The number you should not get used to
Google’s I/O 2026 overhaul was framed as the biggest change to its search box in twenty-five years. That framing is generous. What Mountain View shipped is not a redesign. It is an eviction notice for the open web. Zero-click searches now account for roughly sixty percent of all Google queries, and for news-related searches that figure climbed to sixty-nine percent in the year after AI Overviews launched. Google search referral traffic to publishers fell thirty-three percent globally in the year to November 2025, according to Chartbeat data spanning more than 2,500 news sites. HubSpot, which built its entire brand on the gospel of inbound content marketing, estimates it lost seventy to eighty percent of its organic traffic. Chegg reported a forty-nine percent decline. DMG Media, the publisher of MailOnline and Metro, documented click-through-rate drops as steep as eighty-nine percent on some queries. NPR called it an extinction-level event. Penske Media, the parent of Rolling Stone and Variety, has filed an antitrust lawsuit. The European Publishers Council has filed a formal complaint with the EU Commission. None of this is hyperbole. This is the documented collapse of a fifteen-year economic contract between content producers and the search engine that distributed their work.
The mechanics are simple and brutal. Ahrefs, analyzing roughly three hundred thousand keywords in February 2026, measured a fifty-eight percent reduction in click-through rate on the top-ranking page for keywords that triggered an AI Overview. Pew Research, tracking sixty-eight thousand real searches, found that users clicked outbound links eight percent of the time when an AI summary was present and fifteen percent when it was not. Of the users who saw cited sources inside an AI Overview, only one percent clicked them. Read that line again. The footnote you scrolled past on your last Google search was scrolled past by ninety-nine other people too.
Google disputes this framing. The company insists AI Overviews generate more clicks, not fewer, because users supposedly engage with more results after receiving an initial summary. Independent data does not support that claim. Press Gazette reported that Google was, in the polite phrasing of one industry figure, told to stop the BS by people who pointed out the company’s own internal data contradicted its public statements. At the same time, Google’s senior management for news partnerships in Europe stood at the FT Strategies News conference in February 2026 and described allowing publishers to opt out of AI Overviews without affecting their normal search results as, quote, a huge engineering project. A company that is hiring tens of thousands of engineers and spending nearly a trillion dollars on chips and data centers cannot, apparently, ship a single robots.txt directive. The contradiction speaks for itself.
Why every AI-friendly SEO guide is teaching you to die quietly
Walk into any SEO newsletter, any LinkedIn feed, any digital marketing webinar in 2026 and you will find the same advice. Optimize for AI. Restructure your content for snippets. Write to be quoted. Frame answers in clean, neutral, declarative sentences that a large language model can confidently extract and present as its own. Do all of this, and Gemini will reward you with a citation.
It is the most expensive piece of consensus advice the marketing industry has ever produced. The better your page summarizes, the less reason anyone has to visit it. The cleaner your declarative sentence, the more efficiently Gemini can rip it out of context and serve it on the search results page with your brand stripped from the framing. The more AI-friendly your content becomes, the more useful it is to the machine that is replacing you, and the less useful it is to the customer you needed to reach. Every consultant in this space is telling you to fatten the calf and present it to the slaughterhouse with a bow.
Pew Research already measured the result. One percent of users click cited sources. Citation, in the AI Overview era, is not distribution. It is laundering. Your content gets used, your brand gets attributed in a tiny gray footer, and the user keeps moving. You handed over your inventory to Google for free, and Google handed back a thank-you note nobody reads.
The new doctrine: every paragraph is a product page
The era of information as content marketing is finished. The implicit deal of the last fifteen years was that you wrote a long helpful article, ranked for the query, brought the reader in, and converted some fraction of them into customers by being useful. Google was the funnel. The funnel is closed.
