tellaprompt

AI prompts for copywriters

Last updated

AI won't replace your taste — it industrializes your throughput. Used right, it mines customer language, spits out structured variant sets you curate, holds brand registers you define, and critiques drafts without ego. The five templates below are that workflow, in order.

The difference between slop and usable raw material is mechanical: specific levers instead of "more ideas", locked registers instead of "make it snappy", and facts instead of adjectives.

How to prompt as a copywriter

  • Feed precedents, not descriptions: three past winners teach tone better than ten adjectives about it.
  • Keep a register file per brand — formality, taboo words, CTA style, emoji policy — and paste it into every session. Context beats prompting.
  • Demand quantity with structure: 20 headlines in 4 named categories beats "some ideas", because curation is your job and structure makes curating fast.
  • Ban the AI-isms up front — "unlock", "elevate", "game-changer" — or you'll edit them out of everything.

The five templates

Headline variant engine (Osborn levers)

One approved concept, and you need a wall of genuinely different headline options — not five rewordings.

Product: [PRODUCT]. Core benefit, as a fact: [E.G. "SETS UP IN 4 MINUTES"]. Audience: [AUDIENCE]. Register: [E.G. DRY, CONFIDENT, NO EXCLAMATION MARKS].

Generate 16 headlines: 2 per lever — adapt (borrow a structure from another industry), modify (unexpected tone), magnify (dramatize the benefit), minify (max 4 words), substitute (replace the claim with a story), rearrange (effect before cause), reverse (say what it's NOT / flip a cliché), combine (merge benefit with an objection).

Label each with its lever. The benefit number must appear in at least 6 of them.

Why this works: Named levers force real variety — the Osborn checklist turns "brainstorm" into eight distinct operations, and the number-quota keeps fact density in the set.

Osborn leversFact densityOutput format

Brand register lock

Every output for this client should sound like the client — starting now, not after three correction rounds.

Here is the register for [BRAND] — treat it as law for everything in this conversation:
- Formality: [DU/SIE/FIRST NAMES; "YOU" STYLE]
- Sentence rhythm: [E.G. SHORT, PUNCHY; OR FLOWING]
- Words we love: [LIST]
- Banned: [WORDS, PHRASES, EMOJI POLICY, EXCLAMATION MARKS?]
- CTA style: [E.G. IMPERATIVE, LOWERCASE, MAX 3 WORDS]

Calibration examples of on-register copy:
[PASTE 2–3 APPROVED PIECES]

Confirm the register in one line, then wait for my first brief.

Why this works: A standing register file plus calibration examples is context-first prompting for voice — the examples anchor what rules alone can't describe, and one paste beats re-briefing every session.

Context firstPrecedent exampleRole definition

Voice-of-customer mining

Before writing a landing page, you want the audience's own words — pains, desires, objections — as raw material.

Below are verbatim customer quotes about [PRODUCT/CATEGORY] (reviews, tickets, forum posts).

Extract:
1) The 10 most charged phrases, quoted exactly, with what they reveal.
2) Top 5 pains and top 5 desired outcomes — in the customers' wording, not marketing speak.
3) The 3 objections that keep appearing.
4) 5 headline drafts that reuse the strongest verbatim phrases.

Quotes:
[PASTE 20–100 QUOTES — MORE IS BETTER]

Why this works: Copy that converts uses the language customers already speak; this makes the AI a mining tool for that language instead of a generator of plausible-sounding guesses.

Voice of customerFact densityOutput format

The brutal critique pass

Draft's done, deadline's tomorrow, and you need the weaknesses found tonight — not in the client call.

You are a brutal creative director reviewing the copy below. Never open with praise.
1) Reality check: what does this actually promise, in one sentence?
2) Where does the logic or the offer break?
3) What is this copy avoiding (price? proof? a real claim?) — and what does that avoidance cost?
4) What would a top direct-response writer do differently here?
5) Prioritized list: what to CUT, not what to add.
6) Close with the one uncomfortable question about this piece.

Copy:
[PASTE DRAFT]
Context: audience [WHO], goal [ACTION], channel [WHERE].

Why this works: The six-step critic finds the avoidance — the thing the copy is dancing around — which is where most drafts actually fail; the cut-don't-add rule fights bloat.

Brutal criticContext first

Anti-AI style pass

The draft reads competent and dead. Make it sound like a person wrote it on a good day.

Rewrite the copy below with these style rules:
- Vary sentence length radically: at least one 1–3 word sentence per paragraph, at least one long winding one.
- Fragments allowed. Deliberately.
- Kill these if present: "unlock", "elevate", "seamless", "game-changer", "in a world where", rhetorical question openers, tidy summary endings.
- Keep every fact and number exactly as is.
- One striking, concrete image instead of any abstract claim — pull it from the product's physical reality: [ONE DETAIL, E.G. "THE HINGE CLICKS"].

Copy:
[PASTE DRAFT]

Why this works: Burstiness — hard rhythm variation plus permitted fragments — is the strongest single de-slopping lever, and the concrete-image rule replaces abstraction with something only this product can say.

Anti-AI styleFact densityTest & iterate

Frequently asked questions

Will AI copy hurt my clients' brand or SEO?

Unedited, generic AI copy can — it converges on the same phrases everyone else publishes. The workflow above avoids that by construction: customer language in, brand register locked, variants curated by you, style pass against AI-isms. The output is raw material; the byline is still earned.

How do I keep AI from flattening my voice?

Two levers: calibration examples (paste 2–3 pieces of your best work and demand that register) and explicit style rules (rhythm variation, banned words). Never ask for "professional and engaging" — that's the slop setting.

What's the fastest win for a working copywriter?

The variant engine. You stay the judge; the machine does the combinatorics. Twenty structured options in a minute turns curation — your actual skill — into the whole job.

Can I use these prompts for German / French / Spanish copy?

Yes — write the register file in the target language and add local taboos (du/Sie decisions, formality norms). This site itself ships in six languages; the mechanics are language-agnostic even when the idioms aren't.