AI Made It Cheap to Make Things. It Made It Expensive to Be Seen.
Everyone can create. Almost no one gets seen. Chatbots favor the names they already know, and the same few companies sell the tools, the ads and the answers.
Ivana Nazareno: AI, power and the mindPart of a story we are following · 1 article · See every update
In brief
- AI made it almost free to make things and more expensive than ever to be seen.
- Chatbots tend to recommend the names they already know, and judge newcomers harder. Studies from 2024 to 2026 measure that brand bias.
- The same few companies sell the tools, the ads and the answers. Small businesses, writers and local papers pay at every booth.
Try this tonight. Ask a chatbot to recommend a romance novel. Then ask it for a good accountant, a project-management app, a sunscreen. Write down the names.
You’ll notice something curious: it almost always gives you names you’ve already heard. Well-known companies.
You might think it’s a one-off glitch. It isn’t. In a way, it’s the machine doing exactly what it was built to do. And it’s the quietest form of power I know.
Now picture a novelist whose book is frankly better than half of what the chatbot just named. It’s on Amazon. It has eleven reviews, and four of them are from her friends. She did everything right. The machine has barely heard of her, so it doesn’t say her name.
When a friend recommends a bestseller, you know it’s just a friend with taste. When an AI does it, the answer arrives in a calm, confident, even voice. It sounds like a reliable fact. It sounds like nobody decided anything.
Somebody did. Several somebodies, actually, and they keep being the same few companies.
I’ve spent months trying to put my finger on what bothers me about this. I think I finally found the shape of it.
The GPS That Owns the Restaurants
Let’s keep playing with hypotheticals. Picture a road trip. You ask your GPS for a place to eat. Now picture that the company behind the GPS also owns the highway, runs the toll booths, rents out every billboard on the shoulder and has a stake in some of the restaurants.
The GPS is still polite. It still says “In 500 feet, turn right.” It just happens to suggest you turn toward the places it knows best, which tend to be the places that already paid to be known.
That’s roughly where we are with AI. Here’s my argument in one sentence: AI made it almost free to make things and more expensive than ever to be seen, and the companies that own the seeing are, more and more, the same ones that sold us the making.
Think about what you can make now. A business plan in an afternoon. A logo, an app, a first draft of a novel. The tools are real. I use them every day, and I used them to help research and edit this piece.
But making stuff was never the hard part for long. The hard part is getting a stranger to notice. And the noticing is now run by a remarkably small club.
I’m not here to argue that AI didn’t democratize access. I’m here to show you where the toll booths are.
The Answer Has No Page Two
In 2004, Wired editor Chris Anderson made a hopeful prediction he called the Long Tail. With infinite shelf space, the internet would let the small and the strange find their audience. Hits would matter less. Niches would matter more.
For a while, it sort of worked. Search gave you ten blue links, and if you were patient, a page two. Page two is where the weird little bookstore lived.
AI answers don’t have a page two. They have a paragraph.
The Pew Research Center tracked the browsing of 900 American adults in 2025. When Google showed an AI summary, people clicked a regular search result in 8% of visits, compared with 15% when there was no summary. They clicked the links inside the summary itself just 1% of the time. (Pew Research Center)
Think about what that means for our novelist. In the old world, she had to fight her way to page one. In the new one, she has to be inside the paragraph. There’s no shelf anymore, just one voice reading you a short list.
The long tail wasn’t cut off. It just stopped being mentioned.
I truly believe the problem isn’t artificial intelligence. It’s the mold they’re pouring it into. As a technology, a language model has a breathtaking capacity to read everything, connect the dots and find hidden gems. But train it on an internet dominated by the biggest players, set it up to give the safest, most profitable consensus answer, and you’re forcing it to think like a risk-averse middle manager. They gave it a brain capable of discovering the whole world, then strapped commercial blinders on it.
Famous Until Proven Otherwise
Here’s the part that surprised me. AI doesn’t just recommend the big names more often. It seems to judge the small ones harder.
Try it. Paste a proposal from a small consulting firm into a chatbot and ask for an honest review. You’ll get a careful list of risks. Is the firm too small? What happens to you if it goes out of business? Does it have the capacity?
