Optimized for the Machine

New Acronym on the Block, AI Edition

There was always an algorithm deciding what you’d see. The new one just does the reading for you, too.

Here’s something we should probably acknowledge before we go any further: there has never been a moment in the history of the internet when you were seeing everything. Not once. Not ever.

Search algorithms were always making choices—about what counted as relevant, what counted as authoritative, what went on page one and what got buried in the informational equivalent of a filing cabinet in a locked basement. An entire industry—search engine optimization, SEO, a phrase that sounds clinical because it is—grew up around the project of reverse-engineering those choices. For twenty-five years, if you wanted your website to be found, you didn’t just write something good. You wrote something good—but you also ensured readability hit baseline. (There were scores for this. Flesch-Kincaid. An eighth-grade reading level was, for all practical purposes, non-negotiable.) You used the right keywords—but not too many. You wrote compelling text—but not too much. You got the right backlinks. You didn’t become a linkfarm. You structured your headings in the right way.

To be fair, a lot of this made the internet better. Readable content is better for users. Clear structure is easier to navigate. Google will tell you—and they’re not entirely wrong—that what their algorithm preferred and what readers preferred often pointed in the same direction. But those preferences didn’t always overlap perfectly, and an entire economy grew up in the gap between them. Not because anyone was being nefarious. Just because when a system decides the order, people will learn the system.

Layer on paid placements and it’s not until page... something or other that you arrive at a result no one paid for or optimized for. By then, most people have stopped looking. (Most people stopped looking after result three. Google knew this. Everyone knew this. That was the whole game.)

The point isn’t that this was bad. The point is that the results were never just the results. They were the output of a system that ranked, filtered, and prioritized before you ever saw them. You got ten blue links. You knew they were in an order. You could scroll. You could choose. You could develop a healthy suspicion about whatever showed up first.

I bring this up because there is now a new version of this story, and it is being told with the breathless urgency of something unprecedented. It is not unprecedented. But it is different in one important way.

The old engine ranked the answers. The new one writes them. That’s a meaningful difference, and it’s worth sitting with for a second—not because it’s scary, but because it changes the relationship between the person asking and the system answering.

When Google gave you ten blue links, you were doing the synthesis yourself. You clicked, you read, you compared, you decided. The algorithm chose the order, but you chose what to do with it. When ChatGPT or Perplexity gives you a single, fluent, sourced paragraph, that synthesis has already happened. The AI read the sources, extracted what it determined was relevant, and assembled an answer on your behalf. It’s convenient. It’s often good. And it’s a fundamentally different kind of mediation than a ranked list—because the choices that went into it are no longer visible to you.

Once answers became a product, visibility became a billable line item.

Which brings us, inevitably, to the acronyms.

If you work in or adjacent to digital marketing, you’ve probably already heard that this shift requires an entirely new discipline. You may have heard several names for it.

GEO (Generative Engine Optimization). AIEO (AI Engine Optimization). AEO (Answer Engine Optimization). LLMO (Large Language Model Optimization). AIO, SXO, ASO, AISO, AISEO, AI SEO, and GSO. Search Engine Land counted more than ten in a single study. Rand Fishkin—who has been doing this long enough to have earned the right to be exasperated—advocates for “Search Everywhere Optimization.” Google’s own John Mueller offered, in August 2025, what I consider the definitive take: the higher the urgency and the stronger the push of new acronyms, the more likely it’s spam and scams.

The urgency is considerable. Agencies are charging $2,500 to $30,000 per month for “GEO services.” At least one firm offers “AIEO audits” starting at $95,000. A startup called Profound reached a valuation north of $100 million around its first birthday and has since gone on to raise at a reported $1 billion valuation. Tools claim to “do GEO for you.” Vendor-sponsored surveys show numbers like 97% positive results, which is the kind of claim that should make you check whether the survey was conducted by the vendor’s marketing department. (It was.)

I’m going to use GEO for the rest of this piece because it has the most academic credibility, and because if I type “AIEO” one more time my fingers are going to stage an intervention. But I want to walk through what the actual research says—because when you do, something interesting happens. Something the $95,000 audit doesn’t want you to notice.

First, the context. Google still processes on the order of 8–14 billion searches per day, depending on whose estimates you use. It still commands roughly 90% of global search market share overall, and in many mobile segments that figure creeps toward 95%. All AI tools combined—ChatGPT, Perplexity, Claude, Gemini, all of them—account for a tiny fraction of total internet activity, well under 5% of global query volume. Google alone sends vastly more traffic to websites than ChatGPT, Gemini, and Perplexity combined; one recent analysis put the gap at about 345× more visits.

So: SEO is not dead. The twenty-five-year practice of optimizing for search still matters more than anything else in this piece.

