As AI platforms transfer towards ad-based monetization, researchers warn the expertise may form conduct, beliefs, and selections in methods customers could not see
Eighteen months in the past, it was believable that synthetic intelligence may take a completely different path than social media. Again then, AI’s growth hadn’t consolidated underneath a small variety of huge tech companies. Nor had it capitalized on client consideration, surveilling customers and delivering advertisements.
Sadly, the AI trade is now taking a web page from the social media playbook and has set its sights on monetizing client consideration. When OpenAI launched its ChatGPT Search function in late 2024 and its browser, ChatGPT Atlas, in October 2025, it kicked off a race to seize on-line behavioral information to energy promoting.
It’s a part of a yearslong turnabout by OpenAI, whose CEO Sam Altman as soon as referred to as the mixture of advertisements and AI “unsettling” and now guarantees that advertisements will be deployed in AI apps whereas preserving belief. The rampant hypothesis amongst OpenAI customers who imagine they see paid placements in ChatGPT responses suggests they don’t seem to be satisfied.
In 2024, AI search firm Perplexity began experimenting with advertisements in its choices. A couple of months after that, Microsoft launched advertisements to its Copilot AI. Google’s AI Mode for search now more and more options advertisements, as does Amazon’s Rufus chatbot.
As a safety knowledgeable and information scientist, we see these examples as harbingers of a future the place AI firms revenue from manipulating their customers’ conduct for the good thing about their advertisers and traders. It’s additionally a reminder that point to steer the route of AI growth away from personal exploitation and towards public profit is shortly operating out.
The performance of ChatGPT Search and its Atlas browser is just not actually new. Meta, business AI competitor Perplexity and even ChatGPT itself have had related AI search options for years, and each Google and Microsoft beat OpenAI to the punch by integrating AI with their browsers. However OpenAI’s enterprise positioning alerts a shift.
We imagine the ChatGPT Search and Atlas bulletins are worrisome as a result of there may be actually just one method to generate income on search: the promoting mannequin pioneered ruthlessly by Google.
Promoting mannequin
Dominated a monopolist in US federal courtroom, Google has earned greater than US$1.6 trillion in promoting income since 2001. You might consider Google as an internet search firm, or a streaming video firm (YouTube), or an electronic mail firm (Gmail), or a cell phone firm (Android, Pixel), or perhaps even an AI firm (Gemini). However these merchandise are ancillary to Google’s backside line. The promoting section sometimes accounts for 80% to 90% of its whole income. The whole lot else is there to gather customers’ information and direct customers’ consideration to its promoting income stream.
After 20 years on this monopoly place, Google’s search product is way more tuned to the corporate’s wants than these of its customers. When Google Search first arrived many years in the past, it was revelatory in its capacity to immediately discover helpful data throughout the still-nascent internet. In 2025, its search consequence pages are dominated by low-quality and sometimes AI-generated content material, spam websites that exist solely to drive site visitors to Amazon gross sales — a tactic referred to as affiliate internet marketing — and paid advert placements, which at instances are indistinguishable from natural outcomes.
Loads of advertisers and observers appear to assume AI-powered promoting is the way forward for the advert enterprise.
Extremely persuasive
Paid promoting in AI search, and AI fashions typically, may look very completely different from conventional internet search. It has the potential to affect your considering, spending patterns and even private beliefs in way more delicate methods. As a result of AI can interact in energetic dialogue, addressing your particular questions, considerations and concepts somewhat than simply filtering static content material, its potential for affect is far larger. It’s just like the distinction between studying a textbook and having a dialog with its writer.
Think about you’re conversing along with your AI agent about an upcoming trip. Did it suggest a selected airline or lodge chain as a result of they are surely finest for you, or does the corporate get a kickback for each point out? In case you ask a couple of political concern, does the mannequin bias its reply primarily based on which political celebration has paid the corporate a price, or primarily based on the bias of the mannequin’s company house owners?
There may be mounting proof that AI fashions are at the least as efficient as folks at persuading customers to do issues. A December 2023 meta-analysis of 121 randomized trials reported that AI fashions are pretty much as good as people at shifting folks’s perceptions, attitudes and behaviors. A more moderen meta-analysis of eight research equally concluded there was “no vital total distinction in persuasive efficiency between (massive language fashions) and people.”
This affect could go nicely past shaping what merchandise you purchase or who you vote for. As with the sector of search engine marketing, the motivation for people to carry out for AI fashions may form the way in which folks write and talk with one another. How we categorical ourselves on-line is prone to be more and more directed to win the eye of AIs and earn placement within the responses they return to customers.
A distinct approach ahead
A lot of that is discouraging, however there may be a lot that may be completed to vary it.
First, it’s necessary to acknowledge that at this time’s AI is basically untrustworthy, for a similar causes that search engines like google and yahoo and social media platforms are.
The issue is just not the expertise itself; quick methods to seek out data and talk with family and friends will be great capabilities. The issue is the priorities of the companies who personal these platforms and for whose profit they’re operated. Acknowledge that you just don’t have management over what information is fed to the AI, who it’s shared with and the way it’s used. It’s necessary to maintain that in thoughts if you join gadgets and companies to AI platforms, ask them questions, or think about shopping for or doing the issues they recommend.
There may be additionally so much that folks can demand of governments to restrain dangerous company makes use of of AI. Within the US, Congress may enshrine shoppers’ rights to manage their very own private information, because the EU already has. It may additionally create an information safety enforcement company, as primarily each different developed nation has.
Governments worldwide may spend money on Public AI — fashions constructed by public companies provided universally for public profit and transparently underneath public oversight. They may additionally limit how companies can collude to take advantage of folks utilizing AI, for instance by barring commercials for harmful merchandise comparable to cigarettes and requiring disclosure of paid endorsements.
Each expertise firm seeks to distinguish itself from opponents, significantly in an period when yesterday’s groundbreaking AI shortly turns into a commodity that can run on any child’s telephone. One differentiator is in constructing a reliable service. It stays to be seen whether or not firms comparable to OpenAI and Anthropic can maintain worthwhile companies on the again of subscription AI companies just like the premium editions of ChatGPT, Plus and Professional, and Claude Professional. If they’ll proceed convincing shoppers and companies to pay for these premium companies, they might want to construct belief.
That can require making actual commitments to shoppers on transparency, privateness, reliability and safety which can be adopted by way of persistently and verifiably.
And whereas nobody is aware of what the long run enterprise fashions for AI shall be, we will be sure that customers don’t need to be exploited by AI, secretly or in any other case. – Rappler.com
This text initially appeared on The Dialog.
Bruce Schneier, Adjunct Lecturer in Public Coverage, Harvard Kennedy College
Nathan Sanders, Affiliate, Berkman Klein Middle for Web & Society, Harvard College
