AI search is creating an attribution problem for marketers just as ChatGPT and other artificial intelligence assistants become more influential in deciding which brands consumers consider, buy from or ignore. With ChatGPT already on a $US1bn run rate for ads after just 200 days, the attribution issue is becoming more pressing. Now Australian MMM vendor, Mutinex is having a stab at the problem, with news out this morning that it has added AI search capabilities to its GrowthOS platform. Unlike conventional search, an AI assistant can recommend a product, shape a shortlist or influence a purchase without generating the click that digital marketing measurement was built to follow. That leaves marketers trying to determine whether appearing in an AI answer has commercial value, what marketing helped put the brand there and whether spending more money to improve that visibility will produce a return. "More and more consumers are asking AI on what the best options are for their many needs when it comes to their internet or mobile, and right now we have limited ways of understanding whether Belong is showing up in those conversations or the impact it is having," said head of marketing at internet and mobile provider Belong, Josh Bailey. "That's exactly the problem we need solved as AI becomes a real channel for customer acquisition." The problem becomes harder because the underlying measurement target keeps changing. “The moment in time we’re in right now is you have to sort of acknowledge that this is a moving target,” Tracksuit chief marketing officer, Tom Mansfield said. He is coming at the issue from two perspectives: The implications for Tracksuit's own platform, but also as a CMO working out how to get the best value from search as a channel. "The nature of the responses from an LLM's perspective is super dynamic, and I think that's part of the hard challenge where attribution comes into play. It's one thing if you have an algorithm where you know you can estimate, based on a certain number of inputs, what's going to come out of the search query," Mansfield said. "But because AI and LLMs are constantly learning, and the answers are constantly being basically personalised for your preference style, it's much more complicated to have a consistent view of how you, as a brand, show up. "How do you basically build a consistent view when the outputs from those models are dynamic in real time, and even the models are changing in real time?" Mansfield believed marketers should experiment with AI visibility while continuing to measure the wider brand environment influencing consumer decisions. People enter AI interactions with existing preferences and can reprompt an assistant about a preferred brand if it is absent, while brands appearing together in a response must still compete for the consumer’s preference. Agencies are also just beginning to get a handle on the issue.  Last month, MultimediaX managing director, Najib Raad told Mi3, "People using ChatGPT are quite early in the funnel stage, so they’re still doing research. They’re still considering, and then your brand has just become aware to that client ... A couple of weeks down the track … maybe they’ll then jump onto Google and search for it. Most brands, because of the attribution gap, will completely miss that business." Raad said his agency tries to follow the money all the way through clients’ CRM systems, from cost per lead to qualified lead, sale and ultimately return on investment. Less sophisticated advertisers may have to settle for knowing what it cost to generate a lead and hope the trail does not go cold from there. Even for those with the plumbing connected, cookie loss and attribution gaps mean the journey from click to cash remains considerably easier to describe than to prove. Meanwhile, Sami Akkawi, co-founder, Petra Labs whose firm revealed the collapse in Reddit and YouTube citations on ChatGPT recently, told Mi3 at the time that his firm tries to connect changes in AI visibility with LLM referral traffic, direct traffic, branded organic traffic and the costs associated with the sources driving improved visibility.  It is possible to follow the advertising money and its impact, he said, "but it's hard for now ... and it requires custom work for each client. It's still quite a difficult problem in the sense that it's still very bespoke, and takes us time. We haven't figured out a way yet to do this at scale. "It still takes a lot of custom work for each client." A problem worth solving The measurement gap has spawned a growing category of tools tracking brand visibility inside AI answers. Mutinex is pushing further into attribution with the launch of AI Visibility and AI Signals within its GrowthOS marketing mix modelling platform. According to Mutinex co-founder and global CEO, Henry Innis, "There are three questions a marketer has to answer about AI search, and they only work in this order: 'Am I visible, what is that visibility worth, and what does it cost to move it?' The industry has answered the first one and declared victory. But a visibility score is not an outcome. It is an input with no price on it, and recommending spending against an input nobody has priced is precisely how the last 15 years of search marketing went wrong. Our goal isn't to ship a better scraper.” Mutinex’s system runs a fixed panel of category questions each week across ChatGPT, Claude, Gemini and Perplexity, recording whether brands appear, their prominence and sentiment. Those readings are fed into Signals alongside marketing mix data covering paid, earned and owned activity. The system identifies which marketing activities move most closely with changes in AI visibility, giving marketers what Mutinex describes as a directional view of investment strategy. It does not establish that an individual marketing activity caused a particular AI response. For marketers, the emerging challenge is increasingly about connecting three things that AI search has pulled apart: What consumers are being told, what caused brands to appear in those answers and what those appearances ultimately contribute to the business. "AI is increasingly becoming part of how travellers research and plan their trips, making it important for brands to understand how they appear in AI-generated recommendations," said Webject CMO, Oonagh Flanagan. Like Belong, it's had early access to the tools. "We’ve recently started exploring AI Visibility through GrowthOS and are keen to explore the additional insights it can provide as customer search and discovery behaviours continue to evolve."