But the trend lines are moving. AI referral traffic grew 357% in a single year, from June 2024 to June 2025. For news and media sites, it surged 770% over the same period. That’s still a small slice of overall traffic, but it’s growing orders of magnitude faster than organic search referrals. Zero-click searches—the ones where you get your answer without ever visiting a website—are rising across every measurement. Pew Research tracked nearly 69,000 real searches from 900 adults and found that when AI summaries appeared, users clicked traditional results only 8% of the time, compared to 15% without them. Ahrefs found that AI Overviews correlate with about a 58% drop in click-through rates for content in the number-one position as of late 2025.

The shift is real. It’s worth understanding. And there is, it turns out, genuine academic research on what to do about it.

GEO has one thing most marketing buzzwords don’t: a real academic paper.

In November 2023, a team of researchers led by Princeton University and IIT Delhi, with collaborators from Georgia Tech and the Allen Institute for AI, released a paper that, for all the hype that followed, was actually quite measured. They built something called GEO-bench—a set of 10,000 search queries—and tested nine different strategies for making content more likely to appear in AI-generated responses. The work was later presented at ACM SIGKDD 2024 in Barcelona, one of the world’s top data science conferences, and has since been followed by additional research with broadly similar findings.

The findings were specific and, in some cases, genuinely interesting. Three strategies worked consistently across every domain they tested: adding citations to credible sources (which boosted visibility by up to 40%), including direct quotations (about 37%), and incorporating statistics (roughly 22–30%). The most effective combination was improving the writing’s fluency while adding statistics—which outperformed any single technique.

The results were also domain-specific in ways that matter. Authoritative language worked best for historical content. Statistics had the biggest impact in law and government topics. Quotations performed best in people and society contexts. There was no universal formula—which is both frustrating and, if you think about it, kind of encouraging. It means the system is responding to what’s actually appropriate for the subject, not just pattern-matching on formatting tricks.

Two other findings stood out. First, simply improving the fluency and readability of source text—without changing the substance at all—boosted visibility by 15–30%. The researchers noted that generative engines appear to value presentation, not just information. Second, a more persuasive or authoritative tone made almost no difference. The AI wasn’t swayed by confidence. It was swayed by clarity, evidence, and structure.

And keyword stuffing—once the bread and butter of early SEO, though Google has penalized it for years now—offered no improvement in AI visibility either. That’s less surprising than it would have been a decade ago, but it does confirm that the old playbook doesn’t transfer. What worked for search engines in 2005 doesn’t work for AI in 2026, and what works for Google in 2026 doesn’t necessarily work for ChatGPT in 2026 either. The systems are related but not identical.

GEO techniques also benefited lower-ranked websites far more than dominant ones. The “cite sources” method led to a 115% visibility increase for websites ranked fifth in traditional search, while top-ranked sites saw their visibility decrease by about 30%. This is one of the very few dynamics in digital marketing that actually favors underdogs, and it’s worth paying attention to.

Now. Reread that list of strategies. Cite credible sources. Include statistics. Write clearly. Structure your content well. Answer questions directly. Make it readable.

Does any of that sound... new?

Here is, honestly, the single most useful takeaway in this entire piece, and it’s the one that undercuts about $95,000 worth of consulting:

The best GEO strategy is good SEO.

There’s no real trade-off between the two. Nothing that helps your content get cited by AI systems will hurt your Google rankings. The strategies that the Princeton research identified as most effective for AI visibility—clear structure, cited sources, statistics, direct answers to questions—are also just good SEO practice and, frankly, good writing practice. As one SEO veteran put it: ask your favorite GEO expert for 25 things that are unique to AI search and don’t overlap with SEO, and they’ll block you.

There are a few AI-specific nuances, but they’re mostly about extractability. Think sections that start with a direct answer to a question, short definitions that can be lifted verbatim, tables or lists that package related facts cleanly, and headings that echo the way people actually phrase queries. You’re still writing for humans, but you’re also making it easy for a summarizer to grab the right paragraph and drop it into a synthesized answer.

The “new discipline” is largely the old discipline done well, with some additional attention to structure and extractability. You’re still optimizing for search. The work just has secondary benefits in AI citation now, too. Do the things you were already supposed to be doing, and they’ll now pay off in two places instead of one.

That’s less a revolution than a reinforcement. It’s not nothing—the extractability piece is genuinely new, and the shift in how AI selects passages versus how Google ranks pages matters at the margins. But anyone telling you to throw out your SEO playbook and start over is selling you a problem you don’t have.

So if the tactics aren’t new, what is?

The tactics are a continuation. The mediation is different. And that’s the part worth paying attention to—not because it changes what you should do, but because it changes what happens after you do it. For users, getting a fast, synthesized answer is often a genuine improvement—it just comes with tradeoffs worth noticing.

Here’s what I find most interesting about all of this—more interesting than the tactics, more interesting than the acronym wars. People think the AI is neutral. And it isn’t. And the way it isn’t neutral is, by design, very hard to see.

