Product visibility in AI: what assistants name and what analytics miss
On 14 October 2026, Arbling's Product Visibility in AI page described a gap that many store owners feel without having a name for it: shoppers ask AI assistants what to buy, and the answer names products. Product visibility in AI is the record of which products those answers name, whether they are yours, a competitor's, or your own product sold by another store. Store analytics cannot show that record, because analytics report the visit that follows, not the answer behind it, and they do not tell you which product to fix first. In 2026 that leaves an owner reading sessions while the recommendation that sent the shopper, or failed to send one, stays out of sight. This section sets out the terms used in the rest of the article, so that each stage of the journey stays separate from the next one and no stage is mistaken for a sale.
Five things are easy to blur, so this article keeps them apart. An AI recommendation is a product named in an assistant's answer. A referral visit is a shopper who clicked through from that answer. An agent or crawler request is a machine reaching a page, and the page says crawler and agent requests are never counted as shoppers or orders. An attributed order is an order tied to a visit through recorded identifiers, and the page says every linked order names the method behind the link. Incremental revenue is a further question again, and nothing here treats a linked order as incremental revenue. A recommendation is not a visit, a visit is not an order, and none of them is a guaranteed outcome.
Analytics has its own blind spot. Google Analytics help, read on 17 October 2026, says the traffic-source dimensions in the Traffic acquisition report describe a user's most recent session. Its default channel group has an AI Assistant channel for visits from sources such as ChatGPT and Gemini; a visit whose source is "(direct)" with medium "(not set)" or "(none)" counts as Direct, and one that matches no other channel rule counts as Unassigned. A visit from an assistant that arrives without an identifiable source can land in those last two, so it is worth looking at the answer itself as well as the session. For a deeper look at why counters and attribution differ, see what visit counters cannot tell you.
The seven steps from audit to follow-up for product visibility in AI
Arbling's Product Visibility in AI page, read on 14 October 2026, sets out a sequence of seven steps that carries a store from a buyer's question to a fixed product fact. Work on product visibility in AI begins with an audit of how AI assistants read the catalog and the store. Arbling then turns the catalog into the questions buyers ask and records which products, variants and sellers AI assistants show for each question, next to the competitors in the same answers. Next come AI visits, then orders, then patterns, then fixes, and finally a follow-up. As of 2026 the page presents these as one connected flow and not as separate reports, so each stage hands a cleaner record to the next. The list below keeps the order and the meaning of each step, using the words of the page, so an owner can see where each kind of evidence comes from and where it stops.
- Audit. Arbling sees how AI assistants read the catalog and the store.
- Product visibility. Arbling turns the catalog into the questions buyers ask and records which products, variants and sellers AI assistants show for each question, next to the competitors in the same answers.
- AI visits. AI-referred shoppers, shopping agents and crawlers are counted apart, with the product page each one reached and what it did there. Shopping agents are part of what is often called agentic commerce.
- Orders. Visits are followed to cart, checkout and the order. An order is linked to a visit through recorded identifiers, never by matching a product and a time.
- Patterns. Each product's AI visibility is compared with its real visits and orders, and the page says the comparison shows where they part ways.
- Fixes. The product facts behind each gap are fixed in the catalog. The merchant previews every change and publishes it on confirmation.
- Follow-up. The same buyer questions are asked again after a fix, so before and after are compared on the same terms.
An illustrative example, not a real store. Picture a store that sells an insulated water bottle. The audit shows how assistants read the store's catalog, and the catalog is then turned into buyer questions. The record shows an assistant naming the store's bottle for one question, another seller's listing of the same bottle for a second, and a competitor's bottle for a third. The visit record shows shoppers referred from assistants reaching the bottle page, and separately a crawler requesting the same page. Only a visit with recorded identifiers can be linked to an order, and the crawler request is never an order. The patterns step might show a bottle that is named often but rarely ordered, or the reverse. The fix is a corrected product fact in the catalog, previewed and then published by the merchant. The follow-up asks the same buyer questions again. The example describes a way of reading the evidence, not a result any store has had.
Product visibility in AI for agencies running client work
Arbling's Product Visibility in AI page, read on 14 October 2026, says agencies can run the same workflow for every client, and in 2026 that gives product visibility in AI work a shape an agency can repeat. The workflow has four parts: a baseline, prioritized fixes, a record of what was implemented, and a follow-up the client can read. The baseline is the first audit of which products assistants name for each client. The prioritized fixes come from the patterns step, where a product's AI visibility is set against its real visits and orders. The record of what was implemented shows the client what changed in the catalog and who approved it. The follow-up repeats the same buyer questions, so the client reads a before and after on the same terms. For an agency, that is a report built from evidence rather than from impressions, kept in one format across many accounts.
An illustrative example, not a real agency. Picture an agency with several client stores in different categories. For each client it runs the baseline, then lists the products where assistants name a competitor or another seller. It ranks the fixes by the gap between what the assistants show and what real visits and orders say. The record notes each catalog change the client approved after a preview. A few weeks later the same questions are asked again, and the client reads the comparison. In the report the agency keeps four labels apart: recommendations, referral visits, agent and crawler requests, and attributed orders. It never describes a recommendation or a visit as a sale, and it does not promise the client any outcome.
If you are weighing how to package this kind of work, two guides cover the commercial side: a recurring service for agencies and where a shopping journey loses revenue.
Store owners and e-commerce managers can start with the same baseline. Get early access or run a free scan.
Where Arbling fits. Product Visibility in AI is part of the Arbling suite. Arbling turns the catalog into the questions buyers ask and records which products, variants and sellers AI assistants show for each, next to the competitors in the same answers. AI-referred shoppers, shopping agents and crawlers are counted apart, visits are followed to cart, checkout and the order, and the merchant previews every fix and publishes it on confirmation.