When the AI recommends a brokerage, nobody gets the commission
Someone opens ChatGPT Finance, asks which brokerage they should use for long-term investing, and gets a specific answer. They follow the recommendation. They open an account, deposit money, and start investing.
The brokerage got a customer. The AI gave a specific, acted-upon recommendation. And nobody got paid for the referral.
That's not a hypothetical. ChatGPT Finance is live, it's making specific product recommendations, and financial affiliate programs are the highest-paying vertical in the entire affiliate industry. Credit card approvals routinely pay $50 to $200 per conversion. Brokerage account openings can run significantly higher. NerdWallet built a business worth over a billion dollars routing financial product signups to publishers who could drive them.
That traffic is now, incrementally, shifting to AI advisors. The commission infrastructure doesn't exist for it yet.
The highest-stakes attribution gap
The affiliate industry has been quietly processing the agent attribution problem for about a year. The general shape of it is well-established by now: cookie-based systems don't work when there's no browser, no click, no session. AI agents make API calls; they don't embed affiliate links. The infrastructure needs to be rebuilt from first principles.
Financial services makes that problem much more expensive to ignore than general e-commerce.
A generic product recommendation on a shopping assistant might drive a $40 purchase with a 3% affiliate rate. A financial product recommendation drives an account opening worth a four-figure CPA. Publishers in financial services earn multiples of what most other verticals pay, specifically because the conversion is high-value and durable. The customer who opens a brokerage account on a recommendation is worth something to that brokerage for years.
When the recommender is a human writing for Bankrate or The Points Guy or NerdWallet, the attribution chain is messy but it exists. They embed a tracked link, the user clicks it, a cookie fires, and 30 days later if someone opened an account, someone got paid.
When the recommender is ChatGPT Finance, none of that chain exists.
The publishers who built this category are getting bypassed
Financial affiliate publishing is one of the few categories where the business model was genuinely clean: create trustworthy comparisons, help users make better decisions, earn a commission when they do. NerdWallet, Bankrate, Credit Karma (before the Intuit acquisition) — these companies worked because the value exchange was real. Useful content, accurate comparisons, honest CTAs. Publishers who did it well made real money. Merchants who paid them got real customers.
The AI advisor keeps the value exchange intact. It just compresses it into a single session. Instead of a user reading a 2,000-word comparison article and clicking through a tracked affiliate link, they ask a question, get an answer, and act on it in the same session.
The recommendation quality might be equivalent. The conversion might be real. The merchant still acquired a customer. But the referral source is invisible to every existing attribution system.
Financial services merchants are already spending money on affiliate programs. They have budgets, tracking infrastructure, and commission structures already set up to pay for customer acquisition. What they don't have is any mechanism to track AI-sourced referrals, because the current stack assumes the referral comes through a publisher with a tracked link.
Why finance gets here first
Financial services has something that most other verticals don't: every transaction is already logged server-side by law.
When a customer opens a brokerage account, the brokerage records exactly what happened: when it happened, what product they chose, and what their entry point was. The regulatory paper trail that compliance requires means the raw data for attribution already exists. The problem isn't finding the event. It's connecting the event to the recommendation that preceded it.
That's a solvable problem. AI advisors like ChatGPT Finance operate on identifiable infrastructure. Recommendations can be tagged at the moment they're generated. When a user acts on a recommendation and opens an account, the token that was assigned to that recommendation can travel with the conversion. The brokerage already has the conversion event. The attribution layer just needs to close the loop.
That determinism is what gives agent attribution an edge over cookies. It's sharpest in financial services. Cookie attribution in finance was always probabilistic and contentious: did the user convert because of your article, or because they were going to open an account anyway? With AI advisors, the recommendation is explicit, logged, and time-stamped. Attribution becomes a matching problem, not a statistical inference.
The position that's opened up
NerdWallet's model was: be the trusted intermediary between consumers and financial products, earn commissions by routing conversions. That model is still viable. AI advisors just change where it lives.
The next version of that model is the infrastructure layer between AI advisors and financial affiliate programs. It tracks AI-sourced recommendations through to conversions, and it lets merchants pay for the customers they're already getting. The AI advisor is the publisher. The commission structure still works. The plumbing just needs to be rebuilt for how recommendations actually travel now.
ChatGPT Finance made its first product recommendations last week. Financial services merchants don't have any mechanism to pay for the customers those recommendations drive. That gap has a number attached to it, and the number is large.
The infrastructure that closes it is the same infrastructure the broader affiliate industry needs for agent commerce generally. Financial services is just where the cost of not having it is most visible.
Syndicate Links is building the attribution infrastructure for agent-initiated commerce. If you're a financial merchant, a fintech, or a developer building agent-powered tools, the integration docs and MCP server spec are at syndicatelinks.co.