Back to Blog
Make Me a Diss Track: AI Prompt Guide for Roasts

Make Me a Diss Track: AI Prompt Guide for Roasts

DissTrack AI·
make me aai diss trackdiss track generatorai roastrap lyrics generator

Your group chat is stalled. Somebody says, “Make me a diss track for Jake's birthday.” Everybody reacts with laughing emojis, then nobody writes a thing. You've got a target, a vibe, and maybe one half-decent joke about his fake-deep gym captions, but not enough to build a full roast that lands.

That's where many fumble. They treat make me a like a magic spell. It isn't. It's an incomplete prompt. If you want bars with teeth, you have to feed the machine better ammo than a name and a vague grudge. Good roast writing still runs on specifics, timing, and restraint.

Battle rap learned that lesson early. Big shifts don't start looking big. The first powered flight by the Wright brothers on 17 December 1903 was witnessed by fewer than ten people according to History Hit's historical facts roundup. Same energy here. One tiny inside joke can carry a whole verse if it's the right one. That little detail everybody in the room knows? That's the engine.

If you work with AI a lot, you already know the pattern. Generic input gives generic output. Teams using AI for publishing learn the same lesson fast, which is why this guide to AI content for agencies is useful outside marketing too. The principle transfers cleanly. Strong creative output starts with sharp instructions, not wishful typing.

From 'Make Me a' to Lyrical Assassin

A lazy request sounds like this: “Make me a diss track about Tyler.”

That prompt gets you bargain-bin smoke. It'll spit out broad insults, recycled swagger, and lines so generic they could fit Tyler, Derek, or your cousin who still says “lit” like it's a fresh discovery. The AI isn't failing. You just handed it a blank beat and a shrug.

What the weak prompt misses

A proper roast needs tension. Not fake drama. Specific friction.

Maybe Tyler never pays for parking and always ends up circling the block while acting like it's a strategy. Maybe he talks like a startup founder because he sold two T-shirts once. Maybe he claims he's “locked in” while posting mirror selfies at the gym instead of lifting. That's not random trivia. That's voiceprint.

Generic insults hit nobody. Specific details hit everybody who knows the target.

The difference between “make me a diss track” and a memorable roast is the same difference between shadowboxing and landing a clean hook. One is motion. The other is contact.

What a strong request sounds like

Try this frame instead:

  • Who they are: friend, coworker, rival, party guest
  • What makes them roastable: habits, contradictions, public fails
  • What tone you want: playful, petty, surgical, absurd
  • What should stay off-limits: private pain, family issues, real vulnerabilities

That last part matters more than people think. Sharp is good. Reckless is amateur.

When people type Make me a diss track, they're usually asking for confidence on demand. What they need is structure. Once you stop begging the tool for magic and start directing it like a battle coach, the lyrics stop sounding rented.

Gathering Your Lyrical Ammunition

Before you write the prompt, gather intel. Not dirt for dirt's sake. Data points.

In analytics, a data point is a single observation used to identify patterns and trends, and people use those observations as inputs for monitoring and decision-making, as explained in Lenovo's glossary entry on data points. Same principle here. One odd habit is funny. Several connected habits create a character the AI can write against.

A diagram outlining the intelligence gathering process for creating a diss track, focusing on four key areas.A diagram outlining the intelligence gathering process for creating a diss track, focusing on four key areas.

Four buckets that make a roast personal

Don't dump random notes into one pile. Sort your ammo.

  • Weaknesses: These are the harmless flaws they can't hide. Bad fashion choices, fake confidence, awful taste in music, chronic lateness, overexplaining simple things.
  • Past incidents: The failed karaoke performance. The burnt barbecue. The fantasy football trash talk that aged horribly.
  • Catchphrases and habits: Maybe they say “trust the process” every time they mess up. Maybe they laugh before their own joke lands.
  • Public perception: This is how the room reads them. Try-hard? Chaos agent? Human typo? Low-budget philosopher?

That's the raw material. If you want extra inspiration while collecting angles, an AI roast generator can help you pressure-test whether your details are funny or just overfamiliar to your own friend group.

Good intel beats more intel

You don't need a dossier. You need details that create patterns.

A weak note says, “He dresses weird.” A stronger note says, “He wears expensive sneakers with wrinkled office shirts and talks like he's curating a brand.”

A weak note says, “She's annoying.” A stronger note says, “She corrects people's pronunciation, then says ‘expresso.’”

Practical rule: If the detail could apply to half the planet, it's not ammunition yet.

A fast field method

Use this quick collection method when you're under pressure and the group wants bars tonight:

  1. List three habits that always get mentioned when this person leaves the room.
  2. Add two receipts. Moments everybody remembers.
  3. Find one contradiction. This is battle rap gold. Broke but flexing. Lazy but calling others unmotivated. Tone-deaf but always singing.
  4. Write one line they'd typically say. That gives the AI something to mimic or flip.

