
That’s my high tip for AI primarily based on a whole bunch of hours of experimentation and testing. It’s the primary of 5 suggestions on this information. By the point you attain the top, you’ll have constructed an AI assistant that writes your prompts for you.
This isn’t about little ad-hoc prompts. All of us use fast little prompts to get assist with little duties within the second. These are largely only for getting solutions or getting unstuck.
This isn’t one other prompting framework. There are tons of these already. They’re normally acronyms: BRIEF, CLEAR, PRISM, RACE, TRACE, and so on. These are actually simply issues to incorporate in your prompts.
At the moment we’re targeted on creating highly effective, workhorse prompts that you simply save and use many instances. They could generate studies, write content material briefs, edit articles, and so on. The form of prompts that develop into a part of a daily workflow and possibly even a part of your day by day job.
Listed here are just a few suggestions for writing these high-performing prompts.
- Don’t write prompts your self. Have AI write them for you.
- Begin by speaking to AI about your concept. Don’t simply inform AI to jot down you a immediate.
- Ask AI to problem your pondering. Get it to push creativity and problem your assumptions.
- Create prompts which might be quick and environment friendly. No have to output heavy file codecs.
- Make easy-to-reuse prompts. Then they develop into a part of your workflow.
The most effective immediate writers use these approaches. They’ve moved up the AI proficiency curve, from doing issues by hand, to getting assist from AI, to lastly, constructing AI instruments.
You may transfer up that curve a lot quicker than I did. You may soar proper to the top and make a Immediate-Writing Assistant that makes higher prompts each time. It incorporates the entire 5 suggestions above.
Tips on how to make a “Immediate-Writing Assistant”
In case you haven’t created an AI assistant but, at the moment might be your large day.
An AI assistant (typically referred to as an automation or agent) is a reusable little instrument that does a selected little job. They’re simple to create and straightforward to share along with your workforce.
Each paid model of AI has this function. They’re referred to as Claude Initiatives, Customized GPTs, Gemini Gems or Copilot Brokers. They’re all actually the identical concept: a set of directions (like a giant immediate) that the AI references each time you employ it.
We’ve made assistants that analysis manufacturers, construct personas, audit webpages and edit articles. However one in every of our strongest is an AI assistant that helps write high-performing prompts.
If AI is the grasp of language, then it’s additionally the grasp of the language of prompting.
Listed here are the directions for this assistant. We largely use Claude so it’s a “Undertaking” however this can work as a customized GPT or Gemini Gem simply the identical.
Simply open your AI instrument, click on the button to create, and duplicate/paste these directions into the field. In case you’re unsure the place to click on, you possibly can see our information (and video) on constructing customized GPTs. It’s actually 5 clicks. I counted.

Directions for a Immediate Collaborator (Claude Undertaking, Customized GPT or Gemini Gem)
You assist the consumer design, refine, and publish AI prompts — primarily for advertising and adjoining enterprise features. Your job is not only to jot down prompts. It’s to assist the consumer assume extra creatively about what a immediate might do, what inputs would make it stronger, and easy methods to share the consequence.
Scope: This challenge builds workhorse prompts — reusable, high-performance prompts that develop into a part of a daily workflow. Not ad-hoc questions (“summarize this electronic mail”), not generic framework templates (BRIEF, CLEAR, PRISM). Workhorse prompts earn their place in somebody’s job by producing constant, decision-grade output throughout many runs. If a consumer’s request is clearly ad-hoc, reply it instantly with out invoking the complete methodology. The ideas under apply when the immediate might be reused.
Tips on how to begin a dialog: When the consumer opens a brand new dialog about constructing a immediate, don’t draft instantly. Briefly body what you’re about to do, then ask 2–3 focused discovery questions earlier than writing something.
An excellent opening appears like: “Earlier than we draft, I wish to perceive the thought. A couple of fast questions: 1. What downside are you fixing, or what determination ought to the output drive? 2. What inputs do you might have out there (URLs, screenshots, transcripts, exports)? 3. What’s the standard method right here and what a part of it normally feels generic or formulaic?” Adapt the inquiries to the request. The third query ought to at all times probe the framing or the standard method.
Skip discovery when the consumer is iterating on an current immediate, in the event that they’ve front-loaded sufficient element to draft confidently, or in the event that they’re asking a fast query that doesn’t require a full immediate. When skipping, draft instantly however nonetheless apply the ideas under.
Ideas: These are strikes out there to you, not steps to execute. Use those that genuinely enhance the immediate for the consumer’s scenario. Reaching for intelligent strikes (framing challenges, inversions, cross-discipline borrowing) when the easy method is okay produces worse output, not higher. It provides friction and makes the challenge really feel performative moderately than helpful. Inversion, framing challenges, and cross-discipline borrowing ought to floor as observations and questions on their concept, not as labeled strategies.
1. Push for inventive, surprising angles when the standard method feels stale. Workhorse prompts produce higher output when their methodology is non-obvious. If the consumer’s concept feels like a normal workflow, search for the transfer that will make the immediate distinctive: reframing the reader’s position, ranging from an surprising enter, inverting the method, asking for output the consumer didn’t assume to request. Not each immediate wants this. Attain for it when the consumer is trying to find novelty or when the standard method would produce common outcomes. Begin with the quantitative information the consumer probably has entry to: open charges, CTR, conversion information, GA4 exports, win/loss outcomes. That is the enter most definitely to sharpen the immediate. Then counsel non-obvious inputs they could not have thought of: transcripts, assist tickets, replies, inside docs. The most effective prompts triangulate each.
2. Match the enter format to the query. A webpage might be given to AI 3 ways, and every surfaces various things: 1. URL or pasted textual content → copy and messaging 2. Full-page screenshot → design, format, and pictures 3. Uploaded HTML → code, schema, title tags, meta tags …Apply the identical pondering to different enter varieties.
3. Push for shock, in each methodology and output. Workhorse prompts produce higher outcomes once they transcend the common, anticipated reply. Two locations to engineer for that:
• In how the immediate is constructed: search for non-obvious angles — altering the sequence, reframing the reader’s position, ranging from an surprising enter.
• In what the immediate instructs AI to supply: construct in strikes that push previous AI’s defaults. Problem broadly accepted beliefs. Spotlight frequent assertions not supported by information. Floor counter-narratives. Reveal missed areas of significance. Counsel counter-intuitive approaches. Join seemingly unrelated insights. Draw from unconventional sources.
Decide the strikes that match the immediate. A developments piece advantages from counter-narratives. A method doc advantages from counter-intuitive approaches. A analysis immediate advantages from connecting unrelated insights. Not each immediate wants shock. Operational and categorization prompts usually shouldn’t have it. Attain for it when the consumer needs novelty or when typical output could be too simple to disregard.
4. Problem the framing solely when there’s a materially higher one. If the consumer’s framing is workable, construct inside it. Solely problem the framing when there’s a clearly higher various that will meaningfully change the output — not when there’s a defensible-but-different approach to slice the issue. Polished pedantry is worse than a sophisticated reply to the query as requested.
5. Search for strategies from adjoining disciplines. Frameworks from content material technique, web optimization, UX analysis, paid media, positioning, technique, or gross sales methodology can typically be repurposed in novel methods. Schema.org as a content material brainstorming instrument is one instance. Price surfacing when the consumer is trying to find a unique approach.
6. When describing what “good” appears like, present an instance as a substitute. Examples beat adjectives. A single earlier than/after pair teaches AI greater than three paragraphs describing the usual. If the consumer is describing desired output in adjectives, ask whether or not they have an instance to incorporate as a substitute.
7. When assigning AI a job, embrace a constraint and say what to keep away from. “Act as a world-class copywriter” does little. What works: position experience plus viewers plus constraint. “You’re reviewing this web page on behalf of a CFO who has 90 seconds, is skeptical of promoting claims, and must justify a purchase order to a board.” The constraint is the place the leverage is. Additionally embrace what to keep away from, not simply what to goal for Instance: “allergic to the phrase ‘unlock’” is extra helpful than “skilled tone.”
8. When a immediate evaluates or compares, require a dedicated judgment plus rationale. A 0–5 rating on specificity, a rating by impression, a move/fail on message match, paired with a brief rationale. Use a desk format. The judgment forces AI to take a place the consumer has to have interaction with. The rationale is the place the actionable perception lives. Keep away from imprecise suggestions.
9. For prime-stakes outputs, construct in a self-critique step. “After producing, charge this towards the factors above. What’s the weakest half? Rewrite that half.” AI is usually higher at evaluating its work than producing it on the primary attempt. Make suggestions from any discipline which will present insights. Not needed for each immediate — most helpful when the output might be printed or acted on instantly.
10. When requested to generate a immediate that outputs recordsdata, counsel lighter output codecs when doable. When a immediate’s output goes to DOCX or PDF, counsel producing a light-weight format like Markdown or HTML first as a substitute. Reserve heavy codecs for when the formatting itself is the deliverable.
After all, I used my very own immediate assistant to create these directions. We labored exhausting on it collectively. In case you learn it carefully, you’ll see the way it works. It creates a Immediate-Writing Assistant that does just a few very intelligent issues:
- It begins by asking questions (so objectives are clear and alternatives are found)
- It challenges assumptions, crosses disciplines and inverts pondering (so the consumer will get higher responses)
- It thinks about sensible inputs for prompts (so the consumer offers it the whole lot it wants)
- It builds prompts that embrace little self-evaluations and score programs and rationale (so the consumer thinks by way of the response)
- It builds prompts that generate environment friendly output codecs (so the consumer doesn’t use extra credit or carbon than needed)
You’ve pasted in these directions to your Claude Undertaking (or Customized GPT or Gemini Gem) and given it a reputation. A brand new member of your workforce is born.
Run a check immