Most landing pages are written on instinct: read it aloud, cut the adjectives, make the button say what happens next. That process produces pages that work and pages that do nothing, and it gives no way to tell which one you have before launch.
Code: MindFluence, MIT licensed.
The repair follows the same pattern. When a page underperforms, the fix is to rewrite the hero, add a testimonial, try another button colour. Sometimes the number moves, and nobody can say why, so the next page starts from zero.
MindFluence exists because “I know good copy when I read it” does not work as a method. The skill is one Markdown file a model loads as a system prompt, and it carries twenty cognitive biases, thirteen anti-patterns and a checklist that runs before anything is published.
The bias that was missing from the page
Run it in audit mode on a services page and the top item is rarely the headline. In one draft the third paragraph said we “help companies grow” and moved on.
The note was that the paragraph carried no cost of inaction. A reader who already wants to grow learns nothing, and a reader who is unsure has no reason to keep reading. The fix is one sentence naming what stays broken without the work: a checkout that loses mobile buyers, a release that slips every quarter.
That is loss aversion, and most teams use it by accident, on the days when the writer feels confident.
Twenty biases is a menu, not a checklist
Applying all twenty at once produces a page that reads like a carnival barker. The skill routes instead. Audience temperature, product type and platform go into a lookup table with 38 rows, and the table returns a stack of three to five biases that fit the situation.
Cold traffic on a landing page gets framing, anchoring and social proof. A skeptical reader who has already tried two competitors gets the backfire sequence, where you agree with their objection before answering it. A warm list that stopped opening emails gets sunk cost and in-group language instead of a discount.
The router makes the decision explicit instead of habitual: which psychological lever this reader needs right now.
Anti-patterns are the section that earns its place
Thirteen of them, each with a detection rule that reads like a search query rather than an opinion.
Fake scarcity: a countdown with no stated reason. Authority without proof: “studies show” and no study. Fear without an exit: three sentences of threat and no remedy in the same paragraph. Transactional reciprocity: a free checklist and a buy button in the same email.
Teams ship at least four of these on a normal launch, and each one works in the short term. A countdown does lift clicks for a week. Then the same subscriber ignores the next three emails, and the list slowly dies.
Provenance decides whether the page survives review
The gate that matters most is the boring one. Every number, name and quote has to trace to the input, to a named source, or to an explicit placeholder. When there is no real case, the model writes a labelled illustrative example and says so.
An invented customer reads beautifully, survives review, and then a prospect asks for a reference call. One fabricated statistic can cost a year of trust, and nobody notices it in the CMS.
If the source data is thin, the honest page is the one that says so. A small real example beats a large imaginary one, and the skill forces that choice before publishing instead of after.
How the file sits in the repository
The file lives next to the page templates. When a page underperforms, audit mode names the biases and the reasoning, and the argument happens with labels instead of mood. When the page is written from scratch, quick mode runs and the provenance gate has to pass.
The writing still sounds like the writer. A skill that flattens every voice into generic marketing prose is useless to anyone with a byline, which is what the seven tone presets are for: pick the cadence, keep the vocabulary.
For the process side of shipping a page, see how a missing spec burns a project. If the copy is fine and the problem is that nobody outside the team uses the same words, that is a naming problem, and naming in code covers the same habit.
Background: cognitive bias on Wikipedia, and the skill itself is MIT licensed on GitHub.