How AI shopping assistants pick what to suggest
A shopper asks an assistant where to buy something. Several shops may be named, and yours is not one of them. The next question is always the same. What was the rule? Companies publish some broad inputs, but not a complete or fixed formula. OpenAI lists factors including the shopper's request, price, availability, reviews, product information, and merchant quality. Here is what is actually known, what is not, and where that leaves you.
1. Start with the part nobody will tell you
Companies publish some broad inputs, but not a complete or fixed formula. OpenAI lists factors including the shopper's request, price, availability, reviews, product information, and merchant quality.
So when you read an article that lists seven ranking factors in order, or a service that offers to move you up, ask where the order came from. It came from watching results and guessing. That guess may be sensible. It is still a guess, and it goes stale quietly, because the underlying systems can change without notice.
This page is short on factors on purpose.
2. The one company that has written it down plainly
Google publishes guidance for site owners about its AI answers, and two lines in it are worth more than most of what is written elsewhere.
The first is that these answers are built on Google's ordinary search systems rather than on a separate one. In Google's words, the best practices for search "continue to be relevant" because the AI features are rooted in its core search ranking and quality systems.
The second is blunter. Google says there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." It goes further and says you do not need to add any special extra code to your pages for this.
Read that carefully, because it cuts both ways. Google says there is no separate AI-only code or paid shortcut. Normal Search and Merchant Center foundations still apply, and meeting them does not promise inclusion.
3. Two things happen, and only the first is about your store
Google describes its AI answers as pulling pages in through its normal search results, and as running several related searches at once behind a single question a person typed.
Put plainly, there is a gathering step and then a choosing step.
You can test parts of access and readability today. A scan may find a block or missing product facts, but it cannot prove complete access across the whole store.
Companies publish some broad inputs, but not a complete or fixed formula. OpenAI lists factors including the shopper's request, price, availability, reviews, product information, and merchant quality.
Anyone who blurs those two steps together is selling the second one on the strength of the first.
4. What an assistant can actually read
You can do this, apart from one part that needs a developer.
Three questions cover most of it, and all three are ours to check rather than anyone's to guess at.
Our 26-point access check reads the site-root rules in robots.txt. Separate response checks may warn that a firewall interfered, but those warnings do not identify a Cloudflare setting.
Our 19-point product-details check examines one product page we can locate for machine-readable price, availability, and product information.
Our nine-point answer-content check looks for FAQ data on the sampled pages or question-shaped headings on the homepage. It does not judge whether sizing, delivery, or returns answers are complete.
5. Some of what gets read is not on your website
Google's own advice to shop owners includes keeping the product information you file with Google up to date, alongside the site itself. So part of what an assistant may lean on lives in a feed, not on your pages.
At the same time, shared standards are being written so that assistants can search a catalogue, build a basket and check out directly. One of them, published openly at ucp.dev, sets out steps for catalogue search, basket building, checkout and order updates.
The point is not that you should chase any of this today. The point is that the surface keeps widening, which is one more reason a fixed recipe cannot exist. A recipe written for last year's surface is not wrong so much as beside the point.
6. How to judge an offer
If someone offers to put your shop into AI answers, ask them one question, and ask it plainly. Which part of this do you control?
An honest answer is some version of "what your store publishes, and nothing after that". That is a real service and it is worth paying for if you do not want to do it yourself. A seller can improve some published inputs, such as product accuracy, availability, price, and service quality. No seller controls the final result or can promise placement.
The same test works on us. We measure what your store publishes on the day we look. We do not touch the choosing, and we would not know how.
What this does not do
This page cannot tell you why one assistant named one shop on one day, and neither can anyone else outside those companies. It will not give you a list of factors to work through. An honest list separates published foundations from unknown weighting: maintain access, accurate product information, current availability, useful content, and sound merchant operations, without presenting them as a placement formula.
There is a limit worth naming on our own work too. A scan tells you what was true about your store at the moment we looked. It is a dated measurement, not a prediction, and a store that reads perfectly can still go unmentioned for reasons nobody publishes.
What we can tell you is what your store is publishing right now, and what to change. On this subject, that is the honest whole of it.