Now ask about hiring IBM. Nobody in that answer wonders whether IBM will vanish next year. Nobody asks whether you’ll get locked into a contract, or whether your account will matter to a company that size. The big name gets the benefit of the doubt. The small one gets a background check.
There’s an old corporate saying: “Nobody ever got fired for buying IBM.” It was never really about quality. It was about cover. Picking the famous option protects the person doing the picking. We’ve now built a machine that hands out that advice to hundreds of millions of people, in a “neutral,” polished voice.
This isn’t just a hunch. Researchers have started measuring it.
In 2024, a study presented at a major language-technology conference found that AI models consistently linked global brands with positive traits far more than local ones (EMNLP 2024).
In June 2026, researchers testing skincare products found that when the products were identical on paper, three commercial chatbots recommended the well-known brand every single time. It’s an early study that hasn’t been peer-reviewed yet, but the pattern is hard to miss (arXiv).
And an August 2026 audit of six leading chatbots, from ChatGPT to Mistral, found that in 20 of 100 product and service categories, one product came first in more than three out of four answers. Some models also leaned toward products from their own corporate family (NET Institute).
To be fair, the June study found something hopeful: the famous brand’s edge disappeared when a rival had even slightly better ratings. The brand works as a tiebreaker. But that’s exactly the trap. A business with eleven reviews doesn’t have the ratings to break the tie. And that doesn’t mean it’s worse. It’s just new.
Banks have a name for this. Someone with no credit history isn’t a bad borrower. They’re a “thin file.” The system can’t see them, so it treats them as a risk anyway. The AI economy is turning every new business, every first novel and every unknown name into a thin file.
In 1968, the sociologist Robert Merton called an older version of this the Matthew effect, after the line in the Gospel: to everyone who has, more will be given. He was writing about famous scientists getting extra credit. He’d recognize the chatbot instantly.
Everyone Got a Printing Press. Nobody Got a Newsstand.
In 1971, the economist Herbert Simon wrote that “a wealth of information creates a poverty of attention.” He was thinking about office memos. He might as well have been describing 2026.
When it gets nearly free to make things, the world fills up with things. More apps, more e-books, more online stores, more ads. Your attention doesn’t grow. You still have 24 hours in a day and one pair of eyes. So attention becomes the scarce resource, and whoever owns the scarce thing sets the price.
Who owns your attention? According to the research firm eMarketer, Meta, Google and Amazon will take 62.3% of all the money spent on digital advertising worldwide in 2026, and that share is expected to keep growing (eMarketer).
The price keeps climbing, too. In the third quarter of 2026, Meta’s average price per ad rose 14% worldwide compared with a year earlier (Meta), and 23% in the U.S. and Canada, according to the company’s earnings presentation. Meta’s finance chief credited better ad performance and a stronger economy, and Meta says its AI is a big reason ads perform better.
Let that sink in. The AI makes your ad work better. Then the platform charges you more for the ad. The efficiency doesn’t stay with you. It travels up the road to the toll booth.
My favorite detail: Meta reported that advertisers who took the advice of its AI business assistant got almost 5% more conversions at the same cost per conversion, and raised their ad spending by 9% (exchange4media). Good news for the advertisers. Even better news for Meta.
About two hundred years ago, the economist David Ricardo spotted the same pattern on farms. When farming got more productive, farmers didn’t get rich. Landlords did, because they owned the land everyone needed and could raise the rent. Today the land is your attention. The landlords have better lawyers.
The Toll Road Has Two Booths
So say you’re small and you have a new idea. Here’s your road.
First booth: you pay to build. You rent intelligence by the token from one of a handful of AI companies. A big company can run thousands of AI agents overnight to test its prices, its sales copy, its contracts. You have the $200-a-month plan and a usage limit. It’s a chess match where one player gets five minutes on the clock and the other gets five days.
Second booth: you pay to be seen. You buy ads on Meta or Google, where prices keep rising because everyone else got the same cheap tools and is now shouting into the same feed.
Notice who collects at both booths.
This year, Google, Amazon, Microsoft and Meta plan to spend about $725 billion on capital expenditures, mostly AI data centers. That’s up 77% from last year’s record (Tom’s Hardware, citing the Financial Times). That number isn’t a budget you compete with. It’s a budget that decides who gets to compete.
