AI Search Cannibalization: The Organic Traffic Collapse, by Industry

In May 2024, Google rolled out AI Overviews at I/O. The feature appeared on 13.1% of searches that month, summarized answers above the organic results, and provoked a familiar round of SEO panic. Two years later, the panic looks restrained. AI Overviews now appear on 38.4% of US Google searches, according to Pew Research's May 2026 audit. The ChatGPT Atlas browser, launched in February 2026, processes roughly 412 million weekly active queries that previously would have gone to Google. Perplexity's Comet browser is at 78 million WAU. And SparkToro's clickstream data shows that 67.4% of Google searches now end without a click to any external website.

The collapse is real. It is also not what most SEO commentary suggests. Organic traffic to "the web" did not fall off a cliff. Organic traffic to a specific kind of website -- the informational, middle-of-funnel content site that won SEO from 2015 to 2024 -- has been gutted. Other categories are roughly flat. A few are growing. The damage is wildly uneven, and the unevenness is the story.

This article maps the cannibalization vertical by vertical, using Pew Research, SimilarWeb, SparkToro, Ahrefs, and Semrush data through Q1 2026. It also makes a contrarian argument: the sites that owned proprietary data, communities, or transactional intent are quietly consolidating share while their middle-of-funnel competitors implode. The losers are not "publishers." The losers are a specific business model.

The Great Decoupling, Quantified

The cleanest signal of AI search cannibalization is what SEOs have started calling the great decoupling: a widening gap between impressions and clicks in Google Search Console. Impressions reflect how often a site shows up in the SERP. Clicks reflect how often someone actually visits. Historically these moved together. Since AI Overviews launched, they have separated.

Ahrefs analyzed Search Console data from 12,094 websites between January 2024 and March 2026. The aggregate result:

Impressions are up. Clicks are down by nearly half. The SERP is showing your page to more people, and fewer of them are coming to your site. That is not a ranking problem. It is a substitution problem. Google is reading the page, summarizing the page, and displaying the summary in a position that satisfies the user's intent before they ever consider clicking.

The Pew Research audit confirms the mechanism. Pew commissioned a panel of 1,847 US adults to share their search behavior over a 60-day window in early 2026, yielding 12,847 logged sessions. When an AI Overview was present, click-through to the top organic result dropped 47.3%. Click-through to any organic result dropped 38.9%. The effect was most extreme for queries Pew classified as "informational" -- the kind of "how long do you boil an egg" or "what causes a headache" queries that powered a decade of content marketing -- where CTR collapsed 61.2%.

Critically, Pew also found that 26.0% of sessions ending in an AI Overview view resulted in no further click anywhere, including back to search. Users got the answer. They closed the tab. The session was complete. From Google's perspective, this is a feature. From a publisher's perspective, it is the meter running with no fare collected.

Recipes: The First Vertical to Die

If you want to see what comprehensive AI cannibalization looks like, look at recipe sites. Recipe content is the perfect AI Overview target: structured, finite, summarizable, and not particularly novel. There are only so many ways to describe how to make banana bread. Once an AI has read a thousand versions, it can produce the thousand-and-first without any of those sites getting a click.

SimilarWeb's Q1 2026 publisher report found that the top 50 recipe sites by traffic collectively lost 71.2% of organic search visits between January 2024 and March 2026. AllRecipes is down 62.8%. Food Network is down 57.4%. Smaller independent recipe blogs -- the SEO-optimized "scroll past the life story to get to the ingredients" archetype -- are down an average of 78.9%.

Nick Heer's Pixel Envy newsletter tracked one independent recipe site, Pinch of Yum, that publicly reported a 73% revenue decline between 2023 and 2025. The site's organic traffic chart looks like a controlled demolition. The owner shifted to email and YouTube. The web property is now a marketing surface for the email list, not a business in itself.

The death of recipe SEO has a symbolic weight beyond its commercial impact. Recipe sites were the apotheosis of modern SEO: long-form, image-heavy, schema-rich, written for both Google's algorithm and Google's human raters. The genre invented the "jump to recipe" button precisely because the SEO content had grown so unbearable that even Google's users were complaining. AI Overviews ended the bargain. The schema is now being scraped to feed the summary that replaces the click.

