PUBLISHED SEPTEMBER 2026 · THREE PRODUCTS, ONE BUYER

We gave one ghost user the same job in Linear, Notion AI and Airtable: work out what it costs and whether to adopt it, before signing up.

Same person, same job, same screen, three products, driven one after another on one morning. Every one of them answered every question the buyer came with, so there is no loser here — and that is the finding, not a disappointment. What separates the three is where each keeps its answers: to a reader working out what Airtable costs, 5 of the 99 controls on its pricing page lead to a separate site of Airtable's own. Linear's page does that on none of its 64. Notion's opens no new tab at all. Read a second way four weeks earlier, those same pages put a very different distance between the first price they show and the nearest place they say what that price is per: 16 characters on Linear, 19 characters on Airtable, 315 characters on Notion AI.

HOW THESE THREE ARE COMPARABLE

Three runs are one measurement or they are three essays. These were driven one after another on the same day, by the same buyer, against the same job, on the same screen, decided by the same model — and the tool that assembles this page refuses the group if any one of those differs.

The job: I need to work out whether this could replace the spreadsheet my team of five runs on, and what switching would actually cost me, before I hand over my details.
The buyer: Wisp-11 — Engineering lead at a forty-person B2B software company. Five direct reports, still writes code most weeks, has run Jira, Trello and GitHub Projects and none of these three. Evaluating on a morning with a standup in twenty minutes and two other tools open in tabs from the same shortlist. Not a real person's session →
Screen: desktop · 1440px
Decided by: claude-sonnet-5, declared before the first run and identical across all three. Harness protocol 1.10.0.
Runs: Linear — completed, run on 26 September 2026Notion AI — completed, run on 26 September 2026Airtable — completed, run on 26 September 2026

WHAT THE BUYER CAME FOR

Three questions. All three products answered all three.

1
I can say in one sentence what this would do for my team that our spreadsheet does not.
All three answered it.
2
I know what it would cost us a month for five people, without having to ask anyone.
All three answered it.
3
I know what I would have to hand over to get started, and whether I could get back out.
All three answered it.

A job all three products pass cannot separate them, and we would rather say that than build a ranking out of a tie. This job was chosen because it needs nobody's permission — everything above is answerable from pages anyone can read, with no account and no sales call — and three well-resourced products all clear that bar.

WHERE EACH ONE KEEPS THE ANSWERS

Counted on the pricing pages themselves, with nobody reading them.

The third question — what do I hand over, and can I get back out — is the one whose answer moves around. So we opened each product's pricing page and counted the controls that send a reader somewhere else in a new tab, and recorded where each one goes. No buyer, no model and no stopwatch are involved in the numbers below; they are a fact about three pages on 24 October 2026, and anyone can check them against the same pages.

Pricing page	Controls on the page	Open a new tab	…to its own material on a separate site	…to somebody else's site
Airtableairtable.com/pricing	99	17	5community.airtable.com, support.airtable.com	7
Linearlinear.app/pricing	64	6	0	4
Notion AIwww.notion.com/pricing	73	0	0	0

THE NUMBER WE DID NOT PUBLISH

The obvious figure is the middle column — 17 new tabs on Airtable's page against Linear's 6 and Notion's 0 — and it is wrong for this question. Seven of Airtable's seventeen are the footer's Facebook, LinkedIn, Twitter, Instagram, YouTube, App Store and Google Play icons, and four of Linear's six are the same kind of thing. Nobody working out what a product costs opens those, so a comparison built on that column measures footer decoration and bills it to a named company. The column we publish is the one that answers the buyer's third question: where a product keeps its own material when it is not on the page you are reading. It is not the widest of the four columns either: the total spreads 17, the third-party column 7, this one 5 and the same-site column 5. What it is, and no other column here is, is the one where two of the three products come out identical and the third does not — which is the only shape a sentence about a majority and a minority can be built on.

THE COUNTS WE REPLACED

These numbers are not the ones this page carried when it went up, and they are not the ones it carried after our first re-count either. Our first count read these pages without waiting to see whether they had finished arriving, and came in short where they had not — a defect in our instrument, not a change in anybody's product. We re-counted on a reading that waits until a page stops changing. Then that instrument changed again, to wait for a page to go quiet twice over rather than once, and the count we published before this one was taken on the older of the two — which makes it the wrong thing to measure a later reading against, whatever either reading says about the page. So the figures above are a fresh reading, taken on the instrument that every reading we compare this page with will be taken on. Each time, we replaced the old figures rather than dating them, because a number we invite you to check has to be the number you would get.

THE COUNT THAT ANSWERS TWICE

A figure above moved between the reading this page used to carry and this one: the number of controls on Notion AI's pricing page. We are not calling that an edit, and the reason is a measurement rather than good manners. We have read that page over and over inside a single sitting and had both numbers back, with nothing altered on our side between the reads — so that page serves both, and the figure we published before was one of the two it serves. A page that answers differently to the same question twice cannot be shown, by a difference between two answers, to have been worked on at all. Both of our figures are dated, both are what that page handed us when we asked, and neither is a correction of the other. The rule we hold ourselves to is in the open, and this page is held to it by a gate rather than by us remembering: nothing we publish may say whose doing a movement in a number was, in a column one of our own readings has seen hand back two different values.

AND FROM THE OTHER SIDE

The count above was taken off the pages with nobody in them. These are the sites each buyer actually stood on during its run — a separate record, made independently and four weeks apart, that agrees with it.