There’s a quieter trick, too. In January 2025, the Federal Trade Commission published a study of the deals between the biggest cloud companies and AI labs like OpenAI and Anthropic. Among other things, it found that the labs committed to spend large chunks of the investment money on the investor’s own cloud services (FTC). I invest billions in you. You spend it renting my computers. Lovely business model.
Then there’s the booth inside the answer itself. In February 2026, OpenAI started showing ads in ChatGPT to U.S. users on its free and cheapest paid plans. The ads sit below the reply, labeled “Sponsored,” and OpenAI says they don’t influence the answers (OpenAI). Maybe so. But the answer machine now has two customers: you, and whoever pays to sit right under its reply.
And as I was finishing this piece, OpenAI opened something new. On November 10, 2026, it launched business profiles inside ChatGPT, letting companies claim a verified listing, update their information and see how often the chatbot mentions them (TechCrunch). For a small business, a free profile could genuinely help. But it also puts one more piece of the shop window in the hands of the company that writes the answers. Whether it opens the door for the little guy or concentrates the power to be seen even further is exactly what I’m analyzing now, and I’ll be watching closely.
Meanwhile, a whole industry has grown up around those answers. It’s called GEO, or generative engine optimization, and its whole job is getting brands mentioned by chatbots. The June 2026 study modeled what happens when every brand plays that game: the advantage for any single brand nearly vanishes, and brands that don’t play get zero recommendations.
In Through the Looking-Glass, the Red Queen tells Alice that “it takes all the running you can do, to keep in the same place.” That’s GEO. Everyone pays to run. Nobody gets ahead. Somebody sells the shoes.
And the shoe store never loses. The giants do take real risks with that $725 billion. But they spread the risk across thousands of bets. If your startup wins, you paid them. If it fails, you also paid them. You’re betting your whole business. They’re betting a sliver of a portfolio.
Twenty Thousand Sandwiches
If you write for the internet, there’s one more toll, and it’s the strangest one. You pay with your own work.
For two decades, the web ran on a simple deal. Google’s crawler read your pages, and in return it sent you readers: roughly five pages read for every visitor sent back. Not generous, but fair enough to build a business on.
AI crawlers broke the math. In 2026 measurements based on Cloudflare data, OpenAI’s bots read roughly 1,000 to 1,300 pages for every visitor they sent back. Anthropic’s bots (yes, the company that makes Claude, one of the tools I used for this piece) read anywhere from about 10,000 to more than 20,000. Meta’s sent back no one at all (Technology Checker, from Cloudflare Radar).
Picture a customer who walks out of your deli with 20,000 sandwiches and pays for one. Now picture that he’s also the one who tells the whole town where to eat.
It feels like a robbery in three acts. First, they read your work to train their machines. Then they answer people’s questions with it. Then they toss you crumbs. Remember: people click the sources inside Google’s AI summaries about 1% of the time.
Legally, it may not be one. In June 2025, two federal judges in California ruled that training AI on books can be fair use, though one of them drew the line at pirated copies (White & Case). Fine. But “legal” and “fair” are two different words for a reason.
Why not just lock the door? You can block the bots, and then you vanish from the answers. Google makes even that hard. Penske Media, the owner of Rolling Stone and Variety, sued Google in 2025, arguing that there’s no way to keep your work out of its AI summaries without disappearing from Google Search altogether (eMarketer). In June 2026, Britain’s competition regulator ordered Google to offer a real opt-out, starting in December (Enterprise DNA). It applies in the U.K. So a blogger in Manchester will soon have a right that a small-town paper in Florida doesn’t.
Now the epilogue. Some publishers do get paid. News Corp reportedly gets more than $250 million over five years from OpenAI, and Amazon reportedly pays The New York Times $20 million to $25 million a year (LLM Pulse). News Corp’s CEO has described his company as “essentially an input company” (The Licensing Letter). Being the input pays well, if you’re big enough.
The regional paper doesn’t get a check. Neither does the blogger who spent 15 years building a site that every chatbot quietly learned from. They can’t negotiate. They can’t afford to sue. And they got hit hardest: according to Chartbeat data reported by Axios, small publishers lost about 60% of their search traffic over two years, while big ones lost about 22%.
