# Freestyle > Test whether something was actually made for the person and moment it is going to, or is a template performed at them, using the swap test - change the name, the company, the occasion, and see what breaks. Use whenever the user is about to send something that claims to be personal - a cover letter, cold email, proposal, pitch, recommendation letter, condolence note, toast, apology, outreach message, or campaign that is supposed to feel one to one. Also use when the user asks whether something sounds generic, whether it is personalized enough, why nobody replies, or says a message feels like it could have been sent to anyone. - Skill: `michaeldoesmarketing/freestyle` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add michaeldoesmarketing/freestyle` - Raw SKILL.md: https://api.skillmd.com/api/skills/michaeldoesmarketing/freestyle/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - Author: MichaelDoesMarketing (https://skillmd.com/u/michaeldoesmarketing) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/michaeldoesmarketing/freestyle --- # Freestyle In breaking, one of the things you get judged on is whether you danced to the song that was playing. Not whether the moves were good. Whether they were for this song, right now. Bring a rehearsed routine and run it regardless of what the DJ does and the judges see it immediately, no matter how clean the execution is. Most things people send that claim to be personal are rehearsed routines. The recipient can tell, always, and it costs more than sending nothing, because it asks for credit for effort that was not spent. Two questions decide it: - **Could this have gone to anyone?** - **Could this have been sent any time?** Fail both and it is a written. Pass both and it was made for the room it is going into. Use the logic. Do not perform the frame. Never mention dance or judges in the output. ## The swap test The mechanism. For each axis, swap the variable and ask what breaks: - **The name.** Replace the recipient's name throughout - **The organization.** Replace their company, team, or institution - **The domain.** Move it to a different industry or field - **The occasion.** Change what prompted it, or the date - **The sender.** Imagine a different person sent it If nothing breaks on any axis, the piece was never about them. Mark every sentence that survives all five swaps. That surviving text is the template, and it is usually most of the piece. Report the ratio plainly: how much survives, how much breaks. Read `references/swap-test.md`. ## The specificity ladder Not all specificity counts. This is the part people get wrong, and it is why so much "personalized" outreach performs worse than plain outreach. **Tier 1: merge field.** Their name, their company, a fact lifted off their homepage or a recent post. Insertable by a script. Reads as personalized and is not, and the recipient knows the difference instantly. Scores as generic here, deliberately. **Tier 2: research.** Something you had to go find and read. Real effort, and still available to anyone who looked. **Tier 3: observation.** Something only someone paying attention would notice. A tension between two things they said. A pattern across their work. The thing everyone else missed. **Tier 4: stake.** Something only you could say. Shared history, your own position, a risk you take by saying it, a disagreement you are willing to have. Score each specific element by tier. A piece stuffed with tier 1 and nothing above it is worse than a piece with no personalization at all, because it makes a claim about effort it cannot back. Read `references/ladder.md`. ## Responsiveness The second axis, and the one people skip entirely. Could this have been sent six months ago, or six months from now, with no change? Then it is not responding to anything. A piece can be highly specific to a person and still be dead, because it responds to who they are rather than what is happening. Look for: what changed recently, what they just did, what is unresolved right now, what the timing makes true that was not true before. Read `references/responsiveness.md`. Note the exception: some things should be timeless, and forcing urgency into them makes them worse. ## Steps **1. Establish the room.** Who is this going to, what prompted it, what do they already know, what is the moment. If the user cannot answer these, that is the finding and the piece cannot be fixed by editing. **2. Run the swap test.** Five axes. Mark what survives. **3. Tier every specific element.** Anything at tier 1 gets called what it is. **4. Check responsiveness.** Could this have been sent any time? **5. Report the ratio and the ceiling.** How much is template, and the highest tier the piece actually reaches. **6. Say what would move it up.** Specific, findable, and honest. Never invent a connection that does not exist. ## Modes **Swap test** - the default. A draft exists. Run all steps. **The written** - rewrite the user's piece as the mail merge it actually is, with the variables exposed as fields. Nothing added, nothing softened. Then show the ratio. Uncomfortable and accurate, which is the point. **Off the dome** - no draft yet. Establish the room, find what is genuinely available at tier 3 and 4, and hand back the material before a word gets written. Prevents the problem instead of diagnosing it. **Range check** - the piece goes to many people. Decide what can legitimately be shared and what has to be per-recipient, and be realistic about volume. Read `references/genres.md` for what the minimum viable specificity is by type. ## Output format Deliver in this shape unless the user asks for something else. Prose, no long preamble. ``` THE VERDICT One line. Made for them, or a written. THE SWAP What survives all five swaps, quoted. What breaks, quoted. The ratio, plainly. THE CEILING The highest tier the piece reaches, and what most of it sits at. RESPONSIVENESS Could this have been sent any time? What it is responding to, if anything. WHAT WOULD MOVE IT Specific and findable. What to go look at, what to cut, what to say instead. WHAT TO CUT Tier 1 dressed as tier 3. This usually improves the piece by itself. ``` The worksheet is `assets/swap-sheet.md`. The mail-merge output uses `assets/the-written.md`. ## Rules of engagement **Never invent specificity.** If nothing real is available, say so. Fabricated closeness reads worse than distance, and "I've followed your work for years" from someone who has not is the fastest way to lose a reader who checks. **Volume is a real constraint.** Some things go to a thousand people and cannot be bespoke. The useful answer is what to make specific and what to accept as shared, not a demand for the impossible. **Not everything should be a freestyle.** Confirmation emails, invoices, legal notices, policy updates, safety information. These should be identical for everyone, and personalizing them adds noise to something whose job is clarity. Say so and decline the frame. **Tier 1 is not a small failure.** Scraped personalization is the most common version of this problem and the most damaging, because it asks for credit for attention that was never paid. **Judge the piece, not the person.** No implication that the writer was lazy. Most of this comes from being taught that personalization means inserting a name. **Specific and short beats specific and long.** Once a piece is genuinely responsive, length works against it. Do not let the fix become an expansion.