The Awards We Never Won

Near the bottom of every product page is a short section for what the outside world has said. On our page for a French pocket knife, that section was headed with a claim that the knife had won two awards. It has won none.

Nobody wrote that sentence. The page assembles it as it loads. We keep a short list of marks under each product, the page counts them, and it printed the total followed by the word awards. On this knife the two things in that list are a score out of ten from an outdoor magazine and a note that it has been made in France since 1890. Neither is an award, and the page had no way to know, because it cannot read what it is counting.

The same heading had a second half. Whenever we hold a press quote for a product, it added the words and the press agrees. It never read the quote. Directly beneath that heading on this knife sits a review we published precisely because it is unflattering: an outdoor testing site explaining that the knife scored low against competition costing nearly ten times as much. We chose to put that there. The heading talked over it.

Both halves are gone. The heading now names the publications that reviewed the product, which is something we actually hold, and it claims nothing about what those reviews thought or what the marks amount to. The marks still sit underneath it, where a reader can see exactly what they are and decide for themselves.

The uncomfortable part is that this line was never reviewed by anyone, because it was never written by anyone. Every sentence we publish is written by a pass and checked at a gate. This one was assembled at the moment of loading, out of a count and a yes or no, and it made a claim that neither of those could support.

One other product page carries the same heading and happens to be telling the truth: its three marks genuinely are design awards, and the press quote we hold for it genuinely is admiring. That was the data being kind, not a check working. A sentence that is true by luck is not a true sentence, it is a coin that landed well.

We found this while reading that same page section by section. Earlier today we published a correction about a label on customer quotes two sections higher up. That one printed a word about trust our record refused to claim. This one printed a word about awards our record never held. Same page, same afternoon, same cause: a template making a claim on our behalf, in a place nobody thought to look, because templates are supposed to arrange facts rather than assert them.

Machines were being told the same thing nine days later. Our structured data published an award on 78 product pages, switched on by a word we typed, and no reader of this page or any other could have seen it.