A copied prompt encodes somebody else's context, which was the part that made it work.
That is why prompt collections disappoint. The prompt that produced a great result for someone else was carrying their positioning, their customers and their examples — and the wording you copied is the least important thing in it.
Patterns transfer. Prompts do not. Seven patterns cover most marketing work, and each one is a shape you fill with your own material.
1. Extraction
Pull structured information out of unstructured material you already have.
What it is for: turning a pile of things nobody has read into something usable — a year of support tickets, survey responses, sales call notes, reviews, replies to your newsletter.
The shape:
Here are 200 support emails from the last six months. [pasted]
List every distinct question, grouped by theme. For each theme, give the number of instances and quote the two clearest examples in the customer's own words.
Why it is first: it is the highest-value pattern in marketing and the least used. The output is material about your own audience that no competitor has, and it turns directly into a content plan. The model is not supplying knowledge here — it is organising yours, which is the task type it is most reliable at.
2. Variation
Generate a wide spread of options, structurally different rather than superficially different.
What it is for: subject lines, headlines, CTA wording, opening lines, positioning statements.
The shape:
Here is the email. [pasted]
Write fifteen subject lines: three as questions, three as plain statements, three leading with a number, three naming the reader's situation, three as incomplete sentences.
The instruction that matters is the structural breakdown. Asking for fifteen subject lines produces fifteen similar ones. Asking for five structural categories produces genuine spread, which is the point of generating at volume. Choosing between them.
3. Constraint
Define the output by what it must not contain.
What it is for: anything that will be published, because generated marketing prose has predictable tells.
The shape:
No introduction. No conclusion. No summary sentences. No superlatives. No urgency language. No claim the source material does not support. No sentence that would be equally true on a competitor's site. Maximum 400 words.
The last one is the strongest single constraint available. "Would this be equally true for a competitor?" removes generic content at the generation stage rather than the editing stage.
Constraints are cheap to write and disproportionately effective, and they are what most people leave out.
4. Example-led
Show one piece that gets it right instead of describing the style you want.
What it is for: matching a voice, a format, or a structure you already use.
The shape:
Here are two of our published articles. [pasted]
Write the opening three paragraphs of a new one on [topic], matching how these open — including the fact that they state the conclusion before explaining it.
One concrete example outperforms a paragraph of description, and naming what to copy from the example matters as much as supplying it. "Match the tone" is vague; "match how these open" is a specific instruction. Getting voice right.
5. Critique
Ask what is wrong with something rather than asking for a better version.
What it is for: your own drafts, an existing page, a competitor's page, or a generated draft you are unsure about.
The shape:
Here is a landing page. [pasted]
List every claim it makes that is not supported on the page itself. Then list every sentence a sceptical reader would stop at, and say why.
Why this beats "improve it": a rewrite replaces your judgement with a model's. A critique gives you a list and you decide. It is also the only pattern where a mediocre model output is still useful — half a good list is still half a good list.
6. Transformation
Convert something you have into a different form.
What it is for: one article into an email, a call transcript into a case study outline, a long page into a summary for a different audience, a technical explanation for a non-technical reader.
The shape:
Here is a 3,000-word article. [pasted]
Write the email that sends people to it: 150 words, one idea, ending with the single most useful sentence in the article as the reason to read it.
Reliable, because the input and the output are both concrete. Nothing is being invented, which removes the main failure mode.
7. Source-bounded
Restrict the model to material you supplied, explicitly.
What it is for: anything factual, which is most marketing content that has to be true.
The shape:
Using only the documentation pasted below, answer these six questions. If the documentation does not answer one, say so rather than filling the gap.
The second sentence is the pattern. Without it, a gap gets filled plausibly, and a plausible fill is indistinguishable from a sourced answer.
This is the pattern that most reduces fabrication, and it is still not verification — check anything that matters at the primary source. The method.
Combining them
Most real tasks are two or three patterns.
A worked example — turning support tickets into an article:
- Extraction — pull the themes and the customer wording out of the tickets
- Source-bounded — draft the sections using only your documentation and the tickets
- Constraint — no introduction, no superlatives, nothing not in the sources
- Critique — ask what a sceptical reader would stop at
- Then edit by hand. The four passes
Five steps, and only steps two and three are generation. The other three are the model organising, checking, or restricting itself — which is where it is most reliable.
What to do with the ones that work
Keep the shape, not the words.
A pattern you use repeatedly is worth writing down as a template with your context file attached — but keep it short enough to adapt. A prompt long enough to feel authoritative is usually a prompt nobody edits, and an unedited prompt applied to a task it does not quite fit produces confidently irrelevant output. Building a set a team can use.
Frequently asked questions
What are the best AI prompts for marketing?
There is no transferable list, because the prompts that work carry their author's context. What transfers is the pattern: extraction, variation, constraint, example-led, critique, transformation, and source-bounded. Fill each with your own material.
Why don't copied prompts work as well for me?
Because the copied wording is the least important part. The prompt that worked for somebody else was carrying their positioning, their customers' language and their examples, and none of that came with the text you copied.
What is the most useful prompt pattern for marketing?
Extraction — pulling themes and customer wording out of support tickets, replies, reviews and call notes. It produces material about your own audience that no competitor has, and it is the task type models are most reliable at, because they are organising your knowledge rather than supplying their own.
How do I stop AI filling gaps with invented information?
Use the source-bounded pattern: restrict it to material you supplied and explicitly instruct it to say when the source does not answer something. Without that instruction the gap gets filled plausibly, and a plausible fill looks identical to a sourced answer.
Should I ask AI to improve my draft or critique it?
Critique. A rewrite replaces your judgement with the model's, while a list of specific weaknesses leaves you deciding what to change. It is also the pattern where a mediocre output is still useful — half a good list is still half a good list.
How long should a marketing prompt be?
Long where the length is supplied material, short where it would be instruction. A prompt long enough to feel authoritative usually stops getting edited, and an unedited prompt applied to a task it does not quite fit produces confidently irrelevant output.
Free: The 60-Minute Email Authentication Fix
A no-fluff checklist to set up SPF, DKIM & DMARC correctly and pass Gmail & Yahoo's sender requirements.

Muhammad Basim has worked in digital marketing since 2013, focused on email deliverability and AI-assisted content production. He is the author of The Email Deliverability Playbook and The Email Copywriting Playbook.
Related Articles

Does Send Time Optimisation Work?
Send time optimisation models are trained on when recipients opened messages, and open timestamps now include machine activity that has nothing to do with when anyone was reading. Privacy protection pre-fetches message content on its own schedule, regardless of whether the recipient opened it. Security gateways fetch content during pre-delivery scanning, which happens before the […]

Prompting for Marketers: The Part That Actually Matters
Output quality is set mostly by the material you supply, not the phrasing you use. That is the finding people take longest to accept, because it is less interesting than the alternative. A carefully worded request with no context produces a competent generic answer. A plainly worded request with your actual customer language, your positioning […]

What You Grant When You Connect a Tool
Your security perimeter includes every vendor you have ever connected, including the ones you stopped using and never disconnected. A connection is a standing grant. It does not expire because you stopped logging in, it does not lapse because the trial ended, and it does not disappear when you delete the app from your phone. […]

