rate curve, because that is where the effect shows up first.

GrowthSRC’s 2025 study tracked more than 200,000 keywords and found that average CTR across positions one through five declined by roughly 17.92% compared to 2024. Not rankings: click-through rates at those rankings. A site that held every position it had in 2024 would have lost meaningful traffic anyway.

On AI Overview prevalence, the figures cluster. For local business queries, BrightLocal-sourced analysis compiled by SeoProfy puts AI Overview trigger rate at 40.16%. Digital Applied’s 2026 compilation puts local trigger rate at 43% and estimates a 31% reduction in clicks reaching individual business websites from those results.

Those are large numbers and they are also not the whole picture, which is where most commentary goes wrong.

The part that gets left out

The same Digital Applied dataset contains a finding that complicates the doom narrative considerably: 67% of AI Overviews for local queries reference Google Business Profile data directly, and optimised profiles are recording 2.4x more profile views from AI-generated local recommendations.

So for local businesses, AI Overviews are shifting value from the website to the profile rather than destroying it. Clicks to the site fall. Views and interactions on the profile rise. If your measurement stops at website sessions, you will record this as a catastrophe when part of it is a reallocation.

That distinction matters for budget decisions. A business that responds to “AI Overviews cost us 31% of our clicks” by cutting search investment is reading the data backwards. The correct response for a local business is to invest more in the surface that AI systems are citing, which happens to be the free one.

What actually gets cited

The mechanism behind citation is more legible than most people assume, and it maps closely onto things that were already good practice.

Direct answers, stated early. AI systems extract passages that answer the question cleanly. A page that spends four paragraphs on preamble before reaching the answer does not get pulled. A page that answers in the first sentence and then earns the reader’s attention by going deeper does.

Specific, attributable facts. Numbers, dates, prices, named entities, measured results. Generic advice is not citable because it is interchangeable. There are four hundred pages saying the same thing and no reason to prefer any of them. A page with a figure that exists nowhere else is the only candidate for that fact.

Structure that survives extraction. Clear headings that pose the question, short self-contained paragraphs, tables for comparative data, lists where the content is genuinely a list. A passage that only makes sense with three paragraphs of surrounding context is a passage that will not be quoted.

Consistency across the site. If two of your pages give different figures for the same thing, neither is safe to cite. This is a real and underrated problem on sites that have published for years without an editorial process.

Notice that all four of those things also make the page better for humans. The optimisation for AI citation and the optimisation for a reader in a hurry converge almost completely, which is unusual and worth taking advantage of.

The commodity problem

The strategic frame that holds up best comes from Google itself. At Search Central Live in Toronto in April 2026, Danny Sullivan drew a distinction between commodity and non-commodity content, and it is the most useful lens available for planning right now.

Commodity content is information that exists in many places and can be assembled from general knowledge. “What is a sitemap.” “How to choose a hotel.” “Benefits of yoga.” An AI system can synthesise this from its training data without needing your page at all, and increasingly it does. Content in this category is not becoming less valuable gradually. It is becoming worthless quickly.

Non-commodity content is information that exists only because someone did something. Original measurement. First-hand experience. Proprietary data. A photograph of a thing you actually saw. A price you actually charge. A result you actually got.

An AI system cannot synthesise your first-party data because it does not have it. It has to cite you or omit the fact.

This is the entire strategic pivot, and it is more demanding than the tactics it replaces. You cannot outsource first-hand experience to a content mill. That is precisely why it works.

What this changes practically

Stop measuring only sessions. Add impressions, average position on your priority queries, brand search volume, and, if you are local, Google Business Profile views, calls and direction requests. A campaign that increases brand searches and profile calls while website sessions stay flat is working. Measured on sessions alone, it looks like failure.

Audit your content for commodity status. Go through your top pages and ask, honestly, whether the information on them exists elsewhere. Anything a competent writer could produce from general knowledge is a candidate for consolidation or deletion rather than refreshment. This is uncomfortable because it usually means removing pages someone paid for.

Build a first-party data habit. You have data nobody else has. Your pricing. Your project outcomes. What your customers actually ask. Seasonal patterns in your bookings. The failure rate you observe on a particular product. Publishing a small amount of that is worth more than a large amount of general advice, and it is the only defensible position left.