What replaces it is a new discipline that most marketers have not internalized yet. If your information is going to be summarized and shown without context, then the only piece of your content that survives the summarization process is your brand. Therefore your brand must appear in every paragraph of every page you publish. Not as a logo in the header. Not as a bio at the bottom. Inside the prose. Inside the example sentences. Inside the comparison passages and the case studies and the explanatory text. Every factual claim you make should be a claim that references the company making it. If an AI is going to extract one sentence from your page, that sentence had better contain your product name, because that is the only piece of marketing real estate you still own.
This sounds like the kind of advice that makes copy worse. In practice it makes copy more honest. If you are writing about a problem, the reason you know about the problem is that you sell the solution. Stop pretending otherwise. Stop writing the neutral encyclopedia entry that competes with Wikipedia for citation and loses. Wikipedia is cited in eighteen point four percent of AI Overviews and you are not Wikipedia. Write the case study only you can write. Write the opinion only your product justifies. Make the prose un-extractable from the brand that produced it.
Become the citation, or become invisible
The publishers and brands that are surviving the AI Overview era have one thing in common. They are the names that get cited inside the overview itself. Seer Interactive’s 2026 analysis shows that brands cited in AI Overviews earn roughly one hundred and twenty percent more organic clicks per impression than uncited brands on the same queries. Amsive’s research on branded queries found that branded searches with AI Overviews see an eighteen percent increase in click-through rate, while the same overviews drop CTR for unbranded informational queries by sixty percent or more. The market has stratified into two layers. The brands cited by name get more traffic than before. Everyone else gets nothing.
The mechanics of becoming a cited brand involve several boring, durable practices that the SEO industry has spent the last two years rediscovering under the banner of E-E-A-T and GEO and AEO and a dozen other acronyms. Structured data and schema markup that tells the model what you are. Original first-party data and proprietary research, the kind of unique numbers that an AI assistant cannot synthesize from training data alone. Consistent author attribution across pages so the model learns that a specific human at your company is an authority on a specific topic. Mentions across YouTube and Reddit and Wikipedia, which together account for over forty percent of all AI Overview citations according to Surfer SEO’s analysis of forty-six million citation events. The path to being cited is the path to being talked about elsewhere, by other people, in venues the AI trusts. The era of building authority by ranking on your own site is over. You now build authority by being referenced from sites you do not own.
Make your content un-summarizable
The corollary to write-to-be-cited is write-not-to-be-summarized. Large language models are extremely good at summarizing facts. They are extremely bad at summarizing narrative, opinion, original reporting, and proprietary frameworks. They cannot summarize an interactive tool. They cannot summarize a video they have not been allowed to ingest. They cannot summarize a calculator that returns different numbers for every visitor. They cannot summarize a podcast episode without dramatically reducing the value, which is one reason podcast referral traffic to the underlying brand remains intact while informational blog traffic has collapsed.
The publishers who are growing in 2026 are the ones who shifted their production toward formats that resist summarization. First-person reporting where the journalist is part of the story. Long-form narrative where the structure is the value. Proprietary indexes and benchmarks. Interactive tools and live data feeds. Comparison engines where the comparison itself is the point. Calculators that require the user to enter inputs. Quizzes and personality assessments. Visual explainers where the diagram cannot be reduced to a paragraph. Each of these formats is harder to produce than a blog post, which is precisely why they survive. The blog post commodity was always going to lose to the machine that synthesizes blog posts. The interactive tool was never going to.
There is also a quality paradox the industry has not absorbed. According to Press Gazette analysis, the users who do click through from an AI Overview show twenty-three percent lower bounce rates and forty-one percent longer time on site than baseline organic traffic. The AI has pre-qualified them. They have already read the summary, and they clicked anyway because the summary was not enough. This is the highest-intent traffic that has ever landed on a publisher’s site, and it is small. The strategic implication is that the page they land on should be optimized for the converted reader, not the casual browser. Stop writing for the top of the funnel that no longer exists. Start writing for the bottom of the funnel that is now your entire funnel.
