Agent, meet agent: the AI updates agencies need on their radar 

Shoppers get their own AI agents, ads start talking back and efficiency fails to pay off… what agencies need to know this issue 

A shopping cart on a vibrant purple and blue background. The cart is the main focus of the image, with the colors of the background providing a vivid contrast.

Agent, meet agent: the AI updates agencies need on their radar 

Shoppers get their own AI agents, ads start talking back and efficiency fails to pay off… what agencies need to know this issue 

A shopping cart on a vibrant purple and blue background. The cart is the main focus of the image, with the colors of the background providing a vivid contrast.

Shoppers get their own agents 

For the past two years, AI agents have mostly been something brands and agencies have talked about. This autumn, consumers started getting them. 

Meta launched Muse in the US and Canada. It’s a personal agent that can send emails, book travel, fill out forms, lower bills and make purchases. Three weeks later, OpenAI launched Dots, always-on agents in ChatGPT that each get their own cloud computer and browser and keep working after the user closes the tab. Anthropic followed with Claude Duet, a standalone agent app for professionals and small-business owners. Google is taking its new Gemini agent to businesses first and consumers later. 

How many people are using them? Meta says more than 10 million people in the US and Canada set up Muse in its first six weeks, and around a quarter use it daily. Shopping is its most common use, with the average purchase over $200. 

 Why is everyone launching at once? Marcos Angelides, Head of AI Operations at Publicis Media, thinks it comes down to convenience. Nobody wants a different agent for every retailer and brand. People want one that knows their preferences, budget and non-negotiables. “They’re not really competing to build an agent. They’re competing to become your agent,” Angelides writes. Whichever agent wins gets far more than a transaction… they get the context, data and habits of the person it works for. 

Angelides’ advice for brands is to stop thinking about “your agent” and start thinking about “their agent”. The question isn’t just whether Brand X needs an agent of their own, but whether a consumer agent would choose Brand X in the first place. 

A quick reality check… Muse and Dots aren’t available in the UK yet, though Meta is reportedly preparing a UK launch of Muse within weeks, with UK retailer partnerships to follow. But the direction is clear… the thing doing the shopping may increasingly not be a person.

Not everyone thinks brands should leave the talking to someone else’s agent, though…

The rise of the AI sales agent 

A seamless pattern of colorful speech bubbles with lips on a purple background. The speech bubbles are arranged in a repeating pattern, creating a visually pleasing and eye-catching design. The lips are painted in a variety of colors, including red, blue, yellow, and green, and the background is a deep, vibrant purple.

What happens when an advert doesn’t send you to a website, but starts a conversation instead? 

OpenAI is testing what it’s calling “Sponsored Agents”, a new advertising format that lets users move directly from an ad into a chat with an AI representative of the brand. Rather than clicking through to a landing page, users can ask questions, compare options and receive personalised recommendations before deciding whether to buy or learn more. 

The shift is subtle but significant, for decades marketing has focused on driving clicks but Sponsored Agents introduce a new model. 

From ad to conversation to conversion? 

For agencies, this may create a new discipline. Success could depend not only on creative quality and media performance, but on how well a brand’s AI agent answers questions, explains products and guides consumers towards a decision. 

Just as SEO defined the last generation of digital marketing, optimising for AI conversations may define the next with brands competing not just for attention, but for the quality of the conversations they can deliver. 

Sponsored Agents offer an early glimpse of a potential future in which AI assistants become a primary channel for discovery, consideration and purchase, potentially reshaping how brands engage with customers online. 

Editor’s note: Since this piece was written, OpenAI has expanded the idea. Eligible US online shops and service businesses can now build their own Business Agent in ChatGPT and use it as the destination for their ChatGPT ads. It isn’t yet available in the UK. More on OpenAI’s new Businesses in ChatGPT tools below. 

A seamless pattern of colorful speech bubbles with lips on a purple background. The speech bubbles are arranged in a repeating pattern, creating a visually pleasing and eye-catching design. The lips are painted in a variety of colors, including red, blue, yellow, and green, and the background is a deep, vibrant purple.

ChatGPT ads get their first reality check 

Since September, select US brands have been able to buy ChatGPT ads through Amazon DSP. But the first independent study of how they perform, flagged by Sean Betts (the Chief AI & Innovation Officer at OMG UK) in his newsletter, makes for awkward reading. A working paper from researchers Maximilian Kaiser and Christian Schulze, which tracked 1,898 online shops, found a paid ChatGPT visit generated $0.89 in revenue in US traffic, against $2.45 for a paid Google search visit… 64% less. Of 87 advertisers buying both, 84 earned less per ChatGPT ad visit. The paper is still under review, but for agencies advising clients on test budgets, it’s a useful reality check. 

The fight over the front door 

If shoppers send agents to buy things, someone has to decide how those agents get in. That fight is now under way. 

On 6 October, Sierra and Meta published the Personal Agent Protocol, an open standard for how personal agents authenticate with businesses, backed by Walmart, Shopify and Stripe. OpenAI, Google and Visa all have rival standards. 

The practical consequence is already visible. Shopify has embraced shopping through Muse, while Amazon has blocked it. A brand selling through both may find an agent can recommend and buy from one storefront but not complete the same journey on the other. Walmart has gone the other way. As AI search expert Andy Francos notes in his Franc Talking newsletter, it has just rolled out an Agent Gateway to let approved shopping agents in, while Amazon keeps blocking them. 

