Dust Wave · signals report · 24 September 2026
A deep pass over 13,610 posts from 31 Latter-day Saint accounts, going back to 2014. Four independent analyses, each scoring every post against its own author’s baseline so a 128,000-view account and a 900-view account can be compared honestly.
Two of the four disagreed on the biggest question. That disagreement is the most useful thing in this report, and it is resolved below rather than averaged away.
1 · the link question
Rise Zion’s reach collapse is real and the recovery is real. The part worth getting right is what to do about it, and that turns on eight months.
| Month | Posts with links | Median views |
|---|---|---|
| October 2025 | 100% | 38 |
| November 2025 | 61% | 75 |
| December 2025 | 2% | 30 |
| January–July 2026 | 0–15% | 21–46 |
| August 2026 | 3% | 232 |
| September 2026 | 0% | 2,361 |
The linking stopped in December 2025. Reach did not move for eight months — it sat near 30 median views across 393 link-free posts. It lifted in August 2026, and lifted every kind of post at once.
That is how a platform penalty behaves: triggered by a behaviour, then persisting long after the behaviour changes, until it is reviewed or expires. The restriction was on the account, not on each post that happened to contain a link.
Two checks were run before trusting this. First, whether old posts had been edited to remove links — they were not: October is still 100% linked and November still 61%, so nothing was scrubbed (and X does not permit editing a post after about thirty minutes anyway). Second, whether links suppress reach generally — across the other 30 accounts they do not: 0.97× with a link against 0.96× without, and of the seven accounts with enough of both to compare, four do better with links.
What to do: keep links out of the primary post — the trigger theory is sound and the cost of caution is nearly zero. But do not expect removing them to restore reach on its own; on this evidence that takes months, or a review. Put the URL in the first reply and judge the account on the trend, not the next post.
2 · the strongest finding nobody asked for
Sunday posts beat their author’s own normal by a wide margin, and it holds across 29 of 31 accounts.
| When | vs own normal |
|---|---|
| Sunday | 1.29× |
| Monday–Saturday | 0.93–1.02× |
| Sunday 6–11am MT | 1.16× |
| Sunday 5–10pm MT | 0.69× |
It survives the hardest test: remove the two accounts that make up a third of the whole corpus and Sunday still reads 1.10× against a weekday 0.73×. No single account is driving it.
What to do: put the week’s best work out on Sunday morning, before noon Mountain. This audience is in the pews and on their phones on the Sabbath.
3 · format
The most universally supported finding in the study, present in all 30 accounts that could be tested.
Video 1.14× · One image 1.08× · Text only 0.70×
A text-only post costs roughly a third of its own reach. Not against the field — against what that same account normally gets.
What to do: attach an image or a clip every single time. If there is nothing to attach, make something.
4 · the split nobody expects
Some language buys impressions. Different language deepens the people already watching. They barely overlap.
| Language | Views | Engagement rate |
|---|---|---|
| Conflict & controversy | +36% | flat |
| News & urgency | +30% | −15% |
| Devotional & testimony | flat | +14% |
| Naming an apostle | flat | +19% |
News framing buys impressions and costs loyalty. Devotional framing does the reverse. Naming a specific leader — “President Nelson” rather than “the prophet” — deepens response without widening it.
One more, across 22 accounts: quoting scripture in quotation marks runs +38% on views, while merely citing the reference is flat.
What to do: decide which number you are being judged on before choosing the register. They are not the same job.
5 · rise zion, specifically
Retire the “Watch…” opener. It is 157 of your 556 posts, and it runs at roughly half your own baseline. It is the single clearest self-inflicted drag in your account.
Keep the capitalised news headline — it is working, for you. Your caps posts carry a median of 1,668 views against 33.5 for your others. But be clear-eyed: across the rest of the network the same tactic is worth almost nothing (1.08×). It pays because you committed to the whole register, not because capital letters are magic.
Post once, well. One post a day reads 0.91×; two or more drops each to 0.70×. Posting again within an hour of yourself is the worst habit in the data at 0.62× — and going quiet for three days costs nothing at all (1.02×). Save them; do not rapid-fire them.
6 · what we could not show
X did not record view counts before 15 December 2022. Not low — absent. Any multi-year chart of views will show a fake collapse and a fake recovery. Every view-based number here starts in 2023; findings needing longer history use likes, which go back further.
Bookmarks are unusable. Zero on all 13,634 rows.
No quote-posts and no multi-image posts exist in this data, so neither could be tested — a limit of the collector, not a finding about the platform.
Velocity cannot be measured yet. Only 24 posts have been measured twice, 29 minutes apart. “What is climbing right now” needs the collector to run repeatedly over the same posts.
Things that looked true and were not: overnight posting being fatal (one account supplied up to 88% of those hours); posting nine times a day being ruinous (one account was 60% of that group); a full-caps headline being a general lever (4.5× collapses to 1.08× without Rise Zion); em dashes, quotation marks, hashtag counts and question openings — all flat.
method
Every post is scored against its own author’s median over a trailing window, never against the field — account size otherwise swamps everything. Medians throughout, never averages, so one viral post cannot carry a claim. Denominators are floored so an account with a near-zero baseline cannot manufacture a “600× breakout”.
Every finding reports how many posts and how many distinct accounts support it. Anything carried by a single account is named as that account’s behaviour, not as a network pattern — which is how four of the early findings were caught and withdrawn before reaching this page.
Corpus: 13,610 posts, 31 accounts, 2014–2026. Four independent analyses, reconciled; where two disagreed, the one controlling for account composition was kept.