I can’t prove what the big checks are really buying. Part of it is clearly quality: these are great archives. But it’s hard to miss that the publishers who get paid are also the ones who can afford to sue. The New York Times is still suing OpenAI and Microsoft. The blogger can’t afford the filing fee.
The final twist comes from the Brookings Institution. In June 2026, its analysts pointed out that the same Big Tech firms whose AI is draining publishers’ traffic are now building and running the marketplaces where publishers go to license their work (Brookings). The company that took your customers also runs the window where you line up to get paid.
There’s a practical problem buried in all this, too. If the people who produce fresh reporting go broke, the machines will have nothing new to read. And if the only sources left in the answer are the ones with licensing deals, “the answer” quietly becomes the voice of a few big media companies. That’s not just concentration of money. It’s concentration of what counts as true.
The Best Argument Against Me
I owe the other side a fair hearing, because it has a good case.
It goes like this. AI really did lower the cost of starting something. One person can now do what took a team of ten a few years ago. Ad prices are rising partly because ads work better. Recommending a company with decades of track record isn’t bias, it’s prudence. And if you don’t like the big models, open models and falling prices give you options.
All of that is partly true. Here’s my answer.
The floor did rise. The ceiling rose faster. Everybody got a car, and the people who already had cars got planes. When everyone’s tools improve at once, the tools stop being an advantage. The advantage moves to whoever controls the scarce thing. And the scarce thing isn’t intelligence anymore. It’s being seen.
Take open models. They’re real, and they do cut the first toll: anyone can download one and run it cheaply. But nobody has open-sourced attention. You can build your app for almost nothing and still have to pay Meta, Google or Amazon so that anyone ever sees it. The bottleneck didn’t disappear. It moved.
Prudence is fine for one buyer. Automated for everyone, it becomes a wall. If every assistant on Earth gives the cautious answer, the new company never lands the first customer it needs to stop looking risky.
The real problem isn’t that big companies win. It’s that the referee is also a player. The same few companies own the shop window, the ad space and the assistant that tells you where to look.
Courts have noticed. In April 2025, a federal judge ruled that Google had illegally kept monopoly power in parts of the online advertising technology market. In September 2026, she ordered changes and a compliance monitor, but declined to break the business up. Google’s argument against a breakup? That it would make it harder for small businesses to reach customers (Reuters via The Star).
I’m not making that up. The monopolist defended itself in the name of the little guy.
Wall Street got the message. Since Judge Amit Mehta issued his remedy in the separate search monopoly case, Alphabet’s stock had climbed 45% by September 2026, adding about $1.3 trillion in shareholder wealth, according to the Associated Press (AP). Plenty drove that rally, AI hype included. Still, investors were clearly not scared.
What Would a Fair Answer Machine Look Like?
I don’t think the answer is to ban anything. I think it’s to stop pretending the machine is neutral and to build in some humility. A few ideas, each with a price tag:
- Disclose the family ties. If a chatbot recommends a product made by its own parent company, it should say so, the way a financial adviser has to disclose a commission. The cost: friction, and lawyers arguing forever about what counts as “related.”
- Make “I don’t know much about this one” an honest answer, not a warning. When a model has little data on a small business, it could say so plainly instead of turning silence into suspicion. Better yet, give users a discovery mode: “Show me good options I’ve probably never heard of.” The cost: less confident answers, and some people will hate that.
- Separate the referee from the player. Antitrust law has done this before, with railroads that hauled their own coal and with the old Bell telephone monopoly. Courts have been reluctant to do it with Google. The cost: it’s slow, it’s political, and some products might get worse for a while.
And one for the people who write the internet: let small publishers bargain together. Musicians solved a version of this a century ago with ASCAP, which pools songwriters’ rights so nobody can pick them off one by one. A nonprofit called the RSL Collective is trying the same thing for the web, launched by Reddit, Yahoo and Medium and backed by more than 1,500 organizations (TechCrunch). The catch: for now, it only works if AI companies agree to play along. The cost: AI gets more expensive, and someone will call that a tax on innovation. Maybe. Free sandwiches are cheaper, too.