Banks are drawing their own lines. JPMorgan, Citi, Bank of America and Schwab have all said outside agents can’t move customer money, and Schwab says it blocks Muse attempts dozens of times a week. In the UK, HSBC, Nationwide, NatWest, Barclays and Lloyds have reportedly moved to block AI agents from customer accounts too. 

For agencies, that’s a new kind of question to ask clients: not just “can people find us?” but “can an agent buy from us?” Accurate product data, live prices, stock levels and a working route to purchase become media questions, not just ecommerce ones. 

Efficiency isn’t where the payoff is 

AI use is now routine, but the results are lagging. ISBA’s latest Gen AI Survey of 200 advertisers found 65% use it regularly at work, yet only 14% report a significant or transformational impact on their advertising. 

The more interesting finding is about motive. Efficiency is the main focus for 76% of organisations, up from 65% the year before. But only 11% of efficiency-led organisations report significant impact, compared with 25% of those putting effectiveness first. ISBA Director General Simon Michaelides is careful to say that doesn’t prove cause and effect. Creative quality has now overtaken governance as advertisers’ biggest barrier. 

Havas has found the same problem from the consumer’s side. Across more than 21,000 AI-powered brand experiences, only 9% landed in the top band of its Valuable AI Experience Score. Havas calls it the AI Efficiency Trap: when AI saves the brand time without giving the customer anything useful, it creates drag. The experiences that got it right were linked to 4.4 times higher brand engagement. 

If a brand’s first conversation with a customer is going to be with an agent, that’s a gap worth closing. 

There’s a measurement row brewing in search too. OpenAI says ChatGPT now handles search-like queries equal to about a fifth of Google’s volume, yet it sends websites only a fraction of Google’s traffic. Francos argues that judging AI search on referral traffic is like judging a goalkeeper on goals scored, because much of the deciding happens inside the chat. 

OpenAI is now offering some numbers of its own. On 10 November it launched Businesses in ChatGPT in beta. Through a new Business Hub, any eligible business in a market where ChatGPT is available, including the UK, can claim its profile, update its details and product feeds, and see how often it’s mentioned, how many website clicks it gets and the broad themes of conversations where its site is cited. Individual conversations stay private. In the US, eligible ecommerce and online service businesses can also launch a Business Agent to answer customer questions, and even sponsor it as the destination for ChatGPT Ads. In other words, the same brand agent can now appear whether a business has paid for an ad or been recommended for free, which blurs the line between paid and organic more than search ever did. 

Agencies: everyone’s faster, few are better off 

Agencies are stuck in the same efficiency loop as advertisers, according to Spark AI’s latest Spark Report, Where’s the Value?, based on more than 350 responses from agency leaders and staff. 

AI use is now close to universal. 90% of agency staff use AI, and 91% of those users say it saves them time. The savings are also getting bigger. 13% of users now save more than five hours a week, up from 5% in 2025. Only 9% save no time at all, down from 55% the year before. 

A bar chart depicting the percentage of people who use AI in the United States. The chart is divided into four sections, each representing a different type of AI. The first section is labeled "AI" and the second section is titled "New Revenue". The third section is labelled "Ai". The fourth section is colored in shades of blue, green, and yellow. The fifth section is highlighted in red.

The problem is what happens to that time. When staff were asked where it goes, the most common answer was more output (48%). Almost as many (45%) couldn’t say. Fewer said it led to better-quality work (41%) or less pressure (25%). Just 9% said it went into new kinds of work. Spark warns that without limits, faster just means doing more, and the productivity gain becomes a burnout risk. 

How people use AI is also uneven. Spark’s spring report found 45% of staff used it daily, 40% dipped in and out, and 15% barely engaged. 

Agencies aren’t always helping. More than a third (37%) still leave staff to teach themselves AI, and almost a third (30%) have no named person in charge of adoption. 

Most agencies are also using AI to speed up the way they already work, rather than changing it. 22% use it only for productivity tasks. 53% use it to improve client work within their existing processes. Only 16% have set up shared AI assistants for their teams, and just 9% have redesigned a workflow from start to finish. 

Pricing decides who keeps the value. 35% of agencies charge by output, 31% purely by time and 28% by value. Only 6% have a new AI-enabled revenue model. More than half (55%) have no formal position on AI with their clients. Spark’s verdict is that agencies “have got faster at the old way of working”, and that hourly pricing “turns every efficiency into a discount”.

The bigger picture 

In September, the story was ads moving inside AI conversations. This issue, the conversation itself is changing hands. We’re in the early days of brands getting agents to talk for them, consumers getting agents to shop for them, and the platforms fighting over the door between the two. 

Meanwhile, the evidence suggests that using AI to do the same things faster isn’t where the value lies. Very little of the agent world has reached UK consumers yet, and that’s the point. The agencies that work out now how agents find, judge and buy from brands, and what AI is actually for, will most likely be the ones clients call when it does. 

A shopping cart on a vibrant purple and blue background. The cart is the main focus of the image, with the colors of the background providing a vivid contrast.

Published by the IPA, the IPA AI Magazine helps member agencies make sense of the fast-changing AI landscape, with practical insights, expert perspectives and real-world applications of generative AI.
Managing Editor: Conor Nichols
Design: Jenna Betts

© 2026 The Institute of Practitioners in Advertising. All rights reserved. No part of this site may be reproduced without our permission.