What the AI Found on Your Competitor's Page That It Didn't Find on Yours (A Side-by-Side Check)
Put your sofa's page next to the one the AI recommended and compare them on the shopper's questions. It won't tell you why the AI chose, but it shows what yours left out.
You ran the test. You asked an assistant for a sofa like yours and it named someone else's. The temptation is to reach for a theory: they're bigger, they spend more on ads, the AI likes them. Those might be true and you can't do anything about them. There are also things you can't see at all: the shopper's context, availability, reviews, third-party coverage. What you can see is two product pages, and comparing them won't tell you why the assistant chose, but it will tell you what yours left out.
Here's how to do it.
Set up the comparison
Take the request you made and the product the assistant recommended. Open that product's page in one tab and your closest equivalent in another. Then write down the questions the request contained, explicitly or by implication. "Sofa under 80 inches, pet-friendly fabric, ships in under four weeks, under $2,000" is at least six:
- What's the overall width?
- What's the fabric made of, and is it cleanable or durable in a way that means "pet-friendly"?
- How long does it take to ship?
- What does it cost?
- How many does it seat? (Implied by "sofa".)
- Is it in stock? (Implied by the shipping question.)
Add the questions the assistant answered unprompted in its recommendation. If it said "the cushions have removable, washable covers", that is a claim to add to your checklist and verify against the cited source and correct variant.
Score both pages
For each question, record three things about each page: is the answer present, is it present as text (not an image, not a PDF, not behind a click that loads content), and is it attached to the right variant.
Do it honestly and it might look like this. The pair below is hypothetical, built to show the method:
| Question | Their page | Your page |
|---|---|---|
| Width | "Overall width: 78 in" in a specs table | In the dimension diagram only |
| Fabric | "100% polyester performance weave, 50,000 double rubs, cleanable with water-based cleaners" | "Premium performance fabric" |
| Shipping | "Ships in 2 to 3 weeks" on the page | Calculator after ZIP code |
| Price | $1,899, text | $1,799, text |
| Seats | "Seats 3" in the title | Implied by the photo |
| Stock | "In stock" | Not stated |
| Washable covers | Stated in bullets | True, but only in reviews |
Yours is cheaper. Yours may be the better sofa. On this request, theirs answered all seven questions in text and yours answered one. That doesn't establish why the assistant chose theirs. It does show where your page gave it less to work with.
What this does and doesn't tell you
The comparison identifies information gaps worth fixing. It does not establish which factor caused the assistant to recommend one product over another; price, availability, reviews, third-party information, product metadata and the shopper's own context can all play a part. If your page answered everything and still wasn't recommended, look at those, and at structured data, product feeds and whether your site is reachable by the assistant at all. The page-level gap is the first thing to check because it's the cheapest to fix, not because it's always the answer.
It does reframe "we need better copy". Their copy isn't better. It's more complete about facts. The difference between "premium performance fabric" and "100% polyester, 50,000 double rubs" is not writing quality.
The three kinds of gap
Running this across a few products, the gaps sort into three kinds.
The fact exists but is hard to read. Width in a diagram, specs in a PDF, details in a tab that loads on click. The business knows the answer and the page technically shows it, but some systems may struggle to retrieve or interpret it. Fix: get it into text and into the record. See dimensions in the image.
The fact exists but is vague. "Performance fabric", "solid wood", "fast shipping". A category where a shopper and a machine both need a value. Fix: replace the category with the fact, from the supplier spec or your own knowledge.
The fact isn't on the page at all. Removable covers, packed dimensions with the legs off, whether the legs come off, weight capacity. Known to customer service, known to the buyer, absent from the product data. Fix: verify it, then add the attribute to the record so it can reach the page.
Each kind has a different owner and a different cost, which is why sorting them matters. The first is often a template change. The second is a data-quality pass. The third is a catalog project.
Do it for ten products, not one
One side-by-side tells you about one page. Ten tell you about your template and your catalog. If width is in a diagram on all ten, that's one template fix, not ten page fixes. If fabric composition is missing on all ten, the fabric attribute is empty across the range, and the question becomes where the composition lives (the supplier spec sheets, probably) and how to get it onto every record.
That shift, from fixing pages to fixing the record the pages are built from, is the subject of why fixing one product page doesn't fix the range. For the whole argument, see being found by AI.
One more thing to check on their page
Scroll to the bottom and look at the specifications table, if there is one. Then look at yours. Often the competitor who's being recommended has a boring, complete table, and the one who isn't has a beautiful page with the facts in pictures. We can't tell you the table is why. It's a habit worth copying regardless.
