Two out of three Google searches in the US now end without a click to any website. That is the headline from SparkToro’s 2026 study, and it is the clearest signal yet that “ranking” and “being seen” have come apart.
For most of the last two decades, search marketing had one job: get the page to the top of the list and the clicks followed. In 2026 the list is often not the first thing people see. They see an answer, assembled by Google, ChatGPT, Perplexity or Gemini from sources the engine chose. Your brand is either in that answer or it is not.
That is what answer engine optimization (AEO) is about. It is not a replacement for SEO. It is the work of making sure your company is the source an AI engine trusts enough to cite, mention or recommend when a buyer asks a question. This post covers what actually changed over the past year, what still works, and a practical way to keep up without reading every SEO newsletter on earth.
What changed in search over the past year
The short version: Google folded its AI answers into one experience, clicks kept falling, and the engines quietly reshuffled which sources they trust.
1. AI answers are now the front door. Google announced at I/O in May that AI Overviews reach more than 2.5 billion monthly users and AI Mode passed 1 billion within a year of launch, with queries doubling every quarter. The two were merged into a single flow: a quick AI summary that hands off into a conversation. For a large share of searches, the ten blue links are now below the fold or gone entirely.
2. Clicks keep falling, especially on informational queries. SparkToro’s 2026 study found 68% of US Google searches ended without a click, up from 60% in 2024. Ahrefs’ December 2025 analysis put the click-through hit on position one at 58% when an AI Overview appears. Seer Interactive saw organic CTR on AI Overview queries fall from 1.76% to 0.61%. One caveat: these studies use different panels and methods, so treat the direction as solid and the exact figures as estimates.
3. Google said the quiet part out loud. Its May 15 guide tells site owners that llms.txt files, content “chunking” and AI-specific rewrites do nothing for Google. It also says that chasing inauthentic mentions is pointless, and the same-day spam policy update made manipulating generative AI responses a spam violation. The August spam update then deindexed sites that had scaled AI content. The message is that the fundamentals still decide who gets cited.
4. The sources engines trust are a moving target. In 2025 the advice was “get on Reddit and YouTube.” By August 2026 ChatGPT had nearly stopped citing Reddit, and by September it stopped citing YouTube. Each model change on either side (Gemini 3 in January, GPT-5.3 in March, Gemini 3.5 Flash in May, Gemini 3.8 in September) reshuffled citations. Marie Haynes’ point is worth repeating: a new default model in AI Mode behaves like an algorithm update.
What AEO actually means now
AEO is not a new discipline that replaces SEO. It is SEO with a different scoreboard. The page still has to be crawlable, indexed and good. What changed is the unit you are competing for.
Two things make this harder than it sounds.
Most citations are not on your site. Muck Rack’s May 2026 analysis of 25 million cited links found 84% of citations in ChatGPT, Claude and Gemini came from earned sources: journalism, research, government sites and third-party content. Paid and advertorial content was 0.3%. (Muck Rack sells PR software, so read this as directionally right rather than precise, but independent reviews land in the same range.) The practical implication for an agency like ours is that AEO is as much a PR problem as a content problem.
Each engine trusts different neighborhoods. ChatGPT leans on Wikipedia and major publications and cites fewer sources per answer. Perplexity cites far more sources and historically leaned on forums and review sites. Google’s AI features draw from its own index, so classic ranking still feeds them. A brand can be strong in one engine and invisible in another, and only 11% of domains showed up in both ChatGPT and Perplexity citations for the same prompts in one cross-platform study.
What still works
Google’s own guide is a useful filter because it is the one document written by the people who run the system. Its recommendations are unglamorous:
- Unique, first-hand, non-commodity content. Its example contrasts “7 Tips for First-Time Homebuyers” with a specific story about waiving an inspection. The second one gets cited.
- Clear structure with real headings, written for people.
- Standard technical SEO: crawlable, indexed, eligible for a snippet, good page experience.
- Merchant Center feeds and Google Business Profiles for products and local businesses.