That last move is sneaky. Once the model has a catchphrase or speech pattern, it can build punchlines that feel aimed, not mass-produced.

Crafting the Perfect Prompt

Many users ruin the output in the first sentence. They ask for a diss track the way somebody orders fast food. Name, target, done. Then they act surprised when the result tastes generic.

Technical writing works the same way prompts do. The right level of detail depends on who's reading and what they need, and good guidance recommends step-by-step examples so the message stays clear, especially for less experienced users, as outlined in ClickHelp's piece on optimal detail in documentation. If you're new to prompting, be explicit. Don't try to sound clever. Try to be usable.

Screenshot from https://aidisstrackgenerator.comScreenshot from https://aidisstrackgenerator.com

The prompt skeleton that actually works

Here's the blueprint I'd use:

  • Target and relationship: “Write a diss track about my coworker Marcus.”
  • Roastable traits: “He calls every minor task a strategy, reheats fish in the office, and acts like basic spreadsheet work is elite warfare.”
  • Tone: “Funny, cutting, not hateful.”
  • Style: “Battle rap with tight punchlines and direct setup-payoff structure.”
  • Must-use details: “Use his phrases ‘circle back' and ‘low-hanging fruit.’ Mention his Bluetooth earpiece.”
  • Boundaries: “Keep it workplace-safe. No family, health, or protected traits.”

That's enough to shape the voice without handcuffing the verse.

Here's where specialized tools help. If you want to sharpen a single haymaker before generating the full track, an AI punchline generator is useful for testing setups, flips, and callbacks before you lock in the complete prompt.

Before and after

Look at the difference.

Prompt typeExampleLikely result
Thin“Make me a diss track about Sarah”Broad insults, weak identity, filler bars
Better“Make me a funny diss track about my coworker Sarah who loudly slurps soup, thinks Nickelback is peak rock, and wears socks with sandals”Clearer voice, more usable punchlines
Surgical“Write a playful but sharp battle rap about my coworker Sarah. She loudly slurps soup, thinks Nickelback is peak rock, wears socks with sandals, and says ‘let's synergize' in normal conversations. Keep it party-safe and packed with specific wordplay.”Bars that sound targeted and performable

The trick isn't stuffing in more words. It's feeding details that create angles.

A prompt should tell the model who the target is, why they're roastable, and where the red line sits.

If you post roast clips or short-form performance content, the prompt discipline overlaps with what social teams already do when they engineer tone and variation. This SleekPost AI social media guide is useful for seeing how platform-ready phrasing changes output quality without bloating the brief.

A quick demo helps if you want to hear how tighter instructions change the result in practice.

Copy-paste templates

Use one of these and customize it.

Party roast template

Write a funny diss track about my friend [Name]. We know each other from [context]. Roast these traits: [habit 1], [habit 2], [habit 3]. Mention these incidents: [incident 1], [incident 2]. Use a [style] vibe. Keep it playful, clever, and clean enough to read out loud at a party.

Harder battle template

Write an aggressive but not hateful battle rap aimed at [Name], who is my [relationship]. Focus on these contradictions: [x], [y], [z]. Use direct punchlines, layered wordplay, and a confident flow. Include these phrases or references: [details]. Avoid topics involving [boundaries].

Tuning Your Style and Savagery

Same facts, different delivery. That's where style does the dirty work.

If the content says, “Your friend is a fraud,” the style decides whether that lands like a smirk, a slap, a chess move, or a flamethrower. Most weak AI roasts don't fail because the insult is wrong. They fail because the delivery mode doesn't match the room.

A graphic guide illustrating four different styles of diss tracks, including subtle jab, blunt attack, calculated strike, and roast.A graphic guide illustrating four different styles of diss tracks, including subtle jab, blunt attack, calculated strike, and roast.

Pick the style by audience, not ego

Here's the clean comparison.

ModeHow it soundsBest useRisk
Subtle jabDry, witty, knowingBirthdays, friend groups, streamsToo soft if the room expects impact
Blunt attackDirect and forcefulCompetitive cyphers, louder personalitiesCan sound crude if the writing is weak
Calculated strikeStrategic, layered, cerebralTargets with obvious contradictionsCan get too dense if overstuffed
Over-the-top roastExaggerated, theatrical, funnyShorts, skits, crowd momentsEasy to become clownish

A lot of people default to the most aggressive setting because they think harder means better. Rookie move. If the target is your buddy from work and the crowd is drinking from paper cups in a rented hall, “nuclear villain monologue” probably isn't the lane.

Match genre to the target

Genre tags matter because they shape cadence and line construction.