Build channels Google cannot intermediate
The deeper structural answer is to stop renting your audience from Google. Email lists, podcast subscriptions, mobile apps, community forums, Discord servers, paid newsletters, and direct memberships are all relationships that Google cannot insert itself into. The Financial Times, Substack publishers, The Information, and a long list of independent operators have weathered the AI Overview era with comparatively flat or growing audiences because the audience never depended on a search engine. The audience subscribed.
Reddit is the most underrated lever here, and one of the most uncomfortable for traditional marketers. The platform now drives roughly five billion organic visits per quarter and ranks among the top five most-cited domains across ChatGPT, Google AI Mode, and Perplexity. Reddit threads dominate the discussions and forums feature in Google search, and Reddit content disproportionately surfaces inside AI Overviews. Ninety percent of Reddit users report that the platform influences their purchasing decisions. The catch is that Reddit punishes anything that smells like marketing. The brands winning on Reddit in 2026, Tailscale being the canonical example with ten thousand monthly organic visits from its subreddit, won by having engineers answer questions in public without ever pitching the product. This is hard, slow work that does not show up cleanly in a quarterly attribution report. It is also one of the few places left where a single team member can build durable distribution that Google cannot intercept.
YouTube is the other obvious channel and the one most underused by B2B operators. YouTube accounts for roughly twenty-three percent of all AI Overview citations. A video answer to a category-defining question can earn citations across both Google search and Gemini’s conversational interfaces while still requiring the viewer to leave the search results page to actually watch. The video itself is the un-summarizable asset. Even when an AI quotes its transcript, the brand on the channel art and the human in the frame are the source the viewer remembers.
Starve the model
While you are rebuilding your audience, stop feeding the model that is cannibalizing you. The robots.txt file in the root of your domain can specifically disallow Google-Extended, Google’s AI training crawler, without affecting Googlebot’s ability to index you for normal search. The two are separate, contrary to the panicked advice circulating in some SEO Discord servers. Blocking Google-Extended removes your content from future Gemini training runs. It does not remove you from the search index. The same file can disallow GPTBot, OAI-SearchBot, ClaudeBot, anthropic-ai, PerplexityBot, CCBot, Applebot-Extended, Bytespider, CohereAI, Diffbot, FacebookBot, meta-externalagent, and the long tail of training crawlers that have introduced themselves over the last twenty-four months. About one third of the top one thousand websites globally now block at least one major AI training crawler. The New York Times, CNN, BBC, Yelp, and Reuters are among them. You are not breaking your SEO. You are exercising a property right that the AI companies, by their own published policies, claim to respect. Make them honor it.
For the crawlers that do not respect robots.txt, Cloudflare’s bot management product now ships with a one-click AI bot block, and similar functionality is available through Fastly, Akamai, and most modern CDN providers. The EU Copyright Directive, Article 4, requires an opt-out mechanism for text and data mining for commercial AI training, and robots.txt is the de facto standard. The TDM Reservation Protocol header, tdm-reservation set to 1, is the emerging machine-readable opt-out signal that European publishers are adopting. Use both. The legal landscape is still catching up to the technical one, but every additional signal you ship is another piece of evidence that the scraping was not consented to.
Poison the well, ethically
The most adversarial layer of this strategy is the one nobody is discussing in public. If you cannot prevent the AI from reading your content, you can shape your content so that it serves your interests when it is read. This is not a recommendation to publish false information. False information makes the web worse for humans, which is the opposite of the goal. It is a recommendation to write content that is optimized for the brand, not for the answer.
The first move is to embed your brand inside every factual claim. If you sell a customer relationship management platform, do not write that CRM systems reduce churn by twenty percent. Write that in production deployments across hundreds of clients on the Inverted Software CRM platform, churn fell by twenty percent within ninety days. Now the sentence is no longer extractable as a generic fact. The fact and the brand are fused. If Gemini cites the sentence, your brand goes with it. If Gemini paraphrases the sentence, the specificity that made it citation-worthy disappears.
