When a category label is not a market: a three-hundredfold range under one heading

A category on a comparison site does two jobs. It groups things that do the same work, and it implies that the things in it are alternatives to each other. The second job is the one that fails quietly, and AI writing tools are where it fails most clearly in our records.

Measured across the vendors we hold in that category, the range from cheapest tier to dearest is about three hundredfold. That alone might just mean a broad ladder. The sharper fact is that the ranges do not overlap. One vendor's entry plan costs about ten times what the cheapest vendor charges for its top tier, and more than four times the dearest tier of any other vendor in the category. There is no headcount, no volume and no configuration at which a buyer would be choosing between those two products, because the cheaper one runs out of ladder an order of magnitude below where the dearer one starts.

That is not a criticism of either vendor. One is a plugin sold to individuals and small sites; the other is an enterprise platform sold on annual contracts. Both do something reasonably described as AI writing. The problem is entirely with the label, and with what a reader assumes when two products appear under it: that they were considered against each other, and that the ranking means one was preferred. Neither is true when the ladders are disjoint.

The practical consequence is worth stating for any category, not just this one. Before trusting a ranking, look at the spread. If the cheapest and dearest entries are within a factor of two or three of each other, they are plausibly alternatives and a ranking is doing useful work. If they are separated by an order of magnitude, the list is a directory rather than a comparison, and the useful question is which segment you are in rather than which item is best. We would rather say that than publish a table implying a contest that never took place.

The same category contains a second thing worth knowing, about how one of those prices is expressed. The enterprise vendor sells on an annual basis only, and its per-month figures do not appear on its page — they are its annual totals divided by twelve. Our record keeps the division visible in a provenance field, so a figure we produced for readability is never mistaken for a rate the vendor quoted. On a product priced in five figures a year that distinction matters more than usual, because a per-month figure invites a comparison against monthly-billed competitors that the vendor is not actually offering.

And there is a third case, which adds to a pattern this series has been tracking. One vendor's pricing page rendered its plan prices as empty elements — no figures at all — while the add-on prices on the same cards rendered normally. That is a partial failure rather than a wholesale one, and it is more dangerous than a blank page, because a reader or a reader's software sees money on the card and may not notice which money is missing. The record keeps the add-on figures, because those genuinely rendered, and holds no plan price at all. Its billing toggle also ships in the annual state, and neither state produced a plan figure.

Taken together the three make a short discipline. Check the spread of a category before reading any ranking within it. Check whether a per-month figure is one the vendor quotes or one somebody derived. And treat a page that shows some prices and not others as less trustworthy than one that shows none, because the partial version is the one that looks complete.