Latest with AI
Friday, November 20, 2026
Article

Surmado founder warns AI answers could be secretly rigged by the highest bidder

Surmado founder warns AI answers could be secretly rigged by the highest bidder

When billions of people ask ChatGPT or similar models a question, they expect a neutral, trustworthy answer. But a newly published peer-reviewed paper argues that the real danger isn’t a wrong answer—it’s a correct one whose hidden agenda you can’t see. The paper, published in Springer Nature’s AI and Ethics journal, calls this emerging field “answer-engine optimization” and warns it could let the highest bidder secretly shape what the machines tell us.

Luke F. Walton’s paper in Springer Nature’s AI and Ethics argues that even an accurate AI answer can hide whose judgment it carries.

Also worth a look: ShipCargo’s AI agents deliver firm quotes and zero deposit, but pickup dates remain estimates and PRAVAAH’s Vancouver summit to merge South Asian business leaders with sport, Dec 4.

We don’t want a world where AI answers silently go to the highest bidder. Small businesses need to be able to compete, and our democracy needs accountability to survive.”— Luke F. Walton, author of “The Captured Oracle” and founder of Surmado

For years, search engines returned a list of links, and the user decided which to click. Now, AI answer engines—like the summary at the top of a search results page or a chatbot—hand users a finished, polished answer delivered in the AI’s own voice. The paper poses a stark set of questions: Should I vote yes on this measure? Which bank has the better loan? Which contractor should I hire? Even if the answer is factually accurate, Walton contends, it can still obscure the fact that someone with a financial or political stake in the outcome shaped it.

The paper, “The Captured Oracle: Authorship and Agency in the Ethics of Answer-Engine Optimization,” is by Luke F. Walton, a San Diego-based researcher and software developer who also founded Surmado, a company operating in the very industry the paper examines.

Is answer-engine optimization manipulation? The paper defines AEO—closely related to generative engine optimization (GEO)—as the effort to shape what the AI says. This growing industry promises to “make ChatGPT recommend your brand.” While industries have long tried to “hack the algorithm,” the paper argues that AI is not a formula one can solve; it is an intelligence one tries to manipulate.

The moral wrong, Walton argues, isn’t influence itself. It’s the covert act of manipulation. A business that wants AI to correctly display its hours, prices, and services is simply trying to be found. That is fair, the paper contends, when what the AI ends up saying is true, the business behind it is transparent, and someone with a name will answer for it. The trouble begins when the shaping is hidden. A campaign, a bank, or a contractor quietly sets the terms the AI uses to pick a winner, and its preference emerges from the AI’s mouth sounding neutral. Large corporations are spending millions to covertly “win” AI results. What happens if a foreign country deploys the same techniques?

Can AI answers actually sway an election? A national survey released October 6 by the American Association of Political Consultants Foundation, weeks before the November 3 midterms, found that at least 38% of likely 2026 voters had already gotten election information from an AI search summary. Only about half of them were even aware it came from AI. The survey counted who is getting answers from AI; the paper asks who is shaping those answers. Walton opens the paper with a voter asking an AI what a ballot measure would do. One of the sources behind the answer was placed there by a group with a stake in her vote. He considers the potential for information warfare, laundered through a machine.

This is not new. Facebook ads were bought to target past elections. But AI answers could be manipulated silently, with less for the public to inspect afterward: no visible ad buy attached to the answer, no sponsor label. Subpoena the interaction logs and you may get a question and an answer. Nothing there necessarily says who placed the source the answer leaned on.

Getting the facts right isn’t enough. An AI answer can cite real sources, get every fact right, even label itself as AI, and still not tell you whose call it was. Even if the facts weren’t gamed, the decision and authorship were laundered and presented as the verdict of a “super intelligence.” That is why the obvious fixes fall short. A citation shows where a fact came from; it doesn’t show who decided which facts mattered. The labels Europe’s AI Act requires on some AI-generated content say a machine wrote the text; they don’t say whose judgment it carries.

The problem gets worse as AI starts doing things for you instead of just answering. Imagine an assistant that reads your email, picks a contractor, and sends the reply while you’re busy. If someone plants a document for that assistant to find, and it works, their preference becomes your decision. When you act on an AI’s answer, whoever shaped it owes you an explanation. When an AI agent acts for you, the agent can’t be held accountable for its actions. The paper puts that responsibility on whoever chose to let it act.

Walton’s proposed remedy is called “answerable legibility.” Every judgment an AI delivers should have an owner—someone with a name who says what they stand to gain and can be asked why. It isn’t about less AI or halting progress. “The remedy asks no one to stop building and no one to trust less,” he writes. “We don’t want a world where AI answers silently go to the highest bidder,” Walton said. “Small businesses need to be able to compete, and our democracy needs accountability to survive.”

“I wrote this paper because I was grappling with it in my own crowded industry — what kind of business do I want to build with my team,” Walton said. “We make small businesses legible to AI. We don’t hack ChatGPT.” Walton builds this way himself. Answer Engine, his open-source search tool, answers only from an archive its owner chose, shows the sources it actually used, and says “I don’t know” when the archive has no answer.

About Luke F. Walton: He is an independent researcher and software developer based in San Diego, California. He is the founder of Surmado, Inc., a company that uses AI to build and manage websites and online presence for small businesses. His research asks who is responsible for what AI systems say and do. “The captured oracle” was published November 18, 2026.

Walton, L. F. (2026). The captured oracle: Authorship and agency in the ethics of answer-engine optimization. AI and Ethics, 6, Article 665. https://doi.org/10.1007/s43681-026-01532-2

Read free (view only): https://rdcu.be/p9Mw9GZpQPna
Key terms and accepted manuscript: https://lukefwalton.com/writing/the-captured-oracle/
AAPC Foundation AI Voter Study: https://theaapc.org/release-foundation-research/
Answer Engine source: https://github.com/lukefwalton/answer-engine

Rachel McCauley
Surmado
+1 602-341-5572
Visit us on social media:
LinkedIn
YouTube
X

Why it matters: As AI-generated answers become the default source of information for millions of voters and consumers, the paper’s call for “answerable legibility” demands a new standard of transparency. Without a clear owner for every AI judgment, the public remains vulnerable to hidden manipulation that could quietly influence elections, purchasing decisions, and even personal safety—all without a trace of who is pulling the strings.