The Word Our Own Record Would Not Say

Under every customer quote on one of our product pages sat two words: Verified Review. Our own file on those quotes says the opposite.

When we gather reviews, we record whether the platform marked each one as a confirmed purchase. On this product, the shop marks the newer reviews and does not mark the four oldest, which happen to be the four we quoted. We wrote that down at the time, in the evidence file, in plain words. Then we set the verified flag on all four to no, which is why the small checkmark our template shows for a confirmed purchase correctly never appeared.

The label appeared anyway. It came from a default, not from the record.

The cause is dull and worth explaining, because it is the kind of thing that hides for a long time. The field holding a review’s source used to be a dropdown with a handful of fixed options. We changed it to a plain text box so we could write where a review actually came from. The page that displays reviews was never told. It kept checking each source against the old list of options, and anything it did not recognise, which after the change meant everything, fell through to a fallback someone once wrote as Verified Review.

So the fallback did two things at once. It deleted the name of the shop the quotes came from, and it printed a word about trust that we had specifically declined to claim.

Both are fixed. Each quote now names where it was published, as we recorded it. Where a source is missing, the line is simply absent, because the honest response to a missing name is nothing, not a reassuring word. We also corrected one of our machine-readable files, which had been publishing the retired list of options as though it were still the rule.

What we have not fixed yet, and will not pretend is finished: the star rating and review count at the top of that section come from a different shop than the four quotes below it, and the page still names neither. We keep those two records separate on purpose and never blend them into one flattering average, which is a decision we would rather show than describe. Right now the page explains it much further down and not where a reader meets the number. That needs a change to how we store the information, not just to how we display it, so it is written up and queued rather than patched.

The uncomfortable part is that we did the careful work here and then undid it in the display. The check happened. The result was recorded. A default that nobody had looked at in a long time then overrode it on the page, in the one place a reader goes to decide whether to believe us.