This article is part of our Vogue Business membership package. To enjoy unlimited access to our weekly Technology Edit, which contains Member-only reporting and analysis and our NFT Tracker, sign up for membership here.
Many people now associate artificial intelligence and social media with fictional personalities or highly improbable scenes — such as the Pope wearing Balenciaga or Renaissance paintings coming to life — but that’s not the full picture. New uses of AI are starting to quietly influence decisions behind the scenes on social apps.
Thanks in part to the AI revival, startups are creating new tools that use the tech to increase likes, revenue and accuracy on organic, traditional content. Now, AI can estimate how a post will perform on social media, or compare best choices among various potential options. It can mimic a human talent agent to negotiate the terms of sponsored content between a brand and an influencer, and it can help legitimise luxury resellers by authenticating products. The end result is still the same familiar, human-created content, but the means are heavily influenced by new uses of AI.
This avoids the risk of the ‘uncanny valley’ effect (which appears eerily inhuman) that often plagues AI-altered human imagery, designs and text. It also sidesteps the intense scrutiny or squeamishness that AI-generated content invites.
“If we do our job right, it will go unnoticed,” says Gary Meehan, a third of the founding team behind Retrograde, a startup that uses AI to help social media influencers negotiate and monetise brand deals. (Meehan is co-founder and CTO.)
Many of these innovations are driven by a larger, louder push from social media giants to bring AI to their platforms. Last September, Meta introduced a number of virtual personalities based off of the likenesses and voices of those including Kendall Jenner, Paris Hilton and Snoop Dogg. Advertisers can now dramatically alter images in ads. Earlier this month, CEO Mark Zuckerberg said that he envisions advertisers being able to eventually use AI to entirely generate content for advertisers, based on business objective and budget, and personalise what people see. Snapchat has introduced a customisable ‘My AI’ chatbot that offers recommendations, answers questions and more. Back in June, TikTok released ‘Symphony Digital Avatar’, enabling businesses to animate the likenesses of paid actors in ads, including speaking words from a script.
But many of these tools have been met with mixed results. This summer, Meta phased out the celebrity AI programme, instead promoting its AI Studio that enables those with business accounts (such as creators and brands) to build their own AI chatbots. Snapchat’s tool fell under scrutiny when it experienced a glitch that resulted in the bot posting content on its own; the model has also been labelled potentially risky for teens. Similarly, one version of TikTok’s tool erroneously allowed users to create videos that say anything, seemingly without guardrails.
Public-facing usage of AI-generated content also invites the need to navigate disclosures. Consumers are sensitive to feeling misled; a partnership between bag brand Baggu and Collina Strada drew criticism when some commentators said they wished they had known up front that the brand used generative art to create some of its prints. Meta, YouTube and TikTok all now require accounts to disclose whether content has been AI-generated. (This is not so easy to pull off, with Meta having accidentally mislabelled non-AI content in the past.) Governmental bodies are getting involved: the EU AI Act requires AI chatbots or other systems that interact with consumers — and deepfakes — to be labelled as ‘AI generated’.
While flashier uses are driving interest, utilitarian tools offer more practical uses, as brands and creators establish a more discerning lens. “There is a bit more due diligence than before. A year ago, anything ‘AI’ was like, ‘Yes, we want it now,’” says Parham Aarabi, CEO of AI startup Pre. Now, he says, “it’s like, well, prove that it actually works. Once you get that bar, then it’s a very easy sell”.
Negotiating brand deals
To secure brand deals, many mid-sized creators typically fork over 20 per cent of their earnings to work with an agency, or self-manage — which translates into a lot of time spent on emails, says influencer-turned-entrepreneur Grace Beverley, who originally gained a following of more than a million on Instagram through lifestyle content focused on fitness as @GraceFitUK. “It essentially means that they end up having a glorified desk job… you’d be so surprised at how bad the efficiency is within the whole industry.”
