Stockouts and AI visibility: what is known and what is not

The question brands are asking

A common worry runs like this. If a product is often out of stock on the apps where shoppers buy, will AI assistants stop recommending it, or recommend it and send shoppers to an empty listing? It is a reasonable worry, because both matter to a shopper's journey: an assistant names a brand, the shopper goes to an app, and the product is not available.

It is also a question where confident answers are easy to find and hard to justify. This guide separates what can be said from what cannot, and describes a way to look at your own evidence.

What can be said with confidence

Availability is a real, observable state. On a shopping app, a product can be in stock in one pin code and unavailable in the next, and a dated reading shows which it was at a time. That is a measurement.

AI answers are also an observable record. You can ask an assistant the questions shoppers ask, save each answer word for word and record which brands it names. That is also a measurement, with the caveat that answers vary between askings and between assistants.

A third fact follows from how shoppers behave, not from any assistant's internals: an answer that sends someone to a product they cannot buy is a poor experience for the shopper and the brand, whether or not the assistant knew about it. That gives a brand a reason to watch both surfaces even if they turn out to be unrelated in the data.

What cannot be said

There is no published evidence we can rely on that an assistant checks live stock on Indian shopping apps before naming a product. Assistants differ in how they gather information and how current it is, and this changes as they are updated. Treat claims that stockouts reduce AI mentions, or that they do not, as unproven unless they come with the method and the dates.

One example of such a claim: in a piece dated 12 May 2026 ("Why stockouts quietly destroy AI visibility", on eclerx.com), the consultancy eClerx argues that AI search tools deprioritise or leave out products that are out of stock, and that visibility may lag for a while after stock returns. It names no assistants, products, dates or method, so we can neither confirm nor refute it. If anyone sells you this claim, ask which assistants were tested, which products and dates, and how a stockout was recorded: by pin code and time, or only nationally.

There is also a correlation trap. If a brand's stock is often out and its AI visibility is low, both may come from a third cause, such as a small distribution footprint, a weak product content or a recent launch. Seeing the two together does not show that one causes the other.

For these reasons we do not claim a link, and you should be wary of any tool that promises to raise AI visibility by fixing availability. Saved answers show what was said. They do not reveal the assistant's reasoning, so any explanation of why a brand was or was not named is inference.

How to look at both together

You can build a simple comparison from records you collect yourself. Choose a short list of products and the pin codes that matter for them. Record availability on the shopping apps on a regular cadence, with the time. Separately, run the same fixed panel of shopper questions in an assistant at a regular cadence, save the answers and record whether each product's brand is named.

After a few weeks, lay the two timelines side by side. Look for periods when availability changed and check whether the AI answers changed in the following days. One or two coincidences mean nothing. A repeated pattern across products and weeks is worth reporting, with its limits stated. An absence of pattern is a finding too, and it saves you from spending effort on the wrong lever.

• Keep the panel of questions, the assistant and the pin codes fixed during the comparison.
• Record gaps as gaps. A reading that failed is not a reading of in stock or of not named.

What to do in the meantime

Whatever the connection turns out to be, the two problems have separate remedies, and both are worth doing. For availability, use the record to find the products and pin codes where the shelf is empty most often and take that to your supply and account teams. The guide on out-of-stock alerts explains what proof a stock reading should carry.

For AI visibility, work on the pages assistants can read. Make product pages specific and consistent, and check the sources assistants draw on. Measure how often you are named using a fixed panel, as described in the share of AI guide.

Also consider the order of spend. If a product is often unavailable, bringing shoppers to it through advertising or content is wasteful. The guide on checking availability before retail media covers that.

Keep the two records separate

Record the two surfaces separately. A shelf reading describes what a shopper saw on a retail app. An assistant-answer record preserves the question and response. Combining them into one score would hide which source changed.

A stock gap and an assistant omission may occur at the same time without sharing a cause. Review each record on its own evidence before deciding what your team should change.

How to report this to your leadership

Leadership teams like a clean story, and this topic invites one that the evidence does not support. A slide that says stockouts are hurting our AI visibility sounds decisive, and if it is wrong it will cost you credibility the next time you bring data.

A more useful report has three parts. First, the two observations in plain terms: where the shelf was empty and for how long, and how the assistant answered the panel on the same dates. Second, what you can and cannot say: the records show both were observed, and they do not show that one caused the other. Third, what you propose to do: fix the stock gaps for their own sake, improve the pages assistants can read, and keep watching both.

Give the reader the boundaries too. Say which assistant was used, how many questions were in the panel, which pin codes and apps were read, and over what period. A leader who sees the boundaries stated will trust the parts inside them and is less likely to stretch them further than the data goes.

If a pattern does appear over several months, report it as a pattern and label it as an observation in your category. Suggest a way to test it, for example by tracking a product before and after a known change in availability. If no pattern appears, say so plainly. Knowing that two things you feared were linked are not, in your data, lets the team put its effort where it will pay off.

Keep a separate action log for each source. A supply team may need to investigate repeated stock gaps, while a content team may need to inspect unclear product descriptions. Record the question each team is checking and the evidence it will review next. This makes it easier to spot progress without treating a change on one surface as proof of improvement on the other.

Common questions