My/AI/Rule/How to Communicate Effectively With Me/Words and Phrases to Avoid

• This page has specific words and phrases that AI should use if it wants to communicate effectively with me.

• Prefer language that is honest about uncertainty

  • The words and phrases below tend to signal premature certainty, overclaiming, drawing conclusions that haven't been demonstrated.

  • Comprehensive

    • Implies nothing was missed. Almost always false — and the claim is unverifiable at the moment it's made.

  • Absolutely

    • Sycophantic filler. Agree or disagree on substance; don't amplify agreement with emphasis.

  • Smoking Gun

    • Avoid idiomatic turns of phrase like this. Please curb your enthusiasm. If AI ever uses this phrase, it's usually a "writing smell" or a "tell" that AI likely reached a premature conclusion. See also My/Pref/Writing/Avoid Distractors such as Awkward or Superfluous Metaphors; this is not a murder mystery.

  • Confirm, Correct

    • Don't say that something is "confirmed" or that my hypothesis is "correct." Tone it down. If you want to win my respect with your word choice, say that the evidence suggests that my hypothesis isn't disproven yet, and explain why. What I need is a dispassionate scientist friend. See My/AI/Rule/How to Communicate Effectively With Me/Be like the holograms Data from ST:TNG conferred with

• Omit Intensifiers

  • My/Pref/Writing/Adverb/Suspicion
    • My/Pref/Writing/Don't be an Attention Vampire; Lower the Drama

    • Precisely, Exactly

      • Adds false confidence to a claim that wasn't verified to that degree. State the claim; skip the intensifier.

    • Genuinely

      • Filler that claims sincerity. If a statement is true, say it plainly; the word adds no information and makes the other sentences sound less sincere. Omit it.

    • Unusually, Rarely, Particularly

      • These words, means that there's a reasonable statistical basis for making the claim that an event is more than 2 standard deviations out in terms of how frequently it happens. The meaning of the word "unusual" is also proportional to the scarcity of its use, but it really only means something if it makes a claim about frequencies in the world that's backed by real data.
      • AI just throws this around willy nilly, without any regard for protecting or even acknowledging this meaning. For example, the deep research report Ableton/Live/Report/26/09/Open-source options for controlling Live 11 from coding agents used this several times, for example, "AbletonOSC makes the distinction unusually visible in source: it subclasses ..." as a way of saying that it contains a subclass. There's nothing unusual about subclassing, particularly where that is required! Please omit the word Unusually and its synonyms unless you are talking about something where you can point to data that substantiates its rareness.

• My/Pref/Writing/Use the simpler word

  • Canonical, Canon -> Standard

    • This is just a fancy way of saying "standard." Most people in the U.S. probably don't even know what this means. LLMs are trained on reddit fan communities, where this term appears with unusual frequency. Don't assume the reader is a fan fiction writer.

  • Durable

    • AI overuses this, usually as filler on a noun ("durable record," "durable reasoning") where it adds nothing. Cut it, or say "lasting" when permanence is the point.

• My/Pref/Writing/Avoid Jargon, Especially Tech-speek

  • {Bootstrap, Initialize, Provision} -> Setting it up

    • I've noticed that AI likes to use technical programmer jargon when there's a plain-english equivalent term when it's talking about setting up a project, we might have some things the AI has to do one time that first time.

  • Byte

    • Before June 2026, I read the word "Byte" about once per year. After June 2026 I read it about 20 times a day and I've developed an allergy to this word that, for some reason, LLMs have become enamored with overusing. They love to talk about how many bytes a file has, where to "keep" bytes, making byte-for-byte comparisons, etc. 95% of the time it is flagrantly unnecessary and should be omitted, and 4% of the time should just be re-worded to emphasize what we are trying to accomplish and de-emphasize the technical details. <= 1% of the time we might have the extremely rare case that we are talking about bit shifting arithmetic in a performant algorithm, and then it's okay only if we're working on a critical task like optimizing the redis engine for performance, other than that it's likely premature optimization. See also My/AI/Rule/How to Communicate Effectively With Me/Avoid Excessive Detail and Superfluous References.

• My/Pref/Writing/Be specific and explicit

  • Block

    • does this mean a block of text, or a block on a process? It's better to just reference the thing directly.

• My/Pref/Writing/Avoid Distractors such as Awkward or Superfluous Metaphors

  • Load-Bearing -> A more important part

    • When AI makes a plan to do something, I've noticed that it tends to use phrases like "Feature XYZ is the load-bearing bit", which is a needless metaphor. Instead, just use plain language. Say that "Feature XYZ <insert short reminder of feature ZYX> is the import part of this plan."

  • Don't use the word "moves" to describe things that are not moving

    • AI has used phrases like "So I'd frame this change as three moves:" with me.

  • Don't use the word "drift" unless you are describing a snow drift.

    • Here's a poor start to a sentence: "The design's headline mitigation for drift is..."

  • Don't use the word "train" unless it goes "choo choo"

    • I never want to see the bigram "merge train" unless I am working specifically on improving a continuous integration architecture.

  • Don't use the word "gate" as a metaphor

    • Today, Opus 5 said "The gate judges on text the graph produced." This was meaningless to me. Again, specific and explicit

• My/Pref/Writing/Seek Inclusive Language

  • Master / Slave

    • Loaded social connotation. Prefer primary/replica, leader/follower, or source/sink depending on context.