The old search funnel was a polite little museum tour: rank, click, maybe convert if the visitor did not wander off to compare toaster reviews. Agentic search is more like sending a sleep deprived intern with a browser, a budget, and suspicious confidence. It does not just surface your business, it opens things, reads them, and decides whether you are worth the next step.
That is the useful provocation in AX is the New AEO, an arXiv paper from ora research. The claim is not that SEO is dead, because nothing in marketing ever dies, it just gets rebranded and invited to more webinars. The claim is sharper: answer-engine optimization gets you noticed, but agent experience decides whether the machine can actually do anything with you.
Ora research says the citation is not the finish line
The arXiv paper AX is the New AEO, dated 28 Sep 2026 and authored by Ido Finder, Assaf Elovic, and Gad Shalev of ora research, frames the shift plainly. According to the paper, AI advice moved from seeding model training knowledge in 2023 to optimizing for live web search, where AEO pushes businesses to spread signals across forum threads, listicles, and off-site citations. But ora research argues that being surfaced is no longer sufficient because an agent opens results, reads them before deciding, and may run several rounds of search and fetch for one buyer question.
@title From AEO to AX
@source AX is the New AEO
Buyer question
│
▼
Search and fetch
│
▼
Results opened
│
▼
Business site read
│
▼
Outcome decided
@caption Agents move from finding results to reading sites before deciding.
That is a subtle but nasty little systems problem. AEO can get your name into the candidate set, which is helpful in the same way a restaurant sign is helpful. AX asks whether the agent can parse your menu, understand availability, find pricing, and avoid mistaking your contact form for a CAPTCHA themed escape room.
The experiment is annoyingly practical
According to ora research, the study ran 37,927 agent journeys, each built around a buyer question about a business. The authors say those journeys ran across four independent harnesses over 1,056 real businesses, with businesses matched on fame, prior model knowledge, and two AEO proxies before being split by AX level. That design matters because the paper is trying to isolate whether agent readability adds signal beyond merely being known or citation friendly.
This is not a benchmark beauty pageant where everyone claps because a bar chart wore a tuxedo. The interesting part is the unit of analysis: a buyer question turning into repeated search and fetch behavior. If your product page looks good to humans but collapses when an agent tries to extract constraints, compare options, or find the next action, your funnel has developed a robot shaped pothole.
Practitioners already see AX as more than formatting
Retoba describes agent experience as extending external AI readiness to personal AI assistants acting as authorised agents that understand and use services on people’s behalf. That lines up with the ora research framing, where the decisive step is not just whether an answer engine recommends you, but whether an agent can fetch and read your own site. In other words, AX is not glitter on top of AEO, it is the part where the agent stops admiring your citations and starts poking your infrastructure with a stick.
Insivia’s AEO and GEO resource gives a more tactical clue, describing extractable formatting as key information presented in tables, lists, or clearly marked sections that AI can extract and cite. That sounds pedestrian until you remember that most websites treat crucial buyer information like a side quest hidden behind modals, carousels, and copy written by a thesaurus trapped in a sales kickoff. Clean structure is not boring, it is affordance design for nonhuman readers.
What builders should change first
Ora research’s core lesson is simple enough to be uncomfortable: optimize the page the agent lands on, not only the places that mention you. If an agent must answer a buyer question, it needs fetchable content, readable structure, and enough explicit information to compare, decide, or proceed. Your beautiful brand narrative can stay, but give the agent something less interpretive than vibes in a velvet blazer.
A practical AX pass starts where Insivia’s extractable formatting point starts: make the important facts easy to extract. Put pricing, eligibility, integrations, policies, product differences, and next actions in clearly marked sections. Then test your site the way an agent would, by asking a concrete buyer question and watching whether the system can reach your page, read the relevant details, and form a sane answer without spelunking through decorative fog.
For readers building products, content, or APIs, the next watch item is whether agent platforms reward sites that are not just authoritative but operationally legible. AEO got brands into the answer box. AX is about surviving what happens when the answer box grows hands.