Marketing Mindset — the marketing OS for AI agents
A free, MIT-licensed skill that gives AI agents a real marketer's operating mindset: how to think, decide and prioritize on any marketing, growth or client-acquisition task. Not another tactic pack — the decision layer that produces tactics as its output.
What it is
Marketing Mindset is an agent skill, not a marketing course. It encodes how a director-level B2B marketer decides: a 3-month horizon instead of 2-year plans, hypotheses that any teammate can test this week, competitors as the first source of truth, minimum volumes before a test can be judged, and a hard rule to scale whatever comes easy. The agent reads it and answers with a real opinion — including honest "this will not work" — instead of a template.
Install
npx skills add axelfreeman/marketing-mindsetOr copy SKILL.md into your agent's skills folder, or point the agent at llms.txt and AGENTS.md.
The problem it solves
A developer can generate code but not content. "How do I do marketing" is genuinely unclear to them, and every other "marketing skill" for agents hands over tactics: 50 CRO hacks, 30 subject lines, a copywriting formula. Tactics do not answer the questions that actually block a founder: should I do X to get Y, where do I get my first client, is this idea worth the next month. Those are decisions, and decisions need a mindset.
Six principles
- Don't learn marketing from stale sources. First Google results and cached model doctrine are globally outdated — work from fresh, recent data.
- Three-month horizon. Evaluate every idea on a 3-month window; nobody pitching a 2-year cycle is being straight with you.
- The user moves first. Bold and hacky first steps belong to the founder; the agent helps instead of blocking.
- Every hypothesis must be testable fast — by anyone on the team, not by a specialist.
- Marketing runs ahead of the product. The first thing anyone touches is the marketing, and it may legally lead the build.
- Marketing never works for free. Marketing is exchange: every action must trade for something, and training time is not free either.
Test limits — how much volume before a test can be judged
Below the limit you are measuring randomness, not the market. Each channel has a minimum volume below which the honest answer is "not enough data yet". These are the ranges the market reports; declare the number before the test starts, not after.
| Channel | Minimum volume before a verdict | What it can tell you |
|---|---|---|
| Cold email — deliverability / wording smoke test | 50–100 sends | Whether the email lands and reads plausibly. Not whether the offer works. |
| Cold email — reply-rate test | ~1,500–2,000 sends per variant (industry testing guides) | Whether one variant genuinely beats another, instead of a quiet week. |
| Cold email — subject line / open-rate test | 100–500 sends per version | How the subject performs, since opens are frequent events. |
| Landing page smoke test | 100–200 targeted visitors | Whether the promise produces interest at all (≈30 leads at 15% capture from 200 cold visitors). |
| Strict A/B test | ~10,000 visitors per variation, ≥300 conversions | Statistical significance — usually out of reach for a startup's first tests. |
| Paid ad | Spend gate of 1–3× target CPA and 48–72 hours | Whether to keep, re-hook or kill the ad — never judge during the learning phase. |
| Cold calls | Volume until a repeatable pattern appears in one segment | Whether the script survives real conversations, not one lucky call. |
The rule that follows: what comes easy, scale it
When something performs noticeably easier than the alternatives, that is a signal, not luck — it gets the next unit of effort first, and money only after that. The limit exists to make the gap visible: below the limit, "easy" and "ordinary" are indistinguishable, so people either chase randomness or ignore the one channel that was actually working.
Declare the volume → run one variable → stop at the limit → scale what came easy. Anything else is interpretation.
FAQ
How many cold emails do I need before I can judge a campaign?
Enough to separate signal from a quiet week. Testing guides put a reliable read on reply rate at roughly 1,500–2,000 sends per variant; the popular 50–100 rule proves the email went out, not that it works. Use 50–100 sends as a smoke test of deliverability and wording — never as a verdict on the offer.
Is 100 sends enough to test a subject line?
For open-rate tests, broadly yes — opens are frequent and 100–500 sends per version is the range vendors quote. For reply-rate tests the same number is off by an order of magnitude.
How many visitors does a landing page need to validate an offer?
100–200 targeted visitors is the common minimum. At a 15% capture rate, 200 unique cold visitors produce about 30 leads — enough for a handful of real conversations before you build anything.
How much traffic for a statistically solid A/B test?
Published rules of thumb sit near 10,000 visitors per variation with at least 300 conversions. That is why first tests are decision tests with a declared stopping point, not significance tests.
How much spend before judging a paid ad?
A spend gate of 1–3× your target CPA and 48–72 hours, never during the learning phase. Below the gate you are reading platform noise.
What if my market is too small to reach the minimum?
Change the unit of the decision: a longer window, another observable such as replies per twenty hand-written conversations, or qualitative signal from real buyers. Say plainly that the sample is small — that is an honest constraint, not a result.
Why do limits matter?
Below the limit you measure randomness. A test with no declared volume cannot be closed, cannot be called a success, and quietly turns into a campaign that is only ever interpreted.
What does "if it comes easy, scale it" mean?
Scale the channel, message or offer that performed noticeably easier than the rest — it goes first. You can only see that gap once you are above the limit, because below it easy and ordinary look the same.
Is this just another tactic pack?
No. It is the decision layer: how to think, choose and prioritize. Competitor analysis, ad copy, landing pages and outreach are the output the agent produces, not the input it starts from.
Is it free, and which agents can use it?
MIT licensed and free. It works with Claude Code, Cursor, Codex, ChatGPT and any agent that reads SKILL.md, AGENTS.md or llms.txt.
How this differs from other marketing skills
| Marketing Mindset | Typical tactic packs | |
|---|---|---|
| Unit of value | Decisions and judgment | Checklists and templates |
| Answers | "Should you do this, and how do we know in 3 months" | "Here are 50 subject lines" |
| Test discipline | Declared minimum volume per channel | None |
| When the idea is bad | Says so plainly | Produces output anyway |
| Freshness | Rejects stale sources and >6-month reports | Cites evergreen blog doctrine |
Sources for the numbers above
Cold email sample-size practice: Getlead, Growtoro, Woodpecker. Landing-page smoke tests: GLIDR. A/B sample size: AB Tasty. Ad kill rules: Clikim. Numbers are market rules of thumb, not guarantees — measure your own baseline.