Buying AI tools for five people

L'article ci-dessous est publié en anglais. Chaque chiffre qu'il contient est le chiffre publié par l'éditeur lui-même, avec la date à laquelle nous l'avons lu.

A team of five is where AI tool pricing stops being a single number. The same headline figure can mean one bill or five depending on a word printed beside it, and the four tools in this directory that charge by the person each use a different word for it. Everything below is read off figures our own profiles already carry, with no arithmetic done for you and none needed.

Four vendors, four words, one multiplier

Four of the twenty three tools profiled here print a per person unit on the plan we quote, and no two of them call it the same thing. Descript says per person, Hex says per Editor, Jasper says per seat, and Notion AI says per member. One concept, four labels, and the label is the part a comparison table drops first.

Every figure there is the vendor's own, and each profile names the document it was read from and the day a human opened it.

The other nineteen profiles do not print such a word, and that is not the same as their price covering a whole team. Our ledger records what the card said. Where a card is silent on the unit, we are silent too, and the question goes back to the vendor's page rather than to a conclusion of ours.

A seat price is not the bill

Hex answers this on one page in two separate places, which is what makes it the clearest case rather than the worst one. Its plan cards are denominated per Editor per month, so the headline scales with how many of you there are. A second table further down the same page prices compute by the hour, and that one scales with what the notebooks actually run.

On the page we read for the figure above, that compute table is free at its three smallest sizes and starts charging beyond them, and its own footnote says it bills by the minute rather than by the hour it quotes. So five people running light notebooks pay the seat figure and nothing else, and the same five running heavy jobs do not.

For one person the second meter is a detail. For a team it is the difference between a quote and a guess, because the two meters scale on different things: one on headcount, the other on work. Neither number alone tells you what the year costs, which is the reason to ask the question before reaching for the figure.

The cheapest plan carrying the thing is not the cheapest plan

Notion AI has no price of its own. It is a set of capabilities gated by which Notion tier a workspace is on, so the question is not what the feature costs but which rung stops treating it as a sample. Notion's two cheapest plans both list the same bullet, Trial of Notion AI, and Business is the first card on that page whose contents change.

That is why the figure in the table above is Business rather than the cheaper paid tier a price comparison reaches for first. Take the wrong rung across five people and the mistake repeats five times, in the direction that makes the tool look cheaper and the comparison look tidier.

An error pointing that way is the one nobody goes back to check. It agrees with the vendor, it agrees with whoever is trying to get the purchase approved, and it agrees with the person who wrote the table. That is exactly why our methodology puts the rule in writing: the figure has to be the price of the product named at the top of the page.

Where the price stops being a web page

Jasper's pricing page shows two plans and puts a figure on one of them. Pro is stated in its billing questions with both cadences in a single sentence. Business, the tier above it, is labeled custom pricing and its button reads Contact Sales, so if the capability you need turns out to live there, no web page settles your bill.

For one person an unpriced tier above is usually irrelevant. For five it often is not, because seat administration, shared brand controls and invoicing are the features vendors most often move up a rung. The tier you can read about and the tier you would buy are not reliably the same tier, and that is worth establishing before a comparison starts rather than after it.

Hex draws the same line in a different place. Its cheapest option is a plan rather than a countdown, called Community and scoped in one sentence to hobbyists and small projects. Jasper's only free entry is a seven day trial attached to its Pro card. A tier that continues and a trial that expires are different products, and for a team evaluating something over a quarter the difference decides whether an evaluation is even possible.

What we will not do with these four figures

The obvious move is to multiply by five and publish the total. We do not, and the reason is not caution. Three of the four figures carry a condition that survives multiplication badly: Hex has a second meter, Notion AI has a rung that has to be the right one, and Jasper has a tier the page does not price at all.

Descript carries a condition too, and it is a different kind. Its Hobbyist card prints two per person figures and a toggle decides which one you are quoting, so what is at stake there is which number you multiply rather than whether multiplying means anything. That distinction is the useful one. Three of these four pages hand you a figure that is not the whole answer, and each of the three does say so somewhere other than beside the figure: Hex in a second table further down, Notion on the cards below the one you are reading, Jasper on the plan it declines to price. None of them is hiding anything. They are just not answering the question in the place you asked it.

So treat the four as the start of a quote rather than as one. What these profiles let you do is check the unit before you compare, which is the step that goes wrong silently. What they cannot do is tell you the year.

The same problem one layer down, in what an allowance is measured in rather than what a seat is, is in why credits, minutes and clips do not convert. The three things a pricing card can do to you before you budget from it are in how to read an AI pricing page.