Is AI Search Killing SEO, or Just Changing the Scoreboard?

Short answer: AI search is not killing SEO, but it is quietly rewriting the rules of the game. The click loss is real and independently documented. Pew Research found people clicked a traditional result on only 8% of searches that showed an AI summary, against 15% without one, and Ahrefs measured the top-result click penalty deepening to 58% by December 2025. But SEO is not disappearing. What is changing is the scoreboard. A top ranking is worth less, discovery is moving into surfaces that GA4 and Search Console do not report cleanly, and a small but fast-growing AI referral channel is appearing that most publishers are still not counting properly. The job of SEO is shifting from winning the click to being the source the answer is built on, and then measuring the traffic that survives.

Is AI search killing SEO?

No, but it is compressing the thing SEO has always chased: the value of a high ranking. The mechanism is simple. When Google answers the query inside the results page, the ranking still exists and the impression still fires, but the click never happens. Your position holds and your traffic falls. That is the signature of the changing scoreboard, and there is now solid primary evidence for it.

The strongest neutral study comes from Pew Research, using real browsing data from 900 US adults across 68,879 unique Google searches in March 2025. On searches that surfaced an AI summary, users clicked a traditional link only 8% of the time, compared with 15% when no summary appeared, roughly half. Only 1% of visits to a page with a summary produced a click on a link inside that summary. So the answer box does not redistribute the click; for the most part it removes it.

Pew also found that AI summaries make people stop looking. 26% of visits to a results page with an AI summary ended the browsing session entirely, against 16% without one. And the summaries are built to satisfy: they cited three or more sources 88% of the time, at a median length of 67 words. That is a complete, sourced answer inside the results page, which is exactly the zero-click outcome publishers fear.

This is not a fringe feature you can wait out. On Google's own numbers, AI Overviews reached 2 billion monthly users across 200 countries and territories by Q2 2025, up from 1.5 billion in May, with AI Mode passing 100 million monthly users in the US and India. Google also says AI Overviews drive more than 10% more queries for the query types where they appear. For informational search, the AI answer is now a default surface.

Ahrefs put a number on the damage over time. Across 300,000 keywords using Search Console data, the presence of an AI Overview correlated with a 34.5% lower click-through rate for the position-one page in April 2025, deepening to 58% by December 2025. In raw terms, position-one CTR for informational keywords fell from 0.076 in December 2023 to 0.016 in December 2025. The effect is worst at the top and decays down the page, from roughly a 50.8% penalty at position two to 19.4% at position ten. So the pages that used to win the most are losing the most.

The counter-argument you should hear

For balance, Google disputes the Pew study directly. It calls the study's methodology flawed and its queryset unrepresentative of Search traffic, says it has "not observed significant drops in aggregate web traffic", and argues that clicks arriving through AI features are higher quality. You do not have to pick a side to act. Both things can be true at once: aggregate clicks may look steadier than any single study suggests, while clicks to informational content specifically are down and the surviving visits may be worth more. That is a scoreboard change, not a death.

Is SEO still worth doing in the age of AI search?

Yes, for two reasons. First, being in the index is the precondition for being in the answer. AI Overviews, AI Mode and standalone AI search products build their responses from pages they can crawl and rank. If you are not visible in search, you are not eligible to be cited in the AI answer that replaced the search. The skills are largely the same, even where the payoff has moved.

Second, the alternative channel is growing quickly. Similarweb found AI platforms drove over 1.13 billion referral visits to the top 1,000 websites in June 2025, up around 357% year on year, with ChatGPT accounting for more than 80% of them. ChatGPT referrals to news sites specifically went from under 1 million between January and May 2024 to more than 25 million in the same months of 2025, a 25-fold jump. And the quality is notable: Similarweb clickstream data puts ChatGPT referral conversion at around 7.1%, second only to paid search. So the new channel is small, but it is real and it converts.