A January 2026 study from the Center for News, Technology & Innovation tracked how people actually use chatbots for information. The researchers didn’t just ask them—they had participants walk through their chat histories and demonstrate their use cases. What they found was striking. Users consistently described chatbots as “neutral” and “balanced”—especially compared to traditional news media, which they viewed with suspicion. They rarely discussed algorithmic bias. They rarely acknowledged that news publishers were being cited in their conversations. Many took the mere existence of citations as proof of accuracy—without clicking through to check whether the sources were being represented correctly. One U.S. interviewee captured the vibe perfectly: “You can always double check if you want to.” He told researchers he never felt the need.

This tracks with broader findings. A PsyArXiv study from early 2026 found that users perceive AI chatbots that agree with them as unbiased, while chatbots that challenge them are rated as highly biased. The researchers called this “blindness to sycophantic AI”—we mistake validation for objectivity. ARTICLE 19, the free expression organization, demonstrated that ChatGPT will shift its characterization of news sources within just a few exchanges if you introduce subtle bias in your questions. Ask it neutrally about the New York Times, you get balanced information. Start hinting it’s liberal, and it starts emphasizing conservative criticism. It’s a mirror, not a window.

And here’s the structural layer underneath all of that: when AI chatbots search the web for you, they’re often starting from the same ranked search indexes that already prioritized the results.

ChatGPT’s web mode runs primarily through Bing’s search index. Perplexity uses a hybrid of its own index with Bing’s ranking signals, and various analyses have shown that Google’s own AI features obviously start from Google Search. These AI tools aren’t starting from scratch when they “research” your question. They’re starting from search results that have already been filtered, ranked, and shaped by the same algorithmic priorities that have governed what you see online for twenty-five years.

So when the AI gives you a synthesized, confident, well-cited answer, what it’s actually done is: taken the already-prioritized results from a search engine, selected the passages it predicts will be most useful, and assembled them into a paragraph that reads like it came from an impartial analyst. The prioritization happened before the synthesis. The synthesis just hid the seams.

This is not a conspiracy. It’s not even a flaw, exactly. It’s architecture. And it means that when you feel like the AI “did the research for you,” what it actually did was summarize somebody else’s ranking of what matters. The AI’s confidence is a design choice. The citations are a design choice. The absence of alternatives you didn’t see is a design choice. None of it is neutral. It just feels that way—which is worth noticing.

There’s one more layer here, and then I’ll let you go.

The old deal—the SEO deal—was imperfect but functional. Google crawled your site, indexed it, and when someone searched for something relevant, it sent them to you. You got the traffic. The traffic funded the content. The content fed the index. Circular, maybe. But the loop closed.

The new deal doesn’t close—not yet, anyway. Cloudflare tracked the crawl-to-referral ratios—how many times an AI system visits a website to ingest content versus how many times it sends a user back. As of June 2025, Google’s ratio was roughly 14:1. OpenAI’s was about 1,700:1. Anthropic’s was around 73,000:1. Put differently: for every one human visit, OpenAI’s crawler hit a page around 1,700 times, and Anthropic’s around 73,000 times. That’s a ratio worth noticing. It means the AI is extracting value tens of thousands of times for every instance it sends a reader back to help pay for it.

The publisher response has split into two camps. One is licensing: News Corp signed a deal with OpenAI reportedly worth more than $250 million over five years; Dotdash Meredith signed a deal worth at least $16 million per year. The other is litigating: the New York Times lawsuit against OpenAI is proceeding, and by early 2026 more than 70 copyright lawsuits against AI companies are pending in U.S. courts. Small publishers, independent creators, niche blogs, recipe sites, local businesses? They get neither the licensing deals nor the legal resources. They just get less traffic.

This is the unresolved tension underneath the entire GEO conversation. You can optimize your content for AI citation all you want—and you should, because the research says it works and it doesn’t cost you anything. But the broader question of who pays for content creation when AI absorbs the traffic that used to fund it? That one’s still being figured out.

So: is GEO real? Yes. The shift in how AI mediates discovery is genuine, measurable, and accelerating.

Is it a new discipline? Mostly, no. It’s SEO—done well, with attention to structure and extractability—that now pays off in an additional channel. The revolution is a reinforcement. The best thing you can do for your AI visibility is the same thing you should have been doing for your search visibility all along: write clearly, cite your sources, include data, structure for readability, and build a brand people recognize.

Is there something genuinely new here? Yes—but it’s not the tactics. It’s the shift in how people encounter information. The old system showed you options and let you choose. The new system makes the choice and shows you the result. The old system was visibly mediated. The new system feels like it isn’t—which is the most important thing to understand about it.

Use the tools. They’re genuinely useful. I use them every day. But notice the difference between a tool that helps you find things and a tool that finds things for you. That difference—between visible mediation and the feeling of its absence—is where the most interesting questions live. Not just for search. For how we encounter information at all.

How have you feel about the recommendations from the chatbot of your choice?

Leave a comment

Thanks for reading Opinions & Conditions May Apply!

Share