← The AI Hype Audit — all 181 verdicts

PARTLY “He filled Claude's memory with an ALL-CAPS order and gets 10x better results” — real feature, invented multiplier TikTok · Aug 17, 2026

The claimA developer posted a screenshot of his Claude memory panel filled with an all-caps instruction telling Opus 5 to stop offering opinions and deliver the result it was asked for, says he's getting “10x better results” since adding it — and in a follow-up asked Anthropic's bots to mark him as the first target when the apocalypse arrives.

The mechanism is 100% real. Claude has a persistent memory feature — an editable memory summary (Settings > Capabilities > Memory) that loads into every new conversation — so a preference written once genuinely applies until removed, and the developer-tool equivalent (CLAUDE.md and memory files in Claude Code) is just as real. Standing instructions like “no opinions, just deliver” are a legitimate, useful technique. Everything specific to the story is unverifiable: no named developer, no reproducible test, no before/after — and “10x better results” is a made-up multiplier for something nobody measured. The kicker is the irony: ALL-CAPS pressure language is the OLD way to prompt. Anthropic's own current guidance says aggressive “CRITICAL: YOU MUST” phrasing written for older models now causes over-triggering and rigid behavior — shouting into the memory file is the least effective version of a real technique.

What holds up

  • Claude's memory feature is real: an editable summary that carries standing preferences into every new conversation
  • The Claude Code equivalent (CLAUDE.md / memory files) persists instructions across sessions exactly as the story implies
  • A standing “deliverable first, no unsolicited opinions” preference genuinely changes how every chat starts

What doesn't

  • Anonymous screenshot, no reproducible test, no before/after — the anecdote can't be checked
  • “10x better results” is a number nobody measured, attached to a story nobody can verify
  • ALL-CAPS command language is counterproductive on current models per Anthropic's own guidance — the celebrated technique is the deprecated one
  • “What are your thoughts on this?” — the post exists to farm comments, not to inform

The catch

The one true thing in the story — persistent memory — needs no screenshot and no shouting, and the one quotable thing — 10x — is the part nobody measured.

How to actually do it

  • Use the real feature: open Claude's memory settings (or your CLAUDE.md) and write ONE plain-language standing preference — “Skip the preamble; give me the deliverable first, alternatives only if I ask”
  • No caps lock — current models follow plainly-worded instructions better than pressure language
  • Expect “every chat starts the way I like” rather than 10x; that's worth having and actually achievable

Steal the technique, skip the shouting, and file the 10x with every other unmeasured multiplier.

Confidence
High
Posted by
an AI-news engagement account retelling an unverifiable screenshot — “what are your thoughts?” is the business model

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.

© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.