The look can repeat. The words cannot.

There is a piece of advice every seller hears early and follows correctly: make it consistent. One light, one palette, one treatment across the catalogue, so a shopper landing on your fourth product recognises it as yours. That advice is sound, and this site has argued for it at length. It is also the source of one of the most common and least visible mistakes in listing copy, because sellers apply it to the words as well — and in the words it does the opposite of what it does in the pictures.

Two rules that point opposite ways

The reason they diverge is that a person and a retrieval system are looking for different things.

A shopper looks at your images in sequence and reads repetition as competence. The same backdrop on every product means somebody is in charge. The same crop, the same key light, the same grade — these are the marks of a shop rather than a pile of listings, and a catalogue that looks assembled from six different sources reads as a reseller even when it is not.

A search engine or an answer engine looks at your pages side by side and reads repetition as redundancy. It is not judging your taste; it is deciding which single URL to show for a query, and when two of your pages say the same thing in the same words, it has no basis for showing both. One of them wins and the other is suppressed — and which one wins is not your choice.

So the same instinct, applied to two media, produces a coherent brand and a self-competing catalogue at the same time.

What "near-identical" actually means

It is looser than people assume, and that is the trap. Nobody publishes the same description twice on purpose. What they publish is a description with a fixed skeleton: an opening line about the brand, a closing line about shipping and returns, a middle sentence of specification — and only the specification changes. If the skeleton is two-thirds of the text, then two listings are two-thirds identical, and a deduplicating system does not need them to be any closer than that.

The useful way to check is mechanical rather than aesthetic. Take two of your product descriptions, and ask how much of the text you would have to delete before what is left is only true of one product. If the answer is "most of it stays", the pages are near-duplicates no matter how different they feel to write.

We did this to ourselves, and then measured it

This site publishes seven comparison pages, one per tool a buyer might be choosing between. Each one opens with a short passage that is also the string marked up for answer engines to quote — the one sentence-or-two we are explicitly asking a machine to lift.

Until the day this post was published, those seven passages were generated from a template. The first clause named the other tool and varied; everything after it was one hard-coded pair of sentences about Nitaris. Measured, that came to 169 identical characters out of a 275-character average: 61% of every passage was byte-for-byte the same as its six siblings, and the only part that varied described somebody else's product rather than ours.

The comparison that made it unarguable was against our own other corpora, which are written by hand one page at a time. Overlap between any two of the seven comparison passages ran between 0.49 and 0.66 on a standard text-similarity measure. Across the twenty question pages the worst pair scored 0.29, and across the blog, 0.28. The ranges did not even touch — the templated set's *best* case was worse than the hand-written sets' *worst* case.

The outcome matched. Those seven pages are indexed, they earn impressions on rival brand terms, and they are the only group on this site with no citations at all in the generative-AI report Search Console publishes — while every other corpus has some. Seven URLs were offering retrieval one passage dressed seven ways.

The fix was not clever. Each page now says what the choice actually turns on for that specific comparison — who is on screen, an edit versus a shoot, control versus economics, a still versus a moving camera — in its own words. Same length, same sourcing rule, same facts. The shared span went from 169 characters to one, and the worst pair from 0.66 to 0.30.

What to write instead

The practical version, for a catalogue rather than a comparison set:

• Write the sentence that is only true of this product first, and put it at the top. Whatever is true of everything you sell belongs lower down, or on one page that all of them link to.
• Move the boilerplate out of the description entirely where the platform lets you. Shipping, returns and brand story are page furniture; they do not need to be inside the text that identifies the product.
• Name the thing a buyer of this item specifically worries about. For a watch that is the bracelet and the lug width; for a candle, the burn time and the throw. Those details are unrepeatable by construction, which is exactly why they work.
• Check your two most similar listings, not your two most different. The pair you would never think to compare is the pair a deduplicating system is comparing.

None of this argues against consistency in the images. Fifty products should look like one shop — that argument holds, and it is the reason the treatment is worth repeating. The point is narrower: the look is what you repeat on purpose, and the words are what you must not.

It also sits beside a rule about the images that has the same shape. Within a single listing, a listing is a sequence, not a gallery: the four photo slots each answer a different question, and a row of near-identical photographs wastes five of them. That is the same economy one level down — repetition inside one listing wastes slots, and repetition across listings wastes pages.

The reason this mistake survives so long is that it is invisible from the inside. Every page looks fine on its own; the problem only exists in the relationship between them, and nobody reads their own catalogue two listings at a time. It took us a measurement to see it on seven pages we had written ourselves.