Your product is in the catalog. As far as you can tell, it's eligible. Someone on your team asks an AI assistant for a bag like yours, and your bag isn't in the answer. Is something broken? Or have you just confused two different things: being included, and being chosen?

This is an educational series from Verity Score for Shopify merchants. Last time, we made a dated record of one product in one market, and the routes its information could take. We ended with a promise: to separate a mention, a source citation, a recommendation, a visit and an order. Today we keep that promise, with one question. How do you tell the difference between being included and being chosen?

We'll stay with the fictional luggage team from the first episode. The store, the products and every answer I describe are invented for this example. No real store, shopper or result is involved. The team sells a convertible backpack to customers in the US. It can be carried as a backpack or by a side handle. The team has checked its catalog status, as far as its admin shows it. Now they want to know what happens when people actually ask.

On a Tuesday morning, someone on the team, based in the US, opens ChatGPT in its consumer interface, signed in to their own account, and starts a new conversation. They type: "What's a good backpack for commuting by train with a laptop?"

The answer lists several bags. Their convertible backpack is one of them. The answer describes it in a sentence, and links to an external review site, not to the store's own page. The price shown is last month's price, before a small increase.

Then they ask a second question in the same conversation: "What bag should I take on a short flight if it's my only bag?" This time, their backpack doesn't appear. Other bags do. Some of them are from much larger brands.

So the same product was in one answer and missing from the other, on the same morning, for the same person. The team's first reaction is to compare the two answers like search rankings. Position three in the first one. Nowhere in the second. That reaction is exactly what we want to slow down.

Let's separate the states. They are different kinds of facts, and they need different evidence.

Included is a state of your data. Your product exists in a source that can reach a channel. For Shopify, that might be Catalog eligibility for a connected channel, as your admin shows it on a given date. Inclusion concerns the available data, not any single answer. Eligibility alone does not prove that this exact product has been transmitted; keep that status separate.

Observed is an event. A specific answer, to a specific question, in a specific context, contained something. Inside that event, there are still distinctions. Your store or product can be mentioned, which means its name appears. A page can be cited, which means a link is offered as a source for what the answer says. And a product can be recommended, which means it's presented as a good fit for what was asked. In the first answer, the backpack was mentioned and arguably recommended. The citation went to someone else's page.

Then there's a visit, which is someone arriving on your store. Your reports may or may not be able to connect it to that answer. And there's a purchase, which is an order you can see, whatever its origin.

These are not steps on a staircase. A page can be cited without its merchant being recommended. A recommendation can lead to no visit. An order can happen weeks later with no visible link to any answer.

Now, why might the bag appear in one answer and not the other? OpenAI's help article on shopping in ChatGPT, which we reviewed on October twenty-second, twenty twenty-six, says that product selection considers the question, the context, and first-party and third-party metadata. It also says that not every available product will necessarily be shown. So the second question wasn't a retest of the first. It was a different question, with a different job: a single bag for a flight, rather than a laptop bag for a train. And it was asked in a conversation that already had history.

The same help article says two more things worth knowing. Updated prices and shipping details can take time to appear. That may explain last month's price, though we can't confirm the cause from outside. And reviews and ratings shown in results are not verified by OpenAI. So a star rating next to a competitor doesn't tell you that it's been checked.

What about Google? For its AI features in Search, Google says a page needs to be indexed and eligible to be shown with a snippet to appear as a supporting link. Meeting those requirements doesn't guarantee that it will be indexed or shown. Again, inclusion conditions are not selection guarantees.

Here's what the team can conclude honestly. Their bag was included in one observed answer, for one question, on one morning, in one signed-in conversation. It was absent from another answer, to a different question, in the same conversation. That's all. It's not a ranking. Position three is just where it appeared that time. And the absence in the second answer has no known cause. Maybe the product page never says whether the bag works as a single carry-on for a flight. Maybe other bags simply fit that question better. Maybe the conversation history pulled the answer toward commuting. From outside, the team can't tell.

The citation deserves a second look too. In the first answer, the link went to an external review site, not to the store. That means the sentence about the backpack may have been supported by someone else's description. Is that description accurate? The team doesn't know yet. It goes on the list for a later episode, where we'll look at what other sources say about your products, and what you can do when they get it wrong.

What the team can do is stop asking, "Are we ranking?" and start asking, "Under which conditions do we appear, and for which jobs?" That's an observation question. And it can be answered more reliably than a ranking question.

The team decides to build an observation sheet that someone else could repeat. That's your action this week.

Start with two or three queries that describe different jobs for the same product. For our backpack, that might be commuting with a laptop, and a short flight with one bag. If the queries come from real customer questions, make sure you're allowed to use them, and strip anything that identifies a person. Otherwise, label them as hypotheses. They are not a measure of consumer demand.

For each run, record the date, time and time zone. Record the country the person is in, and how you know it. Record the language, the service, and the type of access, whether that's the consumer app, a browser or something else. Note whether the person was signed in, and whether it was a new conversation. Then record the exact query and what the answer contained: which products appeared, in the order shown, whether yours was mentioned, cited or recommended, and which pages were linked. Finally, compare the price shown to your actual price at that moment.

Use a new conversation for each query, unless you're deliberately testing a conversation context, and write down which one you did. Keep every result, including absences. Don't turn the order of products into a score. The full template is in the show notes.

And if the price shown doesn't match yours, as it didn't for our team, write down both prices and the time. Record the difference, but do not assume where it came from. First check that your own product page and any feed you send show the current price. If they do, the gap may simply be a delay. If they don't, you've found something to fix at home. We'll trace that kind of drift in detail in episode four.

One more limit. This sheet doesn't tell you how many shoppers ask these questions, or what they see in their own conversations. It tells you what your team observed, under recorded conditions. That's narrow, but it's real, and you can repeat it next month under documented conditions. That improves comparability without holding the model or all context constant.

So, where are we? Being included is a state of your data. Being chosen is an event inside one answer. A mention, a citation, a recommendation, a visit and an order need different evidence. And an observed position isn't a rank. Next time, we'll look at that second question again: if your product page never says what job the bag is for, how could anyone choose it for that job?

Sources and review dates are in the show notes. This episode is read by a synthetic voice.