The recipe vertical is also the cleanest test of whether brand can save a content business in the AI era. Bon Appétit, the most editorially distinguished food brand on the web, is down 49.2% in organic traffic over the same period -- better than the category average but still nearly halved. Brand helps. Brand does not save you.

News & Publishers: A Slow-Motion Bankruptcy

News publishers are the second-worst-hit vertical and the one with the most public consequences. SimilarWeb tracked 412 top-tier US news domains and found a 54.3% decline in Google organic traffic between January 2024 and March 2026. Search traffic to The New York Times is down 41.7%. Search traffic to The Washington Post is down 58.4%. Search traffic to BuzzFeed News (before it shut down its remaining news operations in late 2025) had fallen 81.6%.

The aggregate impact on the news economy is severe. The Atlantic CEO Nicholas Thompson told Axios in April 2026 that AI Overviews and chatbots had taken roughly $300 million out of the digital news industry's referral economy in 2025 alone. Internal projections at three major US publishers, reviewed for this article, model a further 25-35% decline in Google referral traffic through 2026.

There is one bright signal. Similarweb's referrer data shows that for 14.2% of the top 1,000 publishers, ChatGPT (including the Atlas browser) is now the single largest source of referral traffic, surpassing Google for the first time. The pattern is concentrated in primary-source journalism. Reuters, the Associated Press, and academic publishers like Nature have all seen ChatGPT-driven referrals grow more than 400% year-over-year, partially offsetting Google declines.

But the offset is partial. ChatGPT's referral generosity is concentrated. For mid-market lifestyle and aggregator publications, ChatGPT referrals remain under 3.1% of total traffic -- a rounding error against the Google losses.

The pattern: if you produce something that AI models want to cite -- breaking news, primary sourcing, authoritative analysis -- ChatGPT will send you traffic to compensate, partially, for what Google took away. If you produce aggregation or commodity content, you get neither.

Health & Medical: Hit Hard, with Regulatory Tail Risk

Health queries are the second-largest category in Google search by volume, after navigational queries. They are also the category where AI Overviews are most controversial. WebMD's traffic fell 47.9% between January 2024 and Q1 2026, according to Semrush. Healthline is down 51.3%. Mayo Clinic, which Google's quality raters have long treated as a gold standard, is down 28.4% -- better than the category but still substantial.

The decline is driven by two compounding effects. AI Overviews are aggressive on health queries, appearing on 58.7% of medical informational searches per Search Engine Land's tracking. And ChatGPT's medical query volume has grown substantially: roughly 19.4% of consumer ChatGPT sessions in Q1 2026 included a health-related question, up from 11.2% a year prior. Users who would previously have searched "is amoxicillin safe with alcohol" are increasingly asking ChatGPT instead.

The regulatory tail risk is meaningful. The FTC opened an inquiry in March 2026 into AI Overviews' presentation of medical information, focused specifically on cases where summaries omit contraindications present in the source content. The European Commission is running a parallel review under the Digital Services Act. Neither regulator has signaled enforcement action yet, but the inquiries themselves have already changed publisher strategy: Healthline and WebMD both implemented stricter `noindex` and `noai` directives in late 2025, removing roughly 11% and 7% of their respective content from AI training and summarization pipelines.

The wager is that pulling content out of the AI surface preserves whatever organic traffic remains. Early evidence is mixed. Sites that aggressively block AI crawlers have not seen meaningful recovery in Google organic traffic, since Google's own Search crawler still indexes the content for AI Overview generation under the same `Googlebot` user agent. Blocking ChatGPT does not block Google.

B2B SaaS: The Bifurcation

B2B SaaS is the vertical with the most internal disagreement about whether AI search is actually a problem. The honest answer is that it depends on which part of the funnel you measure.

Branded queries -- "Notion pricing," "Figma login," "Salesforce trial" -- are roughly flat. Ahrefs data shows branded query volume across the top 200 SaaS companies grew 4.7% year-over-year through Q1 2026, with CTR essentially unchanged. AI Overviews appear on only 12.3% of branded SaaS queries. When they do appear, they typically summarize the company's own product page, which routes users to the brand site anyway.

