~/codigoagente
AI agents

Marketing psychology for your AI agent: the skill that teaches it how buyers think

At the end of the article you'll find the install command, a ready-to-paste prompt and a .md to download.

marketing-psychology is a free skill from the coreyhaines31/marketingskills collection (MIT) that loads 50+ mental models of marketing psychology into your agent: reciprocity, peak-end, loss aversion, anchoring, social proof, paradox of choice, framing, zero-price effect. It is not a magic prompt: it is a structured reference that Claude Code, Codex, Cursor and any skill-aware agent can consult when you ask "look at this landing" or "design the pricing block". Combined with the sister skills in the same repo (cro, copywriting, copy-editing, content-strategy) your agent stops writing generic copy and starts moving a concrete metric.

In this three-part guide you will see the skill in action, with examples from this same blog. Part 1: what it is, how to install it and the eight models from the skill that explain how your buyer decides. Part 2 covers persuasion and copy and part 3 covers pricing and choice architecture.

Three fronts of the skill: think the problem well, understand the buyer and persuade without tricking · codigoagente

What marketing-psychology is and who makes it

  • Repository: github.com/coreyhaines31/marketingskills, published by Corey Haines (founder of Everyone Hates Marketers, ex-growth at Baremetrics).
  • License: MIT. Free for commercial and closed projects, no attribution required in your product, though keeping the LICENSE notice is good manners.
  • Version analyzed: 2.0.3. The SKILL.md is ~12 KB long and groups models into six blocks: thinking, buyers, persuasion, pricing, design and growth.
  • Which agents it works with: Claude Code (native), Codex CLI, Cursor, OpenCode, Gemini CLI and any Agent Skills-compatible runner. Total installs of the collection sit around 228,000, as we covered in the find-skills post.

The skill does not generate text on its own. What it does is structure your agent's reasoning: when you ask for a marketing decision (a headline, a pricing block, an email), it consults the "Quick Reference" table in SKILL.md, identifies the two or three models that apply and uses them instead of parroting LinkedIn clichés.

Install the skill in 30 seconds

From Claude Code, with the skills plugin active:

terminalsh
npx skills add coreyhaines31/marketingskills@marketing-psychology

If you want the whole collection (they are meant to be used together), install only the root and Claude will load each one when relevant:

terminalsh
npx skills add coreyhaines31/marketingskills

In codigoagente we run it alongside cro, copywriting and copy-editing. The repo's .claude/product-marketing.md acts as context: the first instruction of the SKILL.md is literally to read that file before applying any model, so your agent knows your product, reader and voice before trying to persuade.

The model block that actually moves sales

The SKILL.md lists 50+ models. If you could only teach your agent eight of them today, these are the ones that keep showing up in the blog's decisions.

1. Reciprocity: give before you ask

People return favors. If your agent writes an article that solves the reader's problem (tested commands, real errors, license clearly explained), the reader feels in debt when the sponsor block or the share invite shows up.

Real example: on this blog the Safe Build Labs sponsor appears as a quiet strip around 40 % into the article, never at the start. By then the reader has seen commands that work, exact versions and the why behind each decision. That order is on purpose: reciprocity kicks in once you have given something useful first, not when you ask on arrival.

2. Peak-end rule: they remember the peak and the ending

Daniel Kahneman showed that we remember an experience by its most intense moment and its ending, not by its average. A post with a weak ending is forgotten even if the middle is brilliant.

Real example: the "Use it now" block (resource with the ready-to-paste prompt) was added at the end of actionable posts in PR #92 of this repository. The peak is the prompt you can copy and use in 30 seconds; the ending is "It helps us grow", where we invite sharing. If your agent is writing an article, ask it to design the peak and the ending first, the middle last.

3. Loss aversion: losing hurts twice as much

Prospect theory says losing 100 hurts more than gaining 100 pleases. That is why negative headlines ("don't install this skill without reading this") beat positive ones when the risk is real.

Real example: when we covered the SkillSpector skill we titled it "Scan skills before installing them: one was leaking credentials". The headline does not promise an abstract improvement: it names the concrete loss (leaked credentials), which is exactly what prospect theory predicts will catch a technical reader's eye.

4. Zero-price effect: "free" is not a low price

The jump from 1 USD to 0 USD is psychologically bigger than the jump from 2 USD to 1 USD. That is why a free tier has a disproportionate pull.

Real example: the ebook "Claude Code de cero a producción" sells on Hotmart at 19.99 USD. On the blog we offer the free prompt from the "Use it now" block first: the reader copies it, tries it, and the ebook stops being an expense and becomes the natural next step. Free is not just "cheaper": it lands in a different mental category, which is exactly what Dan Ariely's zero-price effect describes.

5. Social proof: numbers convince

People copy what others do. If a repo has 12,000 stars and 2,100 forks, the reader assumes it is not junk and lowers the technical guard.

Real example: every repo card at the end of a post shows the license and the GitHub link. In the find-skills post we list each skill's install count ("228,000 installs" next to marketingskills@seo-audit) instead of calling it "the most popular". Concrete social data, verifiable in the repo's plugin.json, not adjectives.

6. Paradox of choice: fewer options, more decisions

Seven pricing plans paralyze. Three move traffic. Sheena Iyengar measured it in her jam-tasting experiment: 24 flavors converted at 3 %; 6 flavors converted at 30 %.