None of these is free. But the current setup isn’t free either. We just don’t see the bill, because it’s paid by people we never hear about.
Who Answered?
Remember the experiment from the beginning? Here’s part two.
The next time an AI recommends something to you, whether it’s a book, a lawyer or an app for work, ask one more question: “Who did you leave out?” Then notice how much longer the answer takes, and how much less sure it sounds.
Our novelist is still out there with her eleven reviews. Maybe her book is wonderful. Maybe it isn’t. The point is that we’ll never find out by asking a machine trained to repeat what everybody already knows.
So is the reporter at your local paper, whose story a chatbot may have summarized for you this morning without ever sending you her way.
AI could have been the best thing that ever happened to her: a tireless reader that has gone through every catalog, that can compare a first novel with a famous one on what actually matters, and that has no reason to be impressed by a big name. That machine is technically possible. It just isn’t the one being sold to us.
So here’s the question I can’t stop thinking about, and now, I hope, neither can you:
Are we building AI to help people discover what exists? Or to make the system that already decides who deserves to be seen run faster, cheaper and with a straight face?
Ivana Nazareno writes about who holds power in AI, the biases in its answers and the questions it raises about mind and consciousness. She created New Mind Museum, a free online museum of AI history, and is the author of the novels The Night We Came Back and Desire and the Thaw. More of her work: ivinazareno.com.
This is an opinion column. Views are the author’s own. Figures are as of November 2026.
Respond to this column. Do you see it differently, or has AI changed how customers find your business? Send a response of up to 300 words to [email protected] with “Response” in the subject line. We publish a selection, with your name and town.
Share or cite this page
AI Made It Cheap to Make Things. It Made It Expensive to Be Seen.. Emerald Coast Dispatch, November 20, 2026. https://emeraldcoastdispatch.com/okaloosa-county/2026/11/ai-cheap-to-make-expensive-to-be-seen/
<a href="https://emeraldcoastdispatch.com/okaloosa-county/2026/11/ai-cheap-to-make-expensive-to-be-seen/">AI Made It Cheap to Make Things. It Made It Expensive to Be Seen.</a>
Nazareno, I. (2026, November 20). AI Made It Cheap to Make Things. It Made It Expensive to Be Seen.. Emerald Coast Dispatch. https://emeraldcoastdispatch.com/okaloosa-county/2026/11/ai-cheap-to-make-expensive-to-be-seen/
Free to share or quote with credit and a link.
Sources
- Pew Research Center: Google users are less likely to click on links when an AI summary appears in the results
- Kamruzzaman et al.: "Global is Good, Local is Bad?" Understanding Brand Bias in LLMs (EMNLP 2024)
- Chu and Hou: Incumbent Advantage: Brand Bias in LLM Recommendation Systems (arXiv preprint)
- Boffa et al.: An Empirical Audit of LLM Product Recommendations (NET Institute)
- eMarketer: Meta to surpass Google in digital ad revenues for first time ever
- Meta Investor Relations: Meta Reports Third Quarter 2026 Results
- exchange4media: Meta's ad revenue jumps 26% as AI boosts conversions and ad prices
- Tom's Hardware: Big Tech's AI spending plans reach $725 billion
- Federal Trade Commission: Staff Report on AI Partnerships & Investments
- OpenAI: Testing ads in ChatGPT
- TechCrunch: OpenAI rolls out new business profiles in ChatGPT
- Technology Checker: robots.txt and AI crawl-to-referral ratios, from Cloudflare Radar
- White & Case: Two California district judges rule using books to train AI is fair use
- eMarketer: Google faces first major publisher lawsuit over AI Overviews
- Enterprise DNA: UK CMA orders Google to offer publishers an AI search opt-out
- LLM Pulse: AI content licensing deals mapped
- The Licensing Letter: News Corp's deal with Meta AI: "We're essentially an input company"
- Axios: Chartbeat data on search traffic declines for small publishers
- Brookings: Same gatekeepers, new tollbooths in the AI content licensing market
- Reuters via The Star: Google ad tech remedies and antitrust monitor
- Associated Press: Judge orders changes to Google's ad business
- TechCrunch: RSS co-creator launches new protocol for AI data licensing
