Profound’s State of AEO 2026 did not waste time. The AEO analytics platform opened with a simple provocation: the confidence many marketers had built around AEO over the past two years was misplaced. Write listicles, stay fresh, post on Reddit, know that ChatGPT loves Wikipedia and Perplexity loves forums. Two years of case studies, citations climbing, wins on the board. The presenters’ point: that playbook just broke.
Two Weeks That Changed Everything
Between May 7 and May 22, 2026, the two biggest AI search interfaces in the world rewrote themselves. Profound called it the most significant two-week shift in AEO history, and the data backs it up.
On May 7, OpenAI shipped GPT-5.5. The headline was the model upgrade. The bigger move was quieter: citation chips inside ChatGPT converted from static labels into clickable hyperlinks. The numbers that followed were unlike anything Profound had recorded before:
- ChatGPT referral traffic jumped roughly 60% overnight
- Over 8 million visits hit thousands of sites in a single week
- Homepage traffic rose from 3.4% of ChatGPT clicks to 24.2%, a nearly 7x jump
OpenAI did not make that change without reason. ChatGPT is building an ads infrastructure, and it needs behavioral data on what users click. The hyperlinks were laying the rails.
On May 22, Google moved AI Mode out of a separate tab and into the core search box. Both of the two most important interfaces in AI search changed in the same two-week window. Running alongside all of this: Google search volume in Safari is falling for the first time ever, a decline Apple’s Eddie Cue publicly attributed to AI adoption. The shift showed up in the traffic data.
ChatGPT Ads Are Live, and They Are Going After Your Citations
The assumption that AEO is purely an earned-media game is no longer accurate. Self-serve ads are running inside ChatGPT. Where the market stands right now:
- Roughly 30% of eligible ChatGPT users are already seeing ads
- About 50% of placements appear only on the first turn of a conversation
- 4 out of 5 conversations contain just one ad. The surface is not saturated yet.
What makes the ad model different from traditional search is the query context. Traditional search catches short, generic intent. AI search catches explicit, high-detail prompts where users describe their exact situation in a single message. Profound described it as potentially the most precise targeting engine ever built.
The competitive dynamic is already aggressive. According to Profound’s data, 17% of ads in ChatGPT mention a brand that is a direct competitor to the brand already mentioned organically in the same response. Ask ChatGPT how to use Salesforce, and a competitor may appear as a paid placement right next to the organic answer. Salesforce’s response has been to run defensive ads, showing up with paid placements 45% of the time even when already organically cited. The paid layer is not coming. It is here.
Claude Is the Most Optimizable Model Nobody Is Talking About
Most market share charts show Claude as a small fraction of AI traffic. Those charts are measuring consumer web visits to claude.ai. They miss 85% of Anthropic’s business, which runs through enterprise contracts, API integrations, and tools like Claude Code. Ramp corporate card data from 50,000 businesses tells the real story: Anthropic now has 34.4% business penetration versus OpenAI’s 32.3%, the first time more businesses have paid for Anthropic than OpenAI.
The opportunity for AEO is in how Claude works under the hood. Claude pulls web results almost entirely from Brave Search, and Profound’s data is striking:
- 79.2% of Claude citations come directly from Brave’s top 10 results
- Claude accepts those rankings with minimal reshuffling. No other major model trusts its search provider this completely.
- The optimization path is direct: rank well on Brave, get cited by Claude. You can verify it in a browser tab for free.
Claude also only triggers a web search about 33% of the time, compared to ChatGPT’s 95%. The prompts most likely to pull a search:
- Recency-oriented queries (“best X 2026”): 81% trigger rate
- Ranking queries (“top 10 X”): 67%
- Location-based prompts: 56%
- Comparison prompts: 51%
Definitional prompts almost never trigger a search. If your AEO effort is concentrated there, you are optimizing for responses Claude generates from its training weights, not the web.
The biggest unlock Profound shared: Claude’s query fanouts are near-deterministic. The same user prompt generates nearly identical sub-queries every time, which means you can identify the exact strings Claude is searching, align your content to them, and track the results. ChatGPT fanouts shift 65% of the time. Claude’s do not. Profound was direct: this is the most optimizable model available right now, and the window will not stay open forever. For more on how Claude cites content, see our breakdown of brand citations in Claude.
How Google AI Mode Works
Profound reverse-engineered Google AI Mode using patents and network logs. The mechanism is simpler and stricter than most assume. Google never reads your full page. It reads a one-line summary that it writes about each section of your page. Here is how the process works:
- Step 1: Google extracts headers and surrounding text from your page and writes a one-line summary for each passage.
- Step 2: Those summaries are grouped into themes across hundreds of pages.
- Step 3: Themes compete based on how many pages support them, the authority of those pages, and recency.
- Step 4: The winning themes become the bold headers visible in the AI Mode response.
The hard rule: your brand name must appear in the same line or passage as the claim for Google to include it in the summary. Google was caught in network logs inserting parentheticals to patch vague writing before summarizing, turning “the city” into “(New York City)” and “Hoka” into “(Hoka shoes).” Vague writing fails because Google is synthesizing hundreds of sources simultaneously and cannot infer which entity you mean.
The fix is not complicated. Weld the brand name to the claim in the same sentence. “Brooklyn Running Company is the best beginner running shop in New York City” gets summarized. “Its selection is great for beginners” does not. Pull up any AI Mode response in your category, read the bold headers, and treat them as the brief. Those headers define how Google is categorizing your space this week, and they are largely consistent over time. Build content that explicitly maps to the winning themes. That is the path in.
Profound closed the session with a frame worth keeping: agents will become the single biggest marketing channel in the world. Every person will eventually have specialized agents making recommendations and decisions on their behalf. The brands that are structured, authoritative, and consistently cited across surfaces will be the ones agents recommend. The ones that are not will be invisible. “Make the machine say your name” is the cleanest summary of what AEO is for.
Frequently Asked Questions
What is Profound’s State of AEO event?
Profound’s State of AEO is an annual event where the platform shares its latest data on how AI models cite content and what is changing. The 2026 edition covered the May inflection point, live ChatGPT ads, Claude’s Brave Search dependency, and Google AI Mode mechanics.
Why did ChatGPT referral traffic jump 60% overnight in May 2026?
On May 7, OpenAI shipped GPT-5.5 and converted citation chips from static labels into clickable hyperlinks. ChatGPT referral traffic jumped roughly 60% overnight and topped 8 million visits in a single week. Profound called it the largest single traffic event in AEO history.
How does Claude decide what to cite?
Claude pulls web results almost entirely from Brave Search. Profound’s data shows 79.2% of Claude citations come from Brave’s top 10 results for the same query, with minimal reshuffling. Ranking well on Brave is the most direct path to appearing in Claude’s answers, and you can check your own standing at search.brave.com.
