AI visibility for online shops: named as the shop, the brand or the product

Three things an answer can name: the shop, the brand and the product

A local business is named or it is not. A shop can appear in three ways. An answer to "vegan leather wallet with RFID blocking" might name your shop as a place to buy one, name your brand as a maker of one, or name the wallet itself by its product name, and it can do any of these without the others. Each is a different kind of AI visibility, and a reading that folds them together cannot tell you which page to fix.

The three can come apart. A brand you make and also sell through resellers can be named beside a reseller's link. A product can be named with last season's price. A shop that stocks other makers' brands can be named for something it no longer carries. Each is a finding with its own fix, so the record keeps them in separate columns.

Where the answer sends the buyer is part of the reading

Record what the answer points to as well as who it names: your product page, another page of yours, a marketplace listing of your own product, a reseller, a review article, or no link at all. An answer that names your wallet and links a marketplace has sent the order somewhere else; one that names your shop and links your homepage has left the buyer to find the product. Both tell you which of your pages the answer did not cite.

We hold no figure for how often a shopping answer names a marketplace rather than a shop. In the probe that ran beside our audit of 9,483 small business websites, 99.4% of the names Claude gave were actual businesses and 0.6% were directories or aggregators, but those were local trade questions, such as the best electricians in a metro, and we do not carry the figure over to products. For your own products, the panel is where that number comes from.

A question panel built from needs and conditions

A shop's panel starts from its product lines. For each line you want to be named for, write the need in the buyer's words, then add the conditions buyers attach to it, one at a time and in the combinations they type. Add two or three questions about the shop itself (whether it ships to a country, how returns work, whether it is legitimate) and one comparison. Typed the way buyers type them, these are the prompts that go to every assistant, every month.

For paying clients our panel is twelve questions, set once and then left alone so that readings compare; the method is published. Across three product lines, that is three questions per line and two or three about the shop. How to keep the file, and why answers vary between runs, are covered in setting up prompt monitoring and how to measure AI visibility; neither changes for a shop.

What to write down for each answer

The standard record says whether you were mentioned, whether the answer was an AI recommendation of you, whether you came first, and who else was named. A shop needs five more columns:

• Which of the three was named: the shop, the brand, the product, or more than one.
• Where the answer pointed: your product page, another of your pages, a marketplace, a reseller, a review page, or nowhere.
• The facts it stated about you (a price, a destination, a return window, a material), each marked right or wrong against your own pages on the day of the run.
• The other shops, brands and marketplaces it named, so a competitor that appears for one condition and not another stands out.
• The sources it listed, where the assistant shows them, which tells you which of your pages it cited.

The accuracy column costs the most when it is wrong. A buyer told the wrong shipping cost, or a price from last season, reaches your checkout expecting something else. The seven GEO KPIs count accuracy beside the named rate for that reason, and share of voice counts your part of every name given; both work for a shop once the three kinds of name are kept apart.

Reading the first result

The first reading is a baseline, not a verdict. Read it question by question. A product named for the need but not for the destination points at the shipping page. A brand named without the shop points at the about page and at how you name yourself. A shop named with a wrong price points at a page and a feed that disagree. Each pattern names the page to work on first.

Expect the baseline to be thin. In the probe that ran beside our audit, one local trade question per business went to Claude with web retrieval on, and 14.9% of the 9,240 businesses were named, 17.7% among the 5,928 answers that gave a differentiated list. That was one question, one assistant and one day, about local services, so it is context for a low start and not a rate for shops.

Then choose what to work on. Where a small shop has room to be named helps decide which questions are worth the effort, and the overview for online shops lists what each page needs.

What a panel reading cannot tell a shop

It does not count sales. An answer that names your product may end in an order on your site, on a marketplace, in a store down the road, or nowhere, and the panel sees none of it. The readings that come closer, from assistant referrals to a question on the order, are in measuring AI recommendations for a shop.

It does not say why an answer came out the way it did. The operators do not publish how their answers are composed, and a reading records what was said, not the reasons. What it shows is direction on fixed questions over months; one month's change on one question sits inside the ordinary variation between runs.

And it only covers the questions you chose. A product line left out of the panel can be named or missed without your knowing, which is a reason to review the panel each quarter: add questions as new lines, and never reword the old ones.