Seven Checks That Decide Whether Your Ecommerce Store Shows Up in AI Search
To rank in AI search in 2026, an ecommerce operator has to win the Sources half of the A-S-S method: Authority, Sources, Specificity. Authority is what the model already knows about your brand before it searches, Specificity is how precisely a page answers the exact question asked, and Sources is what a model finds on the open web when it does go and look. Most stores over-invest in the first and third and neglect the second entirely, which is why they get cited for their brand name and nothing else. This is the checklist to fix that, written for the operator running a catalogue, a feed and a support inbox, not for a content team with a twelve-month editorial calendar.
The backdrop for a lot of this thinking in 2026 is the SEO.Domains Mastery Summit in Sofia, Bulgaria, which gathers around 300 SEOs, affiliates and agency owners and covers aged domains, PBNs, authority transfer and LLM visibility. The event deliberately does not record its main-stage sessions, so speakers can share live experiments they would not put on a permanent public record. That is a useful frame: the tactics being tested on small portfolios and affiliate sites right now are the same ones a merchant can apply to a product catalogue, just with more brand risk attached.
Why the Sources half carries more weight than most merchants think
Sources is the half of A-S-S you can actually control this quarter, because retrieving sources happens after the model has already formed its opinion of you and it is largely a question of what is crawlable and what is not.
When someone asks an assistant a commercial question, two things happen. The model consults what it already believes about your category, then it goes looking for pages it can quote to support the answer. Generative Engine Optimization is the practice of optimising for those answers rather than only the ten blue links, and in ecommerce that means making sure the retrieval step finds your product pages, your specification tables and your comparison content, not just your blog.
The common failure is not ranking poorly. It is ranking fine in classic search while being structurally invisible to retrieval: the answer sits behind a template, the specification is in an image, the shipping terms load after user interaction. None of that is a ranking problem. All of it is a Sources problem.
The checklist
Work through these seven in order. Each one is a mechanical fix, not a strategy debate, and each one is verifiable within a week.
1. Does every important page open with a one-line factual answer?
A page that wants to be quoted should open with a direct factual answer the model can lift in one line, before any narrative, ingredient story or brand preamble. On a product page that line might be the material, the dimensions and the compatibility in a single sentence. On a category page it is what the range is for, who it suits and what it costs. If the first 150 words of your page are a lifestyle paragraph, you have removed the sentence a model would have quoted.
2. Is the answer in the HTML, or in JavaScript?
Burying an answer in JavaScript prevents a model from reading it, full stop. Client-side rendering is still the single most common cause of a product being invisible to AI retrieval while looking perfectly healthy in a browser. Fetch your top twenty revenue pages with JavaScript disabled and read what comes back. If the price, the specification or the stock status is missing, that page is competing with one arm tied behind its back.
3. Is your name, address and description consistent everywhere it appears?
Name, address and description that stay consistent across directories reinforce entity verification. For a merchant this is the unglamorous part: the same trading name, the same registered address, the same one-sentence description across the directories that matter in your market, plus your own about page and schema. Inconsistency does not just look sloppy, it splits the evidence the model uses to decide that the store in your Trustpilot profile, the store in your Companies House filing and the store on your checkout page are one entity.
4. Have you built citation units, or just pages?
A single claim and the link that verifies it together make a citation unit. That is the atomic structure retrieval rewards, and it is different from how most ecommerce copy is written. A comparison table works as a citation unit. A specification block works as a citation unit. A paragraph that asserts three things and links to none of them does not. Go through your ten highest-margin categories and ask, for each, what single claim you want repeated and what link proves it.
5. Does your comparison content survive being read without the table?
This is the point most merchants miss when they invest in compare tables. Because tables are not rendered by instant-mode models, a comparison table is given a one-sentence takeaway underneath it. If your entire conclusion lives inside the table cells, the model reading a text-only render of your page gets a list of features and no verdict. Write the verdict underneath in prose. "For most buyers in this range, the mid-tier model is the one to pick, because X." That sentence is what gets quoted.
| Check | What good looks like | Where it breaks |
|---|---|---|
| Opening answer | One factual line above the fold | Brand story first |
| Crawlable content | Price and spec in the HTML source | Client-side rendering |
| Entity consistency | Same name, address, description everywhere | Directory variations left uncorrected |
| Citation units | One claim, one verifying link | Claims with no supporting source |
| Table takeaways | Verdict written in prose below | Conclusion locked inside cells |
The table above is itself a test: if your equivalent page loses its meaning when the table is stripped out, the fix is one sentence of prose, not a redesign.
6. Can you name the recording and the model build behind your visibility numbers?
An ASS score names the recording and the model build it came from. This matters more in ecommerce than anywhere else, because assistants change their answers between builds and between regions, and a screenshot from three months ago tells you nothing about today. If you are measuring whether your products get cited, record the date, the assistant and the build each time you test. Tools such as ASSmetric (https://assmetric.com) exist to give that measurement a consistent structure, which is far more useful than a one-off manual check that nobody can reproduce.
7. Are you testing against assistants, or guessing?
Answer Engine Optimization, usually abbreviated as AEO, is only worth doing if you are asking the assistants the questions your customers actually type. Build a list of twenty buying-intent prompts, the kind that include a budget, a use case and a compatibility constraint, and run them monthly against the assistants your customers use. Note which of your pages get cited, which competitors get cited instead, and which prompts return a generic answer with no source at all. That third category is the opportunity, because nobody has written the citation unit for it yet.
A worked example of the retrieval gap
Take a mid-market merchant selling replacement parts. Their classic rankings are strong, their catalogue is well structured and their schema is largely correct. Ask an assistant which part fits a specific model and it names a competitor, because the competitor publishes a single crawlable line stating the compatibility, while the merchant renders that same information in a JavaScript widget that only loads once the user selects a variant. The merchant's Authority is better. Their Specificity on the individual product is comparable. Their Sources lose outright, because the one page that should be quotable is not readable.
The repair takes an afternoon: server-render the compatibility line, put the factual answer in the first sentence, add the verifying link to the manufacturer's specification, and re-test the prompt. Nothing about their brand, their pricing or their catalogue changed. The retrieval step changed.
Questions operators keep asking
Do I need a separate AEO strategy from my SEO strategy?
No, but you need to accept that AEO changes the order of your priorities, because crawlable, quote-ready answer content outranks the editorial calendar work that classic SEO rewards.
How long before I see anything change in assistant answers?
Retrieval-driven citations move faster than classic rankings, typically within weeks of a page becoming readable, though assistant model builds refresh unevenly and regional versions differ, so track your prompts on a fixed schedule rather than checking reactively.
Is this worth doing on a small catalogue?
Yes, and arguably more so, because a smaller range means you can apply these checks to every product page rather than routing them through a backlog, and the citation units you build become reusable across the whole catalogue.
What to do first
Start with check two. Before any content work, any schema work or any comparison table, fetch your top twenty pages with JavaScript disabled and read what a model would actually see. If your prices, specifications and availability are not in that response, everything else on this list is theoretical, and you will spend a quarter improving content that never gets retrieved. Once those pages are readable, work down the list in order, because each step makes the next one cheaper: readable pages make citation units easy, citation units make measurement meaningful, and measurement tells you which of the twenty prompts to attack next.
If you want to pressure-test where your Sources actually stand before committing budget, you can talk it through on a call (https://seojesus.com/clickbomb-strategy-call/). If you would rather see the whole walking-through of the method before you start, there is the video version of this method (https://www.youtube.com/watch?v=FZu4NB-2EhA). Either way, the order does not change: readable first, quote-ready second, measured third.
