The Consumer Reports Affirm framework, published on 18 November 2026, is a draft set of written instructions that tells an AI shopping assistant to compare costs and risks before it recommends financing, paying in full, another provider, waiting or not buying. Fintechbits checked the framework’s GitHub repository, which says the 71.8% and 48.6% pass rates come from 30 scenarios and 720 simulated conversations across four models, and carries a “Not ready for use” banner. Without the instructions, the assistants passed 48.6% of checks on average, and the two organisations report 71.8% with them.
What the Consumer Reports Affirm framework asks an AI to do
The framework is a text file, not software. According to its repository it works in five steps: verify and contextualise the terms, normalise costs, stress-test the options, gate and rank them, then recommend without acting until the shopper confirms. It also asks the assistant to weigh long-term effects such as a missed payment or an expiring promotional rate.
Consumer Reports brought its Consumer Finance AI Standard, which sets nine principles including a duty of loyalty and answers that are not skewed by a company’s commercial interests. Affirm, according to the release, supplied payment expertise, technical development and testing. Dan Leininger of Consumer Reports, who works in its experimental engineering group, says “we need to ensure they put consumers first” as AI agents take on more purchasing decisions.
The case study numbers and where they stop
The framework lifted the average pass rate from 48.6% to 71.8%, a gain of 23.2 percentage points. The other side of the result matters as much: even with the instructions, 28.2% of checks failed. The repository states that the results measure AI responses, not consumer outcomes, and do not prove the framework is ready for consumer use.
The worked example is a simulated $1,200 purchase. Both responses favoured paying in full upfront, but the framework version priced the alternatives: paying only a $40 monthly card minimum could stretch repayment to 41 to 47 months and add $406 to $651 in interest, an instalment plan cost a flat $60, or 5% of the price, and paying by debit added no borrowing cost. The repository adds that testing used simulated conversations with controlled tools, and that the mode where shoppers paste in their own terms is untested.
Affirm’s agentic payment push in context
The framework arrives with a commercial backdrop. On 6 March 2026, The Paypers reported that Affirm would support Stripe’s Shared Payment Tokens so agents could start purchases without exposing sensitive credentials, and that Klarna had announced similar support through Stripe. On 14 May 2026, The Paypers reported Affirm plans inside Google Search AI Mode and the Gemini app through Google Pay.
On 5 November 2026, PYMNTS reported Affirm chief executive Max Levchin telling analysts “I expect 0% volume to go up a lot in agentic.” Interest-free loans were 28% of Affirm’s gross merchandise volume in the September quarter. Fintechbits has followed both sides of that bet: Affirm’s Sur la Table plans from 0% APR on 31 October 2026, and the checkout-security work in KERNEL’s agent payments on 16 November 2026. Levchin’s release quote is “At Affirm, we only succeed when our consumers do.”
Fintechbits analysis
Why a lender-built test of consumer-first AI needs outside replication
The release matters as a template, not as consumer protection. Written loyalty rules for AI agents are overdue, and a pass-rate benchmark gives rivals something to run against. But the evidence in hand is thin and comes with an interest: Affirm developed the original framework and tested it against a standard that Consumer Reports wrote, while Affirm’s chief executive expects agents to favour the product Affirm sells.
For shoppers, nothing changes until an assistant adopts the text, and the repository says not to use it for real payment decisions. For rivals such as Klarna, the useful move is to publish their own results on the same scenarios. For regulators, Consumer Reports’ duty-of-loyalty language supplies vocabulary regulators could borrow if they write rules for AI agents.
Frequently asked questions
What is the Consumer Reports Affirm framework?
It is a draft markdown file that steers an AI assistant through checking terms, comparing costs, weighing risks and justifying its payment recommendation. Consumer Reports and Affirm published it with a joint case study on 18 November 2026.
Can shoppers use the framework to choose how to pay?
No. The repository says it is not ready for use, is not a standalone AI tool and should not drive real payment decisions or a consumer-facing product. It covers U.S. purchases only.
Does Affirm benefit from the framework?
Possibly. Affirm supplied payment expertise and testing, and Levchin expects interest-free volume to rise in agentic shopping. The framework also asks assistants to keep rankings free of commercial influence, so independent testing would show whether the two goals hold together.
What to Watch Next
Three things will decide whether the draft travels. First, whether other lenders and AI developers publish results on the same 30 scenarios, which would test the 71.8% outside the authors’ own setup. Second, whether any assistant maker adopts the instruction file, since a pass rate on simulated conversations is not a consumer outcome. Third, whether Consumer Reports tests Affirm’s own agent-facing products against its Consumer Finance AI Standard.