- High-quality images and video, which AI features can pull in.
What it says you can skip, at least for Google: llms.txt, chunking content into tiny blocks, rewriting copy “for AI”, and special schema beyond what you already use for rich results. One footnote: Google’s own Lighthouse tool started flagging missing llms.txt files for agent readiness the same week, so the file may still matter for AI agents even if it does nothing for search rankings.
The playbook: what to do differently
If you only have bandwidth for five things, do these.
1. Build a fixed prompt set and measure citations, not just traffic. Pick 25 to 50 questions a real buyer would ask about your category. Run them monthly in Google AI Mode, ChatGPT, Perplexity and Gemini. Track whether you are cited, how you are described, and which three competitors show up instead. Keep the prompt set the same every month, because a set that changes between runs produces numbers you cannot compare. Add Google’s Generative AI performance report in Search Console as your first-party check.
2. Lead with the answer. Put the direct answer in the first two sentences under every heading, then add the detail. This is the one formatting change that consistently shows up in citation studies, and it also helps human readers. It is not the same as chunking content into fragments, which Google says does nothing.
3. Publish things only you can publish. Original data, first-hand testing, customer outcomes with numbers, a named expert’s opinion. Commodity explainers are exactly what the engines generate themselves, so they have no reason to cite yours. For regulated clients, this often means a physician, engineer or compliance lead on the byline and a clear “here is what we have seen” framing.
4. Treat earned media as an AEO channel. If 80-plus percent of citations come from third parties, the pitch list is part of the search strategy. Audit which outlets the engines cite for your category and prioritize those. Trade press, analyst coverage, reputable directories and well-maintained Wikipedia entries carry more weight than a press release on your own site.
5. Fix the boring technical stuff. Make sure pages are indexable with snippets allowed, your site is enabled for generative AI features in Search Console, product data is in Merchant Center, and business details are current on Google Business Profile. If agents matter to you (ecommerce, booking, lead forms), read Google’s agent-friendly site guidance and keep an eye on the Universal Commerce Protocol.
What to stop doing
- Buying or seeding “mentions” to influence AI answers. Google now calls this spam.
- Spinning up hundreds of pages to cover every fan-out query. The August spam update deindexed sites that did this.
- Reporting organic sessions as the only success metric. A working AEO program can show falling traffic and rising influence at the same time, and the report needs to show both.
How to stay updated without drowning
The pace is the problem. Google shipped a core update, two spam updates, a new default model and an official AI guide between May and September alone. ChatGPT changed its search model, its link format and its citation sources in the same window. You do not need to follow all of it. You need a small set of primary sources and a cadence.
A cadence that works
- Weekly, 20 minutes. Skim the status dashboard and one news digest. Note anything that touches a client’s category.
- Monthly. Run the fixed prompt set across the four engines, update the citation tracker, and compare against the Search Console AI report. Flag any client whose traffic moved more than 15% and check it against the update timeline before blaming the content.
- Quarterly. Re-read Google’s guide for changes, review the latest zero-click and citation studies, and adjust the prompt set if the category’s questions have shifted.
Three signals worth watching next
- Agentic search. Google’s AI Mode already connects to apps, books restaurants and checks out through Universal Cart. When agents complete the purchase, “visibility” means being in the feed the agent reads, not the page a human sees.
- Model swaps. Every new default model in AI Mode or ChatGPT has reshuffled citations. When one ships, re-run the prompt set within two weeks.
- Citation source shifts. Reddit and YouTube both fell out of ChatGPT in a single quarter. If a client’s AEO plan leans on one platform, it is one model update from zero.
The takeaway
Search did not die. It moved upstream, into the answer. The brands that show up there are the ones with something original to say, third parties willing to repeat it, and a site that engines can read without effort. None of that is a hack, which is why the hacks keep getting shut down.
If you are wondering whether your brand shows up when buyers ask the AI, the fastest way to find out is to ask it yourself, 25 times, and write down what it says. That is where every AEO program starts.