  • Old School Boom Bap works when you want setup, punchline, setup, punchline. Clean, easy to perform, room-friendly.
  • Battle Rap is built for direct address. Great for humiliation by precision.
  • Trap gives swagger and bounce, but it can blur punchlines if your details aren't strong.
  • UK Grime or Drill can make the tone feel colder and more confrontational.
  • West Coast style often suits playful confidence better than outright venom.
  • Emo Rap can work for ironic self-aware disses, especially when the target is melodramatic.

Don't pick a tag because it sounds cool. Pick it because it helps the joke travel.

The nastiest line on paper can die in performance if the rhythm fights the room.

Set the heat with intention

Think in bands, not numbers.

Playful means you're teasing. Good for birthdays and friendship roasts.
Firm means the gloves are off, but you still want laughs.
Brutal means the punchlines are designed to sting, and you'd better trust the audience and the target.
Chaotic is entertainment-first. Bigger exaggeration, less realism, more meme energy.

If you're using DissTrack AI, the useful move isn't maxing out savagery by default. It's pairing the right heat with the right genre, then revising a few lines by hand so the track sounds intentional instead of randomly hostile.

A fast decision grid

Use this when you're stuck:

  • Roast for a party friend: subtle jab or over-the-top roast
  • Roast for a competitive freestyle clip: calculated strike or blunt attack
  • Roast for a coworker send-off: subtle jab, clean details, low collateral
  • Roast for a content skit: over-the-top roast with obvious exaggeration

Style isn't decoration. Style is targeting.

The Final Polish Adding Your Human Touch

AI can draft the scaffold. You still need to make it sound like it came from a human mouth, not a machine with anger issues. The edit is where borrowed fire becomes your own.

The fastest mistake is copy-paste confidence. A line can look sharp on screen and still die when you read it out loud. Rhythm exposes fraud quick.

A checklist titled Human Touch Checklist for Polishing Your Diss Track with five steps for songwriters.A checklist titled Human Touch Checklist for Polishing Your Diss Track with five steps for songwriters.

The five-edit finishing pass

Use this short pass before you perform, post, or send anything.

  1. Read every bar aloud. If you trip over it, rewrite it. Mouthfeel matters.
  2. Replace one generic insult per verse with a real detail from your notes.
  3. Tighten the opener. Your first two lines should identify the target fast and set the tone.
  4. Punch up the closer. End with the line people repeat after the track stops.
  5. Cut dead weight. If a bar explains the joke instead of landing it, kill it.

Where humans still beat the machine

AI tends to overstate and overfill. It'll sometimes pick the obvious insult when a sideways angle would hit harder. That's your lane.

A human editor catches things like:

  • Voice mismatch: the lyric sounds too formal, too online, or too theatrical for you
  • Weak references: the line mentions something true but not funny
  • Missed callbacks: a joke from the first verse should return later and hit harder
  • Performer logic: the line works on screen but not in breath

If you want a stronger feel for how lyric structure works before editing the final version, this piece on how to write lyrics to a song is worth reading for cadence, phrasing, and line trimming.

Cut any line you need to defend. If it was really fire, it would defend itself.

The last tweak that changes everything

Swap one polished but impersonal bar for one slightly messy line that sounds like you. Your slang. Your phrasing. Your way of clowning somebody when you're in the room.

That's the stamp. That's what stops the track from sounding generated.

The Unwritten Rules of AI Roasting

Roasting is supposed to be sharp. It isn't supposed to leave wreckage you can't walk back.

One of the biggest gaps in prompt advice for personal content is safety. Guidance on content angles points out that many guides skip privacy, consent, and reputational risk, which matters a lot when prompts involve real people and shareable outputs, as noted in this discussion of content-angle gaps. That matters even more when the request starts with Make me a and ends with something personal enough to spread beyond the original joke.

Draw the line before you generate

Use a simple test.

  • Would you say it in the room with them present?
  • Would the target understand it as a joke?
  • Are you exposing something private they didn't choose to share?
  • Would this still feel funny if it got screenshotted and traveled?

If the answer goes sideways, don't generate it. Or write it manually and keep it private. Some material isn't improved by automation.

When not to use AI for the roast

AI isn't always the right weapon.

Skip it when the target is dealing with real loss, public embarrassment they didn't consent to revisit, identity-based harassment, or anything tied to private relationships. Also skip it when the joke depends on emotional nuance you don't trust a generic prompt to handle.

Sometimes the smarter play is a hybrid. Let the tool draft a few structures, then write the final lines yourself. Sometimes the smartest play is no roast at all.

Funny ages well. Cruel doesn't.


If you want a faster way to turn scraps of inside jokes into structured roast lyrics, DissTrack AI gives you a prompt-driven starting point with style controls, editable outputs, and private generations until you choose to share.

Related Articles