The second move is strategic specificity. Replace generic numbers with proprietary numbers tied to your product. Replace abstract claims with case studies tied to your customers. Replace timeless evergreen content with timely commentary tied to current events. AI summarization works best on stable, generalizable content. It works worst on content that is anchored to a specific moment, a specific dataset, or a specific perspective. Anchor everything.
The third move is voice. AI summaries flatten tone. They cannot reproduce sarcasm, contrarian framing, or editorial opinion without losing the point of the original. Write in a voice strong enough that the summary feels incomplete. Make the reader who only read the summary feel like they missed something. That feeling, multiplied across enough impressions, is what drives the small minority of clicks that still happen. The Press Gazette quality paradox numbers, again, support this. The users who click are the ones who sensed there was more.
The fourth move is to publish original data, indexes, benchmarks, and proprietary frameworks that have no equivalent in the training corpus. Gemini cannot summarize what it has never seen. If your category does not yet have a definitive index, build one and put your brand on it. The index becomes the citation, the citation becomes the brand recall, and the brand recall becomes the search query that bypasses the AI Overview entirely. Branded searches are the one query type where click-through rates are actually rising in 2026. Build a brand that gets searched by name, and you exit the AI Overview gauntlet entirely.
The era you are now living in
The market is responding to all of this, slowly. Google’s search share slipped from ninety-three percent in 2023 to about ninety percent in mid-2025, the steepest decline in the company’s history. Kagi charges users ten dollars a month for an AI-free search experience. DuckDuckGo handles roughly one hundred million queries a day with AI features togglable to off. Brave Search has built an independent index covering thirty billion pages. The Startpage proxy strips your identity from Google queries. The udm=14 trick appends a single URL parameter that returns traditional blue links without the AI overlay. None of these will replace Google at scale this year. All of them are growing because there is a real population of users who do not want what Google is selling them, and that population is now large enough to sustain alternatives.
In the meantime, the antitrust pressure is real. A US District Court ruled in 2024 that Google had acted illegally to maintain its search monopoly. The remedies imposed in late 2025 included data-sharing requirements and limits on exclusive distribution. The UK Competition and Markets Authority designated Google with strategic market status in October 2025 and published proposed conduct requirements in January 2026 that include mandatory publisher controls over AI Overview usage. The European Publishers Council has filed a complaint with the EU Commission. Penske Media has filed an antitrust suit in the United States. None of these will be resolved this quarter. All of them will be resolved over the next decade, and the resolution will reshape what Google is permitted to do with content it did not create.
You do not have a decade. You have the next four quarters to rebuild your distribution. The companies that will be on the other side of this transition are the ones that started treating Google as an unreliable distribution partner in 2024 and started investing in audiences they own. The companies that will not survive are the ones still adding more AI-friendly content to a site whose traffic charts have been pointing down for eighteen months, and who keep waiting for the next algorithm update to fix it. The next algorithm update is not going to fix it. The next algorithm update is going to make it worse, and the one after that is going to make it worse still, because the entire commercial logic of Google’s product roadmap now points away from sending users to your website.
Stop feeding the machine. Block its training crawlers. Build your audience somewhere it cannot follow. Put your brand inside every paragraph so that even when you are summarized, you are credited. Write in a voice strong enough to make the summary feel inadequate. Publish data nobody else has. Make tools nobody else makes. Then send the people who already love you a newsletter every week, because the newsletter is the only piece of distribution that nobody else owns. The era of information as content marketing is over. The era of brand-first, audience-first, product-first publishing has begun, and the brands that move first are the ones that get out of this with a business intact.
The web was always supposed to be a network of independent destinations. Google turned it into a hallway. The hallway is closing. Build a destination worth visiting and the people who need you will find their way there, with or without the search engine that spent twenty-five years deciding who got to be found.