OpenAI looks to online advertising deal – AI-driven ads will be hard for consumers to spot

Making AI quicker, smarter and better is proving to be a very expensive business. Companies like OpenAI are investing billions of dollars^[1] in hardware, and the likes of Meta are offering top (human) talent huge salaries^[2] for their expertise.

So perhaps it’s no surprise that these businesses have started exploring new ways of making money as well as spending it.

OpenAI, for example, is exploring a partnership with Shopify^[3], one of the world’s largest e-commerce platforms, which helps businesses manage online selling.

The reported deal between the two companies would see OpenAI receive a cut of any Shopify sales that result from recommendations provided by ChatGPT, creating a new revenue stream for OpenAI and more online traffic for Shopify.

But this relationship could be risky for consumers if OpenAI became incentivised to push people towards products, rather than offering genuinely objective recommendations. It might even push recommendations when users of ChatGPT are not looking to buy anything at all.

This situation reminds me of the early days of online advertising when Google was under pressure^[4] from shareholders to increase revenues, following the dot-com bubble. Google was (and still is^[5]) the world’s leading search engine, in part because it had the best algorithm. But the obvious path to generating revenue – advertising – posed a big dilemma.

Loading search results with adverts would put off users and weaken Google’s position. The company’s solution was to develop targeted advertising^[6], matching ads to search queries to maintain relevance and quality.

Similarly, OpenAI will surely not just flood ChatGPT with links to products. If it did, the quality of its own product would decline, and users would quickly go elsewhere.

So, like Google, it needs to find a subtle way to influence people to shop.

Luckily for OpenAI, the sociable, text-based interface of a chatbot creates ample opportunities to use persuasive techniques to try to influence people’s behaviour.

One way of thinking about online persuasion is in terms of “metacognition”^[7], the ability to think about thinking, which is very important in the world of sales.

Research suggests that when a customer has high metacognition skills, they are more likely to be sceptical^[8] of a salesperson’s tactics, and harder to persuade. When a salesperson has high metacognition, they are good at getting into a customer’s head^[9] and making a sale.

One theory of metacognition^[10] argues that high levels are influenced by how much sellers and customers know about a product, how much they know about persuasion, and how much they know about each other.

In all three cases, AI may have an advantage.

On any given topic, ChatGPT will “know” more about it than an average person. A particularly knowledgeable person might not get caught out. But nobody is an expert on everything, while ChatGPT can at least pretend to be (like any good salesperson).

AI large language models (known as LLMs) are also up to speed on the latest research on rhetoric, marketing and psychology. They can even identify deceptive sales techniques^[11].

AI can also be tweaked to be persuasive. For instance, research has found that people are more likely to buy something when a salesperson or advert mirrors their personality^[12]. One study^[13] found that ChatGPT can accurately predict a person’s personality from relatively little information. Over time then, ChatGPT could be programmed to make predictions about us, and then start acting like us.

When it comes to knowledge about each other, most people probably know little about how AI language tools actually work. And if people are also unaware of the incentive AI companies may soon have to recommend products, these recommendations may be met with less scepticism, because an AI chatbot would seemingly have no motive to manipulate.

Meanwhile, like Google, companies such as OpenAI are gathering huge amounts of data^[15] about the people who use their software. Initially, this was to train future AI models. But these same data could be used to learn more about people, what makes them tick, and what makes them click “buy”.

Product recommendations from ChatGPT, Google or any other company are not inherently sinister. If data is used to suggest products people genuinely love, this can be helpful.

But being helpful is not the primary motivation here. Just as Google introduced ads because of financial pressure, deals like those between OpenAI and Shopify are a response to the economic pressures the AI industry is facing^[16].

It is great if these systems recommend products a person wants to buy. But what might matter most to AI, regardless of the product, is that they buy it.

References

1. ^{^} billions of dollars (www.nytimes.com)
2. ^{^} huge salaries (www.wired.com)
3. ^{^} exploring a partnership with Shopify (www.ft.com)
4. ^{^} Google was under pressure (www.nytimes.com)
5. ^{^} still is (gs.statcounter.com)
6. ^{^} develop targeted advertising (www.nytimes.com)
7. ^{^} “metacognition” (www.jstor.org)
8. ^{^} more likely to be sceptical (www.jstor.org)
9. ^{^} good at getting into a customer’s head (academic.oup.com)
10. ^{^} theory of metacognition (academic.oup.com)
11. ^{^} identify deceptive sales techniques (papers.ssrn.com)
12. ^{^} mirrors their personality (myscp.onlinelibrary.wiley.com)
13. ^{^} One study (journals.sagepub.com)
14. ^{^} Koshiro K/Shutterstock (www.shutterstock.com)
15. ^{^} gathering huge amounts of data (theconversation.com)
16. ^{^} AI industry is facing (futurism.com)