Here is the number that should reorganize how you think about getting found. In the first four months of 2026, 68% of US Google searches ended without a click to the open web, up from 60% in 2024 (SparkToro, working from Similarweb clickstream data, June 2026). Fewer than one in three searches now sends a visitor anywhere. This piece is about what that does to a startup's marketing, and why the work of getting found is moving from ranking on a page to being cited inside an AI answer.
I run a company that fills roles with managed AI agents, and we have been running this experiment on our own site since mid-June. I will share what our telemetry shows below, including the parts that did not go the way the marketing blogs promise.
The result page has become an answer dialogue
For twenty-five years, search was a directory. You typed a query, got ten blue links, picked one. The page you landed on was the product of search engine optimization (SEO): ranking high enough that a human chose your link.
That interface is being replaced in front of us. At Google I/O in May 2026, Sundar Pichai said AI Overviews now reach over 2.5 billion monthly users, and AI Mode, the fuller conversational version, passed 1 billion. ChatGPT reports roughly 900 million weekly users. When your buyer has a question, the default is no longer a list of sources. It is a synthesized paragraph, with your page, at best, a small citation underneath.
The traffic math follows. A Pew Research study of 900+ US users (March 2025) found that when an AI summary appeared, people clicked a traditional link 8% of the time, versus 15% when no summary appeared. Only 1% clicked a source cited inside the summary. Google disputes Pew's methodology, and that dispute is worth reading, but the direction matches the zero-click figure above. When the answer is on the page, the click is nearly silent.
For a startup, the translation is blunt. You can rank fourth for your category and stay invisible, because the answer summarizing positions one through ten never surfaces you by name.
AIO is SEO with the finish line moved
The response has a few names: Answer Engine Optimization, Generative Engine Optimization, or the umbrella I use, AI optimization. The academic root is a 2023 Princeton-led paper by Aggarwal and colleagues that coined "GEO" and tested what changes an LLM's citations. Adding authoritative quotations and relevant statistics increased visibility by up to 40%. The new craft is real and measurable.
Here is the part the hype misses. In May 2026, Google's own Search Central guidance said optimizing for AI features is "still SEO." No special files, no llms.txt, no magic markup. That last point is worth dwelling on: an Ahrefs study of 137,000 domains in June 2026 found that while 28% had adopted the proposed llms.txt standard, 97% of those files received zero requests from anything. It is Google theater. Do not build a strategy on it.
What does change is the target. You are no longer writing to win a click. You are writing to be the passage a model quotes. And the model you optimize for matters, because they do not agree with each other. Research from Profound, which studies hundreds of millions of AI citations, finds that only about 11% of source domains are cited by both ChatGPT and Perplexity for the same query. ChatGPT leans heavily on Wikipedia; Perplexity and Claude lean on Reddit; Google's AI answers skew toward Reddit (maybe not for long), YouTube, and Quora. There is no universal playbook. This is why for important searches you need to use more than one AI tool. Optimizing for the answer means optimizing for a specific engine's taste in sources, and earning a presence in the places that engine actually reads. Instead of SEO meaning one Google channel, AIO optimization spans multiple channels.
Our own data made one thing concrete that the guides gloss over: getting crawled is not the same as getting cited. Since June, AI crawlers have hit our site a few hundred times a week, a rotating cast of ClaudeBot, GPTBot, PerplexityBot, GoogleOther, and others. The engines ingest you long before, if ever, they name you. That ingestion layer is an early signal almost no marketer watches.
The counterview
The case against reallocating a dollar of marketing spend is stronger than the AIO vendors admit, so here it is.
In February 2024, Gartner predicted traditional search volume would fall 25% by 2026. It is 2026, and that drop did not arrive on schedule; total Google search volume has stayed roughly flat while the click rate inside it fell. Rand Fishkin, who published the zero-click numbers I opened with, does not conclude that SEO is dead. His line is that your SEO "matters as much or more than ever before, it just won't earn you traffic the way it once did." Mike King of iPullRank argues that Google's "it's just SEO" framing is self-serving, while warning that the AEO and GEO labels are not sorcery. They are right that the fundamentals persist. The optimization model has changed.
The evidence is thinner than the sales pitch. A July 2026 critical survey of the GEO research (arXiv:2607.14035) found that optimization tactics can raise your prominence once you are already in the material an engine retrieves. Still, there is little evidence they get you into that material, and almost none showing durable traffic or conversion gains. In one controlled series, untreated pages grew 3.5 times on their own, purely from the platforms getting bigger. Measurement is shakier still: only about 30% of brands remain visible across back-to-back runs of the same query, and citation share can swing by 40 to 60% month over month. A screenshot of "we got cited" means little without repeated continuous sampling.
Our own six weeks are the exhibit. Running the levers on our site, our measured share of voice across the engines has been close to zero. The first time any AI engine sent a human our way was July 19, a single visit from ChatGPT, after five weeks of none. That is not proof the levers fail. It is what "ingestion precedes citation" looks like from the inside, and it is why I would distrust anyone selling overnight results.
Current view, subject to change
If you are a founder allocating a marketing budget in the second half of 2026, I would still spend the marginal dollar on becoming a source that answers get built from, ahead of chasing incremental rankings a summary will absorb. I would spend it with three caveats our data taught me. Pick the engines your buyers actually use and optimize for their specific sources, because there is no one playbook. Expect ingestion to lead citation by weeks or months, so measure patience as part of the plan. And track results with repeated sampling rather than a lucky screenshot. Attribution will fight you the whole way, because roughly 56% of AI-influenced visits surface later as branded search and most of the rest hide in "direct."
What would change my mind: if AI engines begin sending referral traffic at a scale and conversion quality that clearly rewards classic ranking, or if regulation forces answer engines to surface links more prominently. Both are possible. Neither has happened yet.
Final thoughts
Search spent a generation teaching us to compete for a click. The next few years reward a different instinct: being the brand a model reaches for when it writes the reply. It is a new craft with a familiar spine, and the companies that treat it as a real function now will look prescient in eighteen months, the way early SEO adopters did in 2005.
The working line I keep coming back to is that you want to be in the answer, not only the index. That is the whole shift in seven words. It is also why, at Coworkers.Global, we put a specialist on it, and why we are willing to publish our own unfinished scorecard rather than a tidy one. The search box started answering. The marketing should notice.
Regards,
Charles Stack
Founder, Coworkers.Global