MMarketing Against The Grain
← All frameworks
CommunicationBarbara Jovanovic

Positive-and-Negative Style Prompt

Define desired prose and explicitly block recurring AI writing habits

Difficulty
Easy
Time to result
~days to results
Steps
6
Confidence
99%

This prompt pattern defines style from both directions. Positive instructions specify the desired output: direct, simple, concrete, brief, honest, punchy, and engaging. Negative instructions identify predictable model habits such as startup buzzwords, clichés, metaphors, question-based transitions, expectation-negating constructions, and a maintained list of overused phrases. The negative list is updated whenever editors repeatedly remove the same pattern, converting recurring corrections into preventive controls. During longer chats, the most important constraints are restated because models can drift after several iterations. The mechanism does not replace a client voice library or factual source; it operates as an editorial guardrail layered on top of them, reducing cleanup and making undesirable AI signatures less likely to reach a draft.

Origin

Barbara Jovanovic developed and continually updates this combination of positive style directions and prohibited language while producing client content with ChatGPT. Extracted from Marketing Against The Grain.

Core principles

  • 01Positive style instructions alone leave predictable failure modes
  • 02Concrete adjectives steer output better than vague requests
  • 03A maintained exclusion list removes repeated editing work
  • 04Honesty, specificity, and simplicity outperform startup jargon
  • 05Style constraints must be repeated during long iterative chats

How to run it

  1. 1

    Describe the target voice

    Specify how the writing should sound using concrete, tested descriptors.

    Pro tip Use direct, no-nonsense, punchy, engaging, clear, and simple when they fit the brand.

    Watch out Vague instructions such as 'make it better' provide little usable direction.

  2. 2

    Define sentence behavior

    Set expectations for sentence length, specificity, examples, and overall brevity.

    Pro tip Request concrete details and data points when the source supports them.

    Watch out Brevity should not remove necessary context or nuance.

  3. 3

    Ban structural habits

    Explicitly prohibit unwanted transitions, clichés, metaphors, and formulaic constructions.

    Pro tip Include exact patterns the model repeatedly produces.

    Watch out An overly broad ban can make prose stiff or unnatural.

  4. 4

    Maintain a vocabulary blocklist

    Record words and phrases that consistently signal generic or off-brand AI writing.

    Pro tip Update the list from actual editing experience rather than arbitrary preferences.

    Watch out Do not ban necessary technical terms merely because they are common.

  5. 5

    Reinforce during iteration

    Repeat the highest-priority instructions after multiple rounds of revision.

    Pro tip Keep a short version of the guardrails ready to paste into follow-up prompts.

    Watch out Later revisions can reintroduce habits removed from the first draft.

  6. 6

    Perform a final style audit

    Scan the completed draft against both the desired qualities and the exclusion list.

    Pro tip Add newly discovered recurring failures to the persistent prompt.

    Watch out Passing the style audit does not establish factual accuracy.

In the wild

Removing recognizable AI transitions

A team repeatedly sees headings framed as questions and phrases such as 'this isn't just about' in social drafts. It adds both structures to the prohibited list while requesting short, direct, concrete sentences in the positive instructions.

Future drafts require fewer repetitive style corrections.

Common mistakes

Using only positive adjectives

The model may satisfy broad tone requests while retaining structural habits the editor dislikes.

Copying an arbitrary blocklist

Exclusions should reflect the brand's needs and observed model failures, not another user's preferences without evaluation.

Forgetting prompt drift

A model can stop following early instructions after a long chain of revisions, allowing prohibited patterns to return.

Is it for you?

Best for

Teams repeatedly generating marketing copy, articles, emails, or social posts with the same models.

Not ideal for

Creative writing that intentionally relies on metaphor, ornate language, or experimental structures.

From the transcript

Instructing it to be punchy, still seems to be one of the main ones that perform great, works better than concise, works better than to…

Barbara Jovanovic · 26:30

And then I usually instruct it not to use metaphors, catchphrases, jargon, because it kind of makes it cringe.

Barbara Jovanovic · 26:30

Never use questions as transitions between topics.

Barbara Jovanovic · 30:00

From the episode

I Run a 6-Figure Agency with ZERO Employees (AI Only)

Barbara Jovanovic