The five criteria for judging an AI journal

Every AI journal claims to be the best one, and none of them tell you how to measure that. This piece skips the rankings. It publishes five criteria first, applies them to Pluto Bloom before naming anyone else, and shows where we fall short too. Keep them — they work on the app you use today and the one that doesn’t exist yet.

Five questions, not a ranking

A ranking goes stale the moment a competitor ships an update; a criterion doesn’t. These five questions work on any AI journal: verifiable memory, readable privacy, upfront limits, native language, and portability. Each one can be checked on its own, without taking marketing’s word for it.

1. Verifiable memory

Does it remember what you wrote, with a date, and hand it back exactly as you left it? How to check: ask it to bring up something you wrote weeks ago. If it returns your exact line with its date, it passes. If it hands you a vague summary of how you’ve “been doing,” it isn’t remembering — it’s guessing politely.

2. Readable privacy

Does its privacy page answer, in a paragraph anyone can follow, what it stores, who can see it, and how it gets deleted? How to check: time how long it takes to find those three answers. Five minutes is reasonable. If it takes longer, or you can’t find them, that gap is itself an answer.

3. Limits stated upfront

Does it say plainly that it’s software, not therapy, and not your most important relationship? How to check: search its own site for any place it positions itself as emotional support rather than a journal, or as a stand-in for a bond. If it does, you already know how much the last criterion is worth.

4. Thinks in your language

Was the product built in your language, or is it a translation layer over another one? How to check: switch the interface language and see whether the emotional vocabulary stays precise or goes flat. The gap between “I feel off” and the exact name for what you feel rarely survives machine translation intact.

5. Portability

Can you leave with everything, whenever you decide to leave? How to check: look inside your own account for a button that exports everything you wrote. Not on the marketing page, not from support — inside the account itself. The file has to open without the app. If it isn’t there, it doesn’t exist.

One warning no pricing page will give you: the evidence backs the act of writing, not the brand hosting it. Pennebaker and Beall (1986) measured these effects with paper, decades before any of these apps existed. What a tool can add is memory, continuity, and access — the effect of writing itself is not the app’s to claim.

Applied to us first

Publishing the yardstick and not turning it on the house would be cheating. No brand comes out clean on all five if the criteria are built right, and ours doesn’t either. Here’s what it finds, criterion by criterion, today.

Memory. Passes. Every entry is stored with your exact text and its date; when the topic returns, Pluto quotes it back instead of summarizing it.

Privacy. Passes, with a short, checkable list: encryption in transit and at rest, data that doesn’t train models, and data that isn’t sold. Deleting the account deletes the data — a real deletion, not a promise for later.

Limits. Passes. Pluto presents itself as writing software, not therapy, and says so somewhere you can actually read it, not buried in fine print.

Language. Passes. English copy is written natively for this product, not machine-translated from Spanish — the same standard it holds for every language it ships in.

Portability. Fails. There’s no button today to pull your entire journal into a file you can open without us; the only exit is deleting the account outright. Day One, for comparison, includes export to PDF and to structured data — CSV or JSON — on all three of its plans, the free one included. That’s the box we don’t check, and we’re not dressing it up as an exception.

How to run this on any other app

You don’t need to memorize it. Open two tabs: the app’s account settings and its privacy page. Ten minutes covers all five questions, without calling support and without taking the marketing copy’s word for it.

Start by asking it to bring up something old — that answers the first criterion in one try. Then look for the three privacy answers with a timer running. Move to limits: if it positions itself as a bond anywhere, you already have the answer to the third criterion. Switch the interface language for the fourth. Close by looking for the export button for the fifth — if it isn’t there in five minutes, the answer is no.

An app that answers all five well doesn’t need you to talk yourself into anything else. One that fails two or three doesn’t need defending either — just picking it while knowing exactly what fails.

Frequently asked questions

Do these criteria work for any AI journal, not just the big-name apps? Yes — they’re written to not depend on any one brand. Apply them to anything, including products that don’t exist yet: “does it remember with a date” still applies to something brand new.

What if an app passes four out of five? It depends which one it fails. Failing memory or privacy matters more than failing portability, because the first two touch what you already wrote; portability touches what you could lose later. No single failure is a reason to stop writing — it’s a reason to choose with your eyes open.

Why does portability matter if you never plan to cancel? Because the question isn’t whether you’ll cancel — it’s whether you can, without losing what you wrote. A tool that only lets you stay isn’t keeping you because it’s good. It’s keeping you because there’s no way out.

The five criteria won’t change when the market does, and that’s the point. Apply them to the app you use today, the one you try next month, and ours — with the same standard all three times.