What an AI commerce service for agencies should include
In 2026, an AI commerce service for agencies has to answer one plain client question: what do AI assistants see on our site, and what happened after they arrived? Arbling's agencies page, read on 13 October 2026, offers to bring the client audit, improvements and outcome measurement into one workflow, and that workflow is a sensible spine for a recurring offer. The same page describes one-time AI-visibility and SEO audits, with an optional recurring monthly monitoring cycle, which gives an agency a clean way to start small and then renew. A recurring AI commerce service for agencies earns its retainer when every month ends with the same report layout: what was checked, what was changed, and what was measured. Keeping those three parts fixed removes much of the manual rebuilding that eats margin on client reporting, and it makes the work easy for a client to understand and to renew without a long explanation each cycle.
The audit is the natural entry point, and the guide on turning AI search checks into a client offer covers how to package it. The retainer then starts when the audit findings become a queue of improvements that you ship and re-measure.
On 13 October 2026, the agencies page also said Arbling recognizes buyers from ChatGPT, Claude, Perplexity and 25 more AI assistants. It describes agent pages too: AI agents get a machine-readable page on the business's own domain with the same facts shoppers see. For an agency working on agentic commerce, a page like that is something a client can open and check. If you need background on how shopping assistants choose, read how ChatGPT Shopping decides which stores to recommend.
Reporting AI referrals, agent fetches and attributed orders separately
Google Analytics help, read on 11 October 2026, says traffic acquisition is attributed per session, and visits that arrive with no referrer are reported as direct or unassigned traffic. For an AI commerce service for agencies in 2026, that has a direct reporting consequence: a report built on referrer data alone cannot assume that every assistant-driven visit arrives labelled. OpenAI's crawler documentation, read the same day, lists separate user agents for search indexing, model training and fetching a page live for a user. So a fetch by an AI agent is a different event from a referral that brings a visitor, and neither is an order. A clear report keeps four lines apart: referrals, agent fetches, attributed orders and incremental lift. Each line answers a different client question, and merging them is how a retainer report ends up promising more than the data can support.
Google's help page on traffic acquisition explains the direct and unassigned buckets. Treat a user agent string as a declared label. OpenAI's documentation separates what its agents do, but a declared name is not a verified identity, and a browser-like visit is not proof that a person was there. Say that plainly in the report notes.
An attributed order is an order that the attribution rules tied to an AI source. It is not proof of extra revenue. Incremental lift asks what would have happened without the AI channel, and counting orders cannot answer that. The guide on what counters can't tell you draws the same line.
Platforms add their own rules. Shopify Help on Agentic Storefronts sets channel eligibility, referral and attribution rules for selling through AI channels, and not every product is exposed. For each client, record which products are exposed before you read any order total.
Pricing the retainer so manual reporting stops eating margin
In 2026, the margin of an AI commerce service for agencies depends less on the headline fee than on how much reporting is rebuilt by hand each month. Arbling's pricing page, read on 13 October 2026, lists AI shopper tracking and attribution on every plan, and multi-store reporting and attribution for up to 10 stores on Scale. The agencies page, read the same day, says the Enterprise plan includes agencies managing multiple clients. Those are the features to line up against your own client count when you plan a retainer, because each client you add either reuses the same reporting layout or forces another round of manual assembly. A retainer that reuses one layout keeps its delivery cost closer to flat as the client list grows, while one rebuilt per client does not. Plan for both cases before you quote a fee.
Here is an illustrative example, not a result. Take one client on a monthly retainer. The inputs are the fee, the labour to deliver, the software, the revisions the client asks for, and the support you give. Margin is the fee minus the four costs, divided by the fee. Cutting manual reporting lowers the labour and revisions inputs, and that is the only lever this example moves. It says nothing about how much time any tool saves, so fill in your own figures before you rely on it.
Arbling's pricing page and agencies page are the places to check the plan details yourself. If you want to see the workflow, get early access or book a call with the founder, as the agencies page offers.
Where Arbling fits. Arbling offers to bring the client audit, improvements and outcome measurement into one workflow. It recognizes buyers from ChatGPT, Claude, Perplexity and more AI assistants, and its agent pages give AI agents a machine-readable page on the business's own domain with the same facts shoppers see. AI shopper tracking and attribution come on every plan, and the agencies page offers one-time AI-visibility and SEO audits with an optional recurring monthly monitoring cycle.