Real example: on the kit page we chose to show a single recommended product with the rest as "also available" in a small list, instead of three cards in parallel. The page asks for one decision, not three.

7. Anchoring: the first number wins

The first number the reader sees conditions every number after it. If you open with "AI agents cost 500 USD/month", 19 USD looks cheap.

Real example: in the slopus post we anchor the cost of a self-made model with "training an LLM from scratch costs 500k to 10M USD". The 200 USD/month Max plan goes from "expensive" to "a steal". More on this in part 3.

8. Default effect: the preselected option wins

Pre-selected choices get accepted 70-80 % of the time. The European organ donation study showed it: opt-out countries have 90 % donors, opt-in countries 15 %. Same people, different default.

Real example: in the "Use it now" block the first tab is "New project" with the safest prompt for someone starting out. The second, "Advanced project", is more aggressive. Readers almost always click the first, which is the recommended path anyway.

How to tell your agent to use these models

The trick is not to list all eight in every prompt. It is to load the skill, let it read .claude/product-marketing.md and ask it a concrete marketing question. For example:

snippettext
Load marketing-psychology. We have a post about skill X. At the end we want the
reader to buy the ebook (19.99 USD). Tell me the three mental models that fit
best and a sample sentence for each.

Your agent will return something like:

  • Reciprocity: "I gave you the full prompt for free; if that saved you an afternoon, the ebook saves you a week".
  • Peak-end: put the ebook link right after the recurso.en.md, not at the start.
  • Zero-price effect: lead with the free prompt, then the ebook price.

In part 2 of the guide we turn models into copy that does not read like a neighborhood salesman: authority, commitment, real scarcity and framing. In part 3 we cover pricing, decoys and how to design a choice page that actually decides.

FAQ

Does it help if I do not sell anything? Yes. Applying peak-end to a README, reciprocity to an open source landing or loss aversion to a security warning improves the result as much as when you are selling.

How much does it cost? 0 USD. The SKILL.md is a static reference on your machine, no service calls. The cost is the tokens your agent consumes when it reads it (around 3,000-4,000 input tokens; with Caveman active, less).

Is it ethical? That depends on you. The skill warns that "ethical application" is your responsibility: using scarcity when it is fake or decoys to trick buyers works short term and burns you long term. We hammer this in part 2.

Does it work for social media? It works for any content decision. On the blog we use the same skill for the script of the Shorts per post, not only for articles.

What is next

Move on to part 2: persuasion and ethical copy with marketing-psychology and learn to turn mental models into paragraphs that convince. Then part 3: pricing and choice architecture teaches you to design the page where the reader finally pays.

// resource · use it now

marketing-psychology ready to use in your agent

Install the skill, load your product context and paste the prompt into your agent. Works with Claude Code (Pro or Max), Codex, Cursor and any Agent Skills-compatible runner.

Repository license: MIT

1. Install

Only marketing-psychology
npx skills add coreyhaines31/marketingskills@marketing-psychology
Whole collection (cro, copywriting, copy-editing, content-strategy, marketing-psychology)
npx skills add coreyhaines31/marketingskills

2. Pick your case

Paste this prompt into your agent
Load the marketing-psychology skill. We will use it in this project.

1. Create a .claude/product-marketing.md file with: what the product sells, to whom, the measurable goal (signups, sales, clicks) and a 3-line brand voice.
2. Read it before applying any model.
3. When I give you a marketing decision (a headline, a pricing block, an email, a post), reply with:
   - The 2 or 3 skill models that best fit.
   - A sample sentence applying each model.
   - Which metric we should move and how to measure it.
4. If a decision is manipulative (fake scarcity, misleading decoy, invented social proof), push back or refuse and suggest an ethical alternative.
5. Never name the model in the final text the reader will see: apply it, do not explain it.

3. How to use it

  1. Open your project and paste the prompt as the first message.
  2. Review the .claude/product-marketing.md it creates: that file is the source of truth for every decision after it.
  3. Ask it a concrete decision ("redesign the end of this post", "pick the headline for this email out of three").
  4. Once you have a variant, measure it in your analytics (Umami, Plausible, GA4) for 7-14 days before calling a winner.
Prompt and short guide in one fileRepositorycoreyhaines31/marketingskills

Help us grow

codigoagente is free, with no paywalls. If this article helped you, any of these keeps the blog going.

// newsletter

One email a week with the repos worth your time

The week's new articles, tested against real code. No spam; unsubscribe in one click.

AI agents

guide · The complete ECC guide for Claude Code · #8

AgentShield, ECC's scanner, checks CLAUDE.md, settings.json, MCP, hooks and agents. We ran it on a deliberately vulnerable setup and on ECC's own install.
  • Security and code review
  • Claude Code
read article
AI agents

guide · The complete ECC guide for Claude Code · #7

We measured ECC's context footprint with /ecc:context-budget --audit (about 12,500 tokens on the minimal profile) and how to trim it: model, compaction and MCP.
  • Claude Code
  • Skills for your agents
read article
AI agents

guide · The complete ECC guide for Claude Code · #6

What ECC's 28 hooks do, how GateGuard stops the first edit, how to save and resume sessions and how continuous learning instincts work, with a real test.
  • Claude Code
  • Skills for your agents
read article