Here is the honest counterweight, because pretending otherwise helps nobody. That growth does not yet fill the hole. Over the same period, news organic search traffic fell from a mid-2024 peak of over 2.3 billion visits to under 1.7 billion, and zero-click news searches climbed from 56% to nearly 69% by May 2025. A 25x rise in AI referrals is still small against roughly 600 million lost search visits. SEO is worth doing, but do it with clear eyes: you are working a shrinking primary channel and a growing secondary one at the same time.

What replaces the ranking as the thing to optimise for?

The ranking is no longer a reliable proxy for traffic, so you need a new target. Two things replace it: citation and value.

Citation and inclusion. The unit of visibility is no longer "am I position one" but "am I one of the sources this AI answer is built from". Pew found AI summaries cite three or more sources 88% of the time. Being one of those cited sources is the new position one, because it is the thing that can still send a click, earn a brand mention, or shape the answer even when no click follows. Informational and question-shaped content is where this concentrates: Pew found AI summaries appeared on 60% of searches that began with a question word, versus just 8% of one-to-two-word searches. So the risk, and the opportunity, sit squarely on your explanatory, answer-style pages.

The value of each source. If clicks fall but AI-referred clicks convert at around 7.1%, then a raw session count hides the story. The metric that matters is what each channel is worth. A thousand AI referrals that convert at 7% can beat several thousand thin organic sessions. Optimising for the ranking alone will lead you to defend traffic that no longer pays, while ignoring a channel that does.

There is also a control decision that used to be invisible and now matters: what you let AI crawlers do. The crawlers split cleanly by purpose, and blocking one does not necessarily block another.

The nuance that trips people up: blocking training does not block search visibility, and blocking indexing does not stop a user-triggered fetch. Per Perplexity's own docs, its indexing crawler PerplexityBot respects robots.txt while its live fetcher Perplexity-User generally ignores it, because a person asked for that specific page. The same pattern holds at OpenAI (GPTBot and OAI-SearchBot are automated crawlers, ChatGPT-User is user-triggered) and Anthropic (ClaudeBot crawls, Claude-User is user-triggered).

Google's control is separate again. Google-Extended is not a crawler and not a user-agent; it is a robots.txt token that governs only whether your content trains and grounds Gemini. It does not affect Search inclusion or ranking, and it does not remove your eligibility for AI Overviews. So you can opt out of Gemini training while keeping full Search and AI Overviews visibility. The training decision and the visibility decision are genuinely separable, and worth making deliberately rather than by accident.

One practical note on verification, because a user-agent string alone is spoofable. OpenAI, Anthropic and Perplexity all publish machine-readable IP-range files (for example Perplexity's perplexitybot.json and separate perplexity-user.json). Correct identification is a user-agent match plus a source-IP check against those ranges, with reverse DNS as a secondary method. That is the backbone of any honest audit of what is actually taking your content.

What metrics show whether my AI-search visibility is improving?

This is where the "changing scoreboard" argument stops being rhetorical and becomes an urgent measurement problem, because by default your analytics is not scoring the new game.

The GA4 blind spot. Historically, GA4 dumped ChatGPT and Perplexity referrals into Direct or Referral, so AI traffic was invisible as a channel. Google only added a native "AI Assistant" channel to GA4's default channel group on 13 May 2026, naming ChatGPT, Gemini and Claude at launch and later folding in Copilot, Grok and DeepSeek. Perplexity is still absent from Google's definition, so its visits keep landing in Referral. If you have not built a custom channel group, you are undercounting your AI traffic and missing Perplexity entirely. The fix is a Custom Channel Group that matches source contains chatgpt.com, perplexity.ai and the other AI domains, ordered ahead of Referral and Organic so those visits are not swallowed by a broader rule.