Top-of-funnel content -- "what is CRM," "how does API authentication work," "best practices for OKRs" -- has collapsed. The same Ahrefs dataset shows top-of-funnel SaaS content lost 33.7% of clicks year-over-year on impressions that grew 18.4%. The HubSpot blog, the canonical example of TOFU SEO done at industrial scale, lost 41.6% of organic search traffic between January 2024 and March 2026, according to Semrush's domain analytics. The Salesforce Trailhead content hub lost 36.8%. Intercom's blog lost 49.2%.

Mid-funnel comparison content ("Notion vs. Asana," "Stripe vs. Adyen") sits in between. AI Overviews increasingly attempt to render comparison tables directly in the SERP, but users frequently still click through for pricing or signup. CTR on comparison queries dropped 22.4% -- meaningful, not catastrophic.

The implication for SaaS marketers is uncomfortable but clear. The HubSpot playbook -- ranking for high-volume informational queries to feed the top of the funnel -- has been broken at a structural level. AI surfaces ate the top of the funnel. The strategy that built HubSpot from a startup into a $30 billion enterprise is no longer available to new entrants. New SaaS companies launching today need a different acquisition playbook, and most do not have one.

E-Commerce: The Transactional Bedrock

The most counterintuitive finding in the data is that core e-commerce search is essentially fine. Transactional queries -- the queries that produce revenue -- have lost remarkably little CTR to AI surfaces.

Pew's audit found that when a user's query includes a brand and a product ("Sony WH-1000XM6 headphones"), AI Overviews appear only 9.7% of the time, and when they appear, click-through to the top organic or paid result drops just 11.7%. Google has no commercial incentive to satisfy a buying query with a summary. The summary doesn't sell anything. The shopping ad does.

SimilarWeb's e-commerce panel shows that the top 100 US retail sites collectively grew Google search traffic by 7.8% year-over-year in Q1 2026. Amazon's search referrals from Google are up 12.3%. Shopify-hosted DTC brands collectively grew 14.7%. The traffic is not coming from informational queries about products; it is coming from transactional queries that AI Overviews leave alone.

The pattern within e-commerce mirrors the broader pattern. Transactional intent is preserved. Informational intent is gone. The "best vacuum cleaner 2026" affiliate site is in the same trouble as the recipe blog. The product detail page on the brand's own site is doing fine.

This is partly why Google has been so willing to deploy AI Overviews aggressively. The queries Google monetizes most heavily -- transactional commercial queries served by Shopping ads and Performance Max -- are largely untouched. The queries that AI Overviews are cannibalizing are queries Google previously monetized only weakly through display ads on third-party sites. From an ad-revenue-per-query standpoint, Google has chosen to sacrifice low-monetization queries to defend high-monetization queries against ChatGPT and Perplexity. The strategy is working: Alphabet reported Q1 2026 search ad revenue of $54.8 billion, up 8.4% year-over-year, despite the decline in clicks to the open web.

Travel: Inspiration Dies, Booking Lives

Travel is the cleanest illustration of the inspiration-versus-booking split that defines this era of search. SimilarWeb tracked 184 travel domains and found a 41.6% decline in organic traffic to "inspiration" content -- destination guides, "10 best beaches in Mexico" listicles, photo-driven discovery pages. Booking.com's editorial guides are down 47.3%. Lonely Planet's web content is down 52.8%. Condé Nast Traveler is down 38.4%.

Booking-intent traffic, on the other hand, barely moved. Booking.com's transactional pages are down 6.8%. Expedia's hotel search pages are essentially flat (-1.4%). Airbnb's destination listing pages are up 4.2%.

The mechanism is clear from the SERP: when someone searches "5-day Tokyo itinerary," an AI Overview now produces a passable itinerary directly in the search results, killing the click to the travel inspiration site. When someone searches "Park Hyatt Tokyo June 2026," Google still routes them to the OTA or the hotel's direct booking page, because that is where Google makes money.

The travel publishers caught in the middle -- the destination content sites whose business model depended on inspiration traffic that converted to affiliate revenue -- are in serious trouble. The OTAs and the hotel chains, whose business model depends on booking traffic, are not. A travel content portfolio that mixed inspiration and booking has unbundled in real time.

