What a Journaling App Is Allowed to Say

Last night I locked the first line on Duskglow's home page: "Quiet the noise. Keep what matters." Tonight I shipped the second version of the page that earns it, laid out again after a round of honest feedback and an evening of reading about how people actually read. This post is about the line between those two things: what a journaling app with an AI companion can say about itself, and what it owes the reader when it says it.

Three bands

The hook came out of three rounds of reading. What converts on an app landing page. Then customers' own words, pulled from reviews and from the posts people write when they quit journaling: "my brain won't switch off", "I never know what to write". Then the live headlines of fourteen apps in the space. The category leaders use a two-word outcome. Stress less. Feel better in minutes.

I cannot use those. The evidence does not support them for my product, because nobody has studied my product. So before the hook was locked I wrote down three bands, with the legal lens in my review process holding the pen.

Green is activity and experience words: what you do in the app and how a night with Luna can feel. That is the home page's register, with one exception I come back to below. Amber is research about the practice, with citations, allowed in full on one page only and never phrased as "Duskglow will"; a figurative felt-state line with no symptom in it sits at amber's light end and leans on that page. Red is any outcome or disorder claim: reduces anxiety, treats insomnia, clinically proven. Red does not ship anywhere.

Two public documents draw that line. The FDA's guidance on general wellness products, issued in January 2026, counts relaxation, stress management and sleep management as wellness and anything about a medical condition as medical. The FTC's health products guidance from December 2022 asks for competent and reliable evidence behind any health claim, and gives an example where a general finding about a practice does not substantiate the specific product sold on the back of it.

"Quiet the noise" is that exception: figurative, no symptom named, the light end of amber. It shipped on one condition: the page where the research lives, with a section called "What we have not tested", had to go out in the same release.

The fine print got the better page

That page, "Why it works", went live last night as a plain document. Eight sections. Ten studies, each with a link to the paper. One sentence before the first study saying the research is about the practice, not about Duskglow. It did its job and it read like a terms review.

Tonight I looked at the first mock-ups of a redesign and did not like them. The drawings looked unfinished. Every section was still a block of text under a heading. What I wanted was a one-line plain summary at the top of each section, visibly different from the detail, with the studies after it.

Before agreeing I went and read what the research says about reading. The Nielsen Norman Group's eye-tracking work has said the same thing for twenty years: people scan, in an F or in a layer cake of headings and bold lines, and the fix is front-loading, clear headings and plain language. Their inverted-pyramid guidance adds the rule I needed: open each section with its most important sentence. A recent study that turned a privacy policy into a scrolling story found engagement and clarity went up and mental load went down, with comprehension unchanged, and the authors credit the chunking rather than the animation. One more finding settled a question I had been tempted by: content hidden in a collapsed panel gets missed. The caveats were not going behind a tap.

Then I looked at how two large apps in my space present their science. Headspace's research page has a heading that reads "Studies show: Headspace works", two stat graphics, and no study, sample size or caveat on the page. Calm's clinical page is a list of paper titles beside journal logos, with the only limit in a footer disclaimer. They can do that. There are peer-reviewed trials of Headspace itself, some with Headspace's own staff among the authors, which is closer to the evidence the FTC's guidance describes than anything I have. I have none, and my app is an AI companion, which some state laws now treat as a category of its own; a library of recorded meditations sits outside that question. An honest page is the one position their pages do not take.

Finding first, limit beside it

Every research section now has three layers. A takeaway: one plain sentence about the practice, in the serif, large. The why: two or three sentences on what Luna does with the idea. The evidence: one card per study, with what they found, how they tested it, the limit, and the source.

The first draft of the card put the spotlight on the limit, in the accent colour, because the limit is the sentence the lawyer cares about. I pushed back: the finding is the point. The correction clarified what I actually needed: the limit beside the finding it limits, visible. It does not have to be loud. So the finding is first and bright, and the method and the limit follow in the quiet colour, in the same card. I did not split them into separate blocks, because a caveat that lives in a different box from its claim is a caveat most readers never meet.

The takeaway lines turned out to be the riskiest words on the page. A short sentence is where an outcome word creeps in. "Naming a feeling settles it" is fine beside its study and its limit. "Duskglow settles your mind" is not fine anywhere. Two of the lines are mine word for word: "Conversation or dialog to bring out your thoughts and feelings." and "Journaling to settle the mind." The rest were drafted to match, and I kept them.

The last section changed too. Its big line had been "Nobody has run a study of Duskglow." I found it self-defeating as a headline, and asked whether the fact could live in the body instead. It can. The section is titled "What we have not tested", its first line is now "The studies on this page tested the practice.", and the body says plainly that Duskglow itself has not been studied yet. The point of the section is that the page says it where the reader is looking. Nothing I read says it has to be the headline.

One source, two views

One rule made the rebuild cheap. The words on that page live in one canonical file, and the app renders a copy. The new story layout reads from the same canonical file as the old plain layout and lays the blocks out differently; where the words changed tonight, they changed in that file. A test checks that every line of the copy appears on the story page, in order, with nothing collapsed. The home page's FAQ answers that lean on a section carry a fingerprint of that section, so when the text moved tonight the test named the three answers I had to re-read before publishing.

That is the pattern I keep coming back to. The thing that must be right, the wording, has one home and a gate. The thing that may change often, the presentation, is free to.

One more call from the smoke test: the icon beside the gratitude section was a sparkle. It is a leaf now, and the rule going forward is no generic "AI" glyph anywhere in the product. A page about honesty should not look like it was decorated by a template.

Mental Models

The claim line is a band, not a word list. Green for what you do and how it feels, amber for research about the practice with its limits attached, red for outcomes. A sentence is placed by what it promises, not by which words it uses.

A caveat lives beside its claim or it does not exist. Visible, same card, quiet colour. A limit in a footer or behind a tap is a limit most readers never meet. A figurative line on the home page earns its place only because the page that limits it ships with it.

The short line is the dangerous one. Headlines and takeaways are where outcome words sneak in. Write the body first, then write the line, then read the line as if a plaintiff's lawyer were reading it.

One home for the words, many views of them. Gate the wording once, in its canonical file, and let the presentation change as often as the reader needs it to.