The Search Console blind spot. Google has confirmed that AI Overview appearances are counted in Search Console impressions, and clicks through an Overview are counted as clicks. But Search Console does not break AI Overviews out as their own dimension. If the same URL appears in both an AI Overview and the blue links on one search, it counts as a single impression at a single position, and the clicks that never happen, because the answer was given in the summary, simply do not appear anywhere. The net effect is the mechanical signature of the changing scoreboard: impressions look flat or even up while clicks fall. If you only watch rankings and impressions, everything looks fine right up until revenue drops.

So the metrics worth tracking now are:

• AI referral sessions by engine. ChatGPT, Perplexity, Gemini, Copilot and Claude, tracked as a distinct channel, not lumped into Referral. This is the growth number, and it is the one your default GA4 still hides for Perplexity.
• Value per source. Conversion rate, subscriptions or ad revenue per session, split by channel. Given AI referrals can convert around 7.1%, a source can be small in volume but large in value.
• CTR on informational queries with AI Overviews present. Watch the gap open between impressions (flat or up) and clicks (down) on your explainer content. That divergence is the tax, and it is where Ahrefs measured the deepest penalty.
• Which AI crawlers are hitting you, verified by IP. Training versus search versus user-triggered, so you can see whether you are being indexed and fetched for live answers, not just scraped for training.
• Citation frequency. How often AI answers are built on your pages. This is the new position one, and it is the leading indicator of AI referral traffic to come.

A worked example

Say an informational page held position one all year. In Search Console, impressions rose 5% and average position stayed at 1.2. Nothing looks wrong. But sessions to that page fell 40%, because an AI Overview now answers most of those searches inside the results. Meanwhile a "Referral" line you never examined turns out, on inspection, to be Perplexity sending small but high-converting traffic that GA4's native AI channel still does not classify. On the old scoreboard, this page is a winner. On the real one, it is quietly leaking your most valuable visits into a channel you are not counting. You cannot manage what you cannot see, and the default tools are still built for the old game.

This is the practical gap Ramprt is built to close. It reads your existing GA4 read-only, with no new tracking script, and gives you a clean view of AI referrals by engine, which AI crawlers are taking your content, how often AI answers cite your pages, and what AI-referred traffic earns against your site average. If you want to see what AI is actually doing to your traffic before you decide anything, try the free live demo.

Frequently asked questions

Is SEO dead because of AI Overviews?

No. AI Overviews compress the value of a top ranking on informational queries, but being crawled and ranked is still the precondition for appearing in AI answers. The evidence shows click loss (Pew: 8% versus 15%; Ahrefs: up to a 58% CTR penalty on the top result) not the end of search-driven discovery. The work shifts from winning the click to being the cited source.

Do AI referrals make up for lost search clicks?

Not yet. Similarweb measured AI referrals up around 357% year on year in June 2025, and ChatGPT referrals to news sites rose 25x, but news organic search still fell from over 2.3 billion visits to under 1.7 billion, with zero-click searches near 69%. The new channel is real and high-converting (around 7.1% for ChatGPT) but too small to offset the loss so far.

Should I block AI crawlers in robots.txt?

It depends on your goal, and the crawlers split by purpose. Blocking a training crawler like GPTBot or ClaudeBot does not remove you from AI search or Google Search. Google-Extended only governs Gemini training and does not affect Search or AI Overviews eligibility. Note that user-triggered fetchers like Perplexity-User and ChatGPT-User often ignore robots.txt because a human requested the page.

Why does my GA4 not show ChatGPT or Perplexity traffic?

Because by default GA4 put AI referrals into Direct or Referral. A native AI Assistant channel arrived on 13 May 2026 and now covers ChatGPT, Gemini, Claude and Copilot, but Perplexity is still excluded and lands in Referral. To capture all AI traffic you need a custom channel group that matches the AI source domains and sits ahead of Referral and Organic.

Does an AI Overview still count as a Search Console impression?

Yes. Google counts AI Overview appearances in impressions and counts click-throughs from them as clicks. But it does not break Overviews out as a separate dimension, and clicks that never happen because the summary answered the query do not appear at all. That is why impressions can look flat while clicks fall.