Now the founder of activewear brand Tala and fitness app Shreddy, Beverley is hoping to make it less laborious for creators like her to monetise their content. This August, she, along with co-founders Meehan and Jake Browne (who co-founded two other startups aimed at content creators), launched Retrograde, positioned as an “AI talent agent” that communicates with the creator via text message and potential clients via email. It is designed to communicate like a person and can read contracts and digest creative briefs, communicating key takeaways with the creator (similar to an agent) through text message. While more complex brand deals, like those with many facets and negotiations on usage of intellectual property, might require more hand-holding, that’s not the majority of deals that happen in the industry, Beverley says.
“One of the greatest applications that we could do with AI is helping to optimise those processes that are admin heavy and don’t need to be,” says Beverley, adding that this type of tech (which was built on tech from AI companies OpenAI and Anthropic) is well suited to this industry because there are relatively few variables in each of these influencer contracts.
Creators are onboarded and given a Retrograde email, and when a brand reaches out with a proposal — say, $500 for an Instagram post — Retrograde will text them an update, and the creator can respond and say that they will only do it for $1,500. Once the negotiation has reached a resolution — Beverley says even simple deals can take more than 10 emails — it generates a contract. Retrograde makes money by taking 10 per cent of the fee and is paid only when the influencer is paid.
Optimising performance
Startup Pre developed an AI simulator that predicts how customers will interact with e-commerce sites, looking at factors such as user intent and likelihoods of interaction, adding to a cart or converting into a sale, based on screenshots. This summer, it added to its arsenal with a new tool, ‘Pre Social’, which uses a similar process to analyse social media posts. It can estimate the number of likes on a visual post with up to 80 per cent accuracy. Using this tool, brands and creators can submit potential posts to Pre to make a decision on what is likely to get the most likes.
“It’s not a hundred [per cent], because it is a very difficult problem,” Aarabi says. “There are a lot of human factors that we can’t model, but I think it’s a really good starting point that can give you a lot of information.”
The tool was trained on 2,400 images — half of them engaging images, half of them non-engaging — and then a separate series of 1,350 images from accounts with more than 100,000 followers, noting the number of likes on each. The research, which was presented in a white paper at this summer’s International Conference on Social Media and Society, can now be used to both make recommendations on which potential posts are likely to get the most engagement and estimate how many likes and comments it will get based on the number of followers an account has. Pre charges through a software-as-a-service model, similar to its previous tool.
Authenticating luxury resale
In September 2023, TikTok launched its Shop function to enable users to buy and sell products directly in the app, via a marketplace or live shopping. This has helped fuel ‘dupe’ shopping, as well as the market for luxury counterfeits. Now, as TikTok courts luxury brands, it’s turning to AI authentication.
Last autumn, TikTok partnered with tech startup Entrupy to be the official authenticator of pre-owned luxury goods in the US. It started with luxury bags, with anyone selling them on the app required to authenticate the products using a special Entrupy-supplied light box-like device. This month, it has expanded the programme to sneakers, which can be authenticated more simply using the Entrupy app — spanning styles from those including New Balance, Alexander McQueen, Balenciaga, Christian Louboutin, Dior, Gucci, Louis Vuitton and Off-White.
A seller can ‘scan’ a product to collect a set of microscopic images. Then, Entrupy’s machine learning compares this to more than 20 million images of similar products dating back to 100-year-old pieces, including both authentic and counterfeit products, to understand the differences. Each scan returns a result as authentic or unverified, with a 99 per cent accuracy rate, according to the company.
CEO and co-founder Vidyuth Srinivasan says that while machine learning (a subset of AI) has always been a part of Entrupy’s offering, it hasn’t focused on AI as its main draw. Instead, it’s focused on utility — perhaps differentiating from the flashier tools that inherently emphasise new, visible uses of generative AI.
“AI is fundamentally a twin-use tool. It could be used to create something positive,” he points to productivity, creativity and new solutions, “or used to disrupt or destroy something,” such as disinformation, division or derision, he says. “It can be a great accelerator for a lot of positive things for humanity, but it all depends on who is wielding it and their intent. Hopefully we can all agree on guardrails that promote the former while preventing the latter.”
Sign up to receive the Vogue Business newsletter for the latest luxury news and insights, plus exclusive membership discounts.
Comments, questions or feedback? Email us at feedback@voguebusiness.com.