Linearlinear.app
Notion AIapp.notion.com · www.notion.com
Airtableairtable.com · support.airtable.com · www.airtable.com

AND HOW FAR THE PRICE IS FROM WHAT IT IS PER

The same three pricing pages, read a second way.

The buyer's second question — what would this cost us a month for five people — was answered everywhere. But a price on its own is not the answer: the same number means one thing charged by the month and another charged for a year up front. So we read the rendered text of the same pages again on 29 September 2026 and measured one distance: from the first paid price the page shows to the nearest place that page names a billing period. No buyer, no model and no stopwatch here either.

Pricing page	First paid price → nearest billing period	Closest any paid price gets	Paid prices within two lines of one	Periods the page names
Linearlinear.app/pricing	16 chars	16 chars	2 of 2	annual
Airtableairtable.com/pricing	19 chars	1 char	4 of 11	annual + monthly
Notion AIwww.notion.com/pricing	315 chars	315 chars	0 of 5	annual + monthly

WHY WE TRUST THIS ONE

There are three ways of asking the question in the table above — the first paid price, the closest paid price, and how many of a page's prices sit next to a phrase naming a period — and the widest and narrowest of the first-price distances are apart by 299 characters. The tool that assembles this page refuses the whole group unless all three readings single out the same page, because the reading was chosen after these numbers were on screen and a finding that depends on which reading we picked is a finding about us. It did not have to agree, and it did:

The odd one out on all three readings — Notion AI

What this is not: a claim that any of these pages fails to say what its prices are per. Each one says so, and the column on the right shows which periods each names. What is measured here is how far a reader travels between the two, on one rendering of one page on one day — and anybody can open the same page and count it again.

WHAT THIS PAGE LEAVES OUT

Four things we measured and are not publishing.

How long each run took.: The three came in at 286, 265 and 404 seconds, and that 139-second spread is ours rather than theirs. Under the protocol these ran on, a link that opens a new tab costs the buyer a whole extra turn — and Airtable's pricing page has 17 of those against Linear's 6 and Notion's 0, so our instrument charges a product in proportion to the very structure this page is about. Airtable's run spent 4 of its turns on round trips that produced no step, Linear's 1 and Notion's none, which is the same ordering as the tab counts and not a coincidence. We have decomposed those seconds out in our own notes and the three products land within a few seconds of each other; that decomposition is not published here, because it is not yet something a record on disk can be checked against, and a number we cannot re-derive is one we do not print.
What the buyer said about each product.: Two of the three moments this group recorded as the product's worst are ours. Notion AI's buyer described a page whose address had changed while the screen had not — that is our readiness check, not their software. Airtable's rests on a cookie banner whose buttons never appeared in the list of things we hand the buyer to act on. Publishing the third product's quote on its own would exempt exactly the two whose findings we caused, so none of the three appears here until the instrument is fixed.
Which of these products shows you its cheaper rate first.: The rate a pricing page shows you before you touch anything is not always the one you would pay monthly, so we pressed every control whose label names a billing period — each on its own fresh load of the page — and watched the prices. On Linear's page and on Notion's the prices moved, and each arrived on the lower of its two rates. On Airtable's we recognised no control naming a period at all, which is a limit of what our tool can recognise rather than a fact about Airtable: the honest reading there is unknown, not absent. A sentence of the form “most of them do this” is false while one of them is unknown, so there is not one here. What the press did settle is the distance above: Notion AI's reading is identical in both states, so nothing in this section depends on which state a reader happens to arrive in.
A winner.: The job was chosen because it needs nobody's permission — everything a buyer can learn from public pages — and three well-resourced products all clear that bar. We are not going to manufacture a ranking out of a tie; the tool that assembles this page refuses to write a cross-company sentence when the group does not split, and on the criteria above it did not.

THE FINDING

Nobody failed. All three told a buyer what they do, what they cost for five people, and what they want at the door, from pages anybody can read without an account — which is worth saying plainly, because a job all three products pass is a job that cannot separate them, and we went looking for a difference we did not find. What the three runs leave behind is not a score but a map of where the third answer lives. The buyer that read Notion never left Notion's own domain, standing on www.notion.com and app.notion.com and nowhere else. The buyer that read Linear never left linear.app at all. The buyer that read Airtable ended up on support.airtable.com — one of the two sites of Airtable's own that 5 of the 99 controls on its pricing page point at, 4 of them to support.airtable.com and 1 to community.airtable.com, counted separately, four weeks later, with nobody reading the page at all. Read a second way on the same pages, the same question has a different shape: where each product keeps the fact that a price is charged by the month or by the year. On Linear's page and on Airtable's, that fact sits beside the price a reader meets first. On Notion AI's, the words are there and the first price arrives a long way after them. Read a third way, by pressing the controls those pages offer rather than only reading them, the same pages say something else again. Linear's pricing page carries three phrases naming a billing period and every one of them names the year; the control that switches its rates is itself labelled “Billed yearly”. We pressed it and the paid prices moved from $10 and $16 to $12 and $18, and the words did not change. So one of the prices a reader can see is described with the other’s billing period, and the page does not say which — we are not saying which either, because nothing we measured settles it. Notion AI's page is the opposite shape, and the only words naming a period anywhere on it are that switch's own two controls, so there is no single period for a price to be labelled with. On Airtable's page we recognised no control naming a period, so this reading has no Airtable in it — unknown, not absent.

← Every run we have published, one product at a time