Fix the answer-first structure on existing pages. This is a rewrite job, not a new content job, and it is fast. Move the answer to the top. Add a table where you had prose. Make each section self-contained.

Invest in the surfaces AI systems cite. For local businesses that means the Business Profile, which 67% of local AI Overviews reference directly. For everyone it means being mentioned by sources the systems trust, which is closer to public relations than to link building.

Where the effect is smaller than you think

Two categories are much less exposed than the headline numbers suggest.

Transactional queries. Nobody buys from an AI summary. Queries where the searcher intends to purchase, book, or contact someone still route to a destination, and the click value on those queries has if anything gone up as informational traffic has dried up.

Local queries with intent to visit. The data supports this directly: near-me searches still convert at extraordinary rates, with roughly 76% of local searchers visiting a business within 24 hours and 28% making a purchase within the same window. An AI Overview can summarise which restaurants are nearby. It cannot feed anyone.

The traffic that is genuinely disappearing is top-of-funnel informational traffic, which, uncomfortably, is a large share of what most content programmes have been producing for a decade, and always the share that converted worst.

There is a case that this is a correction rather than a crisis. Traffic that never converted is now not arriving. That is painful for anyone whose reporting was built on sessions and fine for anyone whose reporting was built on revenue.

The reporting has to change before the strategy can

A practical problem sits underneath all of this: most marketing reports cannot represent what is happening, which means they generate the wrong decisions regardless of how good the strategy is.

A report built on organic sessions will show this shift as pure decline. Sessions fall because informational queries are being answered in the results. Everything else (brand awareness, profile activity, the quality of the traffic that does arrive) is invisible.

Four additions fix it.

Impressions alongside clicks, per query cluster. If impressions hold or grow while clicks fall, you are still being surfaced and the loss is to zero-click behaviour rather than to a competitor. That is a completely different problem from being outranked, and it calls for different work.

Brand search volume as a trend line. When people encounter you in an AI summary and later search your name directly, the credit lands in brand search, not organic. Rising brand volume alongside falling non-brand clicks is evidence the content is working, not failing.

Profile interactions for local businesses. Calls, direction requests and messages from the Business Profile. Given that 67% of local AI Overviews cite profile data and profile views from AI recommendations are up 2.4x, leaving this out of the report guarantees a misreading.

Conversion rate of the traffic that does arrive. If informational visitors are being filtered out upstream, the visitors who still click should convert better. Many sites are seeing exactly that: fewer sessions, higher conversion rate, flat or growing revenue. On a sessions-only report that reads as a crisis.

Until the report shows these, the strategy conversation cannot happen, because everyone is looking at a number that only moves in one direction.

The market context

None of this has reduced spending on search. Mordor Intelligence values the SEO services market at $83.98 billion in 2026, up from $74.9 billion in 2025, and projects $148.86 billion by 2031 on a 12.12% CAGR. Asia-Pacific is forecast as the fastest-growing region at 13.55%.

Money is not leaving the channel. It is moving from volume production toward the things AI cannot replicate: technical work, original research, and earning citations from sources that get referenced.

That shift favours practitioners who can do diagnostic and technical work over those who could only produce content at scale. It is also why an increasing share of buyers now evaluate a provider on what they found rather than what they propose to produce, a filter worth applying whether you are assessing a seo expert nepal businesses recommend, a London agency, or an internal hire.

The honest summary

CTR at the top of page one fell roughly 18% year over year. Around 40-43% of local queries now show an AI Overview, and those results cut clicks to business websites by about 31%. But 67% of local AI Overviews cite Google Business Profile data, profile views from AI recommendations are up 2.4x, and transactional and local-intent queries are largely unaffected.

The channel is not dying. Commodity content is. The response is to publish things only you could publish, structure them so they can be extracted, and measure something other than sessions.

Businesses without the internal capacity to make that shift are increasingly buying it in, and the useful filter when assessing an SEO Company in Nepal or any other provider right now is simple: ask what they would stop doing. Anyone whose 2026 proposal is identical to their 2023 proposal has not been reading the data.