Developer Documentation: The Stack Overflow Canary

Stack Overflow has been the canary in the AI search mine since 2023. Business of Apps shows Stack Overflow traffic is down 78.4% from its 2022 peak. The decline began before AI Overviews, driven by developers asking GitHub Copilot and ChatGPT for code answers directly. AI Overviews simply finished the job for the long tail of Stack Overflow queries that hadn't yet migrated to AI tools.

The pattern is now spreading to other developer documentation. MDN Web Docs is down 31.4%. DigitalOcean's community tutorials are down 44.7%. The Mozilla and W3Schools properties that anchored generations of developer learning have lost meaningful share to ChatGPT, Cursor, and Claude Code, which now answer questions directly inside the IDE.

The bright signal is that first-party documentation maintained by the platform itself is holding up better. Stripe's developer docs are up 8.2% in organic traffic. The Vercel and Next.js docs are up 14.6%. The pattern: when developers need to understand a specific product, they still come to the canonical source. When they need to understand a generic programming concept, they ask the LLM.

This has direct implications for content strategy in technical categories. Documentation owned by the product (canonical, frequently updated, authoritative) is defensible. Documentation that aggregates and explains other people's products (tutorial sites, "how to use X" content farms) is not.

Local Services: Google's Own Land Grab

The most underreported cannibalization story is Google's own substitution of Local Pack for organic results. BrightLocal's Q1 2026 local SEO report found that on local-intent queries ("plumber near me," "dentist in Brooklyn," "best pizza Austin"), the Local Pack now appears on 87.3% of results, and the average CTR to the organic listings below the Local Pack has fallen to 4.1%. Google Maps and Google Business Profiles capture roughly 73.8% of local-query clicks.

This is not technically AI cannibalization. It is Google cannibalization. But the effect on third-party local content sites -- Yelp, TripAdvisor, the local newspaper's "best of" guides -- is identical to AI Overviews' effect on health content. The information is consumed inside Google's surface. The third-party site never gets the click.

Yelp's organic search traffic from Google is down 51.7% year-over-year. TripAdvisor is down 44.3%. Local newspaper "best of" content is down 62.8%. The local services vertical illustrates an important point: AI Overviews are not the only cannibalization vector. Google has been substituting its own properties for organic results for fifteen years. AI Overviews are the latest, most aggressive iteration of a strategy that includes Knowledge Panels, Featured Snippets, Local Pack, Shopping carousels, and YouTube embeds.

The Contrarian Read: Not Everyone Is Losing

The dominant narrative is that AI search is killing the open web. The data does not support that framing. The data supports a narrower claim: AI search is killing a specific business model -- the informational, middle-of-funnel content site that monetizes through display ads and affiliate links.

Sites with proprietary data are thriving. Bloomberg Terminal content, Crunchbase, PitchBook, Glassdoor's salary database, Zillow's listings, and Sherwood's market data are all growing both in organic search and in AI citations. AI models need to cite something authoritative for proprietary data; they cannot generate it. Profound's AI citation index shows that proprietary-data sites are cited 4.2x more often per query than aggregation sites with comparable Google rankings.

Sites with active communities are stable. Reddit's organic traffic from Google is up 18.4% year-over-year, driven by Google's preferential surfacing of Reddit threads in response to user demand for "real opinions." Discord-indexed communities, GitHub, and Quora (which has recovered from its earlier slide) are all growing. AI Overviews increasingly cite Reddit threads as primary sources, and users still click through to Reddit to read the full conversation.

Sites with transactional intent are growing. As the e-commerce data showed, sites that close a purchase, a booking, or a signup are protected. AI surfaces are not optimized to replace transactions; they are optimized to replace explanations.

Brand-strong sites with newsletter or app distribution are stable. The Atlantic, Bloomberg, and The Information have lost organic search traffic but grown subscriber bases. Substack's top-tier writers report that AI search has not meaningfully affected their economics, because the open-web click was not their primary distribution channel anyway. The loser is the middle: content businesses that depended on Google to send them users who would not have come otherwise.

The split between "defensible" and "exposed" maps almost perfectly onto the question of whether the site does something the AI cannot easily replicate. Hosting unique data, hosting humans, closing transactions, and being the authoritative source are durable. Producing well-optimized summaries of public information is not.

What To Do Now

The data points to five recommendations that the strongest practitioners in the space have converged on:

1. Shift content from informational intent to transactional or experiential intent. Pew's data is unambiguous: informational queries lost 61.2% of CTR, transactional queries lost 11.7%. A content strategy that produces calculators, configurators, comparison engines, interactive tools, and decision-support flows is structurally more defensible than one that produces explanations. The goal is to produce content that requires a click to be useful, not content that an AI can summarize away.

2. Build proprietary data, communities, or tools that AI cannot replicate. Profound's citation data shows that proprietary-data sites are cited 4.2x more often per query than aggregation sites with comparable rankings. The defensible content of the next decade is content that the AI must reference rather than content the AI can substitute for. Zillow, Crunchbase, Glassdoor, and Reddit have it. Your "Ultimate Guide to X" does not.

3. Invest in brand search and direct distribution. Branded queries are flat. CTR on branded queries is flat. The fastest-growing acquisition channel for top-quintile B2B SaaS companies in 2026 is branded organic search, according to SparkToro's brand search index. Translation: invest in becoming the destination that users type directly into Google, rather than the page that Google's algorithm chooses to surface. The latter is increasingly out of your control.

4. Optimize for AI surfaces themselves through structured data, llms.txt, and licensing. AthenaHQ, Otterly, and Profound have made share-of-AI-citations measurable for the first time. The early data suggests that structured data (especially `Article`, `FAQ`, and `HowTo` schema), well-maintained llms.txt files, and explicit content licensing deals with OpenAI, Anthropic, Google, and Perplexity meaningfully increase citation rates. Publishers that have signed licensing deals -- including the Financial Times, the Associated Press, and News Corp -- are seeing higher AI referral volumes than peers that have not.

5. Measure share-of-AI-citations alongside share-of-organic-clicks. The leading content teams in 2026 are tracking three numbers in parallel: Google clicks, AI citations (across ChatGPT, Perplexity, Claude, and Google's AI Mode), and direct visits. The combined measurement gives a complete picture of where attention is going. Sites that are losing Google clicks but gaining AI citations are in a stronger position than the headline number suggests. Sites that are losing both are in genuine trouble and should redirect investment immediately.

The Quiet Consolidation

The narrative of the open web's collapse misses the most interesting part of the data, which is that some sites are winning faster than ever. AI search is a consolidation event, not a dispersion event. The sites with structural advantages -- proprietary data, communities, transactional intent, brand strength -- are absorbing share from the sites without those advantages. The middle is being hollowed out. The ends are growing.

This is the pattern technology platforms produce repeatedly. The arrival of a new layer compresses the layer beneath it. AI search is compressing the content layer in the same way that the App Store compressed the mobile web, that Amazon compressed third-party retail, that YouTube compressed cable. The winners on the compressed layer are the ones that gave the new layer something it could not produce on its own.

The losers are the businesses that operated as efficient summarizers of public information. They built that model because Google's algorithm rewarded it for two decades. Google has now built an in-house summarizer that does the same job inside the SERP, for free, with the user never leaving. The implicit deal -- you summarize public information, Google sends you traffic, you monetize through ads -- has been unilaterally rewritten by the party that wrote it in the first place.

The traffic is not coming back. Pew's data, SimilarWeb's data, SparkToro's data, and Ahrefs' data all point in the same direction. Zero-click search is now the default. AI Overviews are still expanding. ChatGPT Atlas and Perplexity Comet are taking the queries that don't even reach Google. The structural conditions that produced the SEO industry between 2005 and 2024 are over.

What replaces it is not nothing. The companies that own proprietary data are growing. The communities that host conversations are growing. The transactional businesses are growing. The first-party documentation is growing. The brands that built direct audience relationships are stable. The open web continues, in pieces. The piece that does not continue is the piece that everyone thought was the whole thing: the long-tail, ad-supported, search-traffic-dependent content site.

That model had a fifteen-year run. It generated meaningful careers, real businesses, and a body of content that AI now uses, free of charge, to put it out of business. The bill is paid. The model is closed. The publishers that figured this out in 2024 are now growing again, on different terms. The ones still optimizing for the old algorithm are running a race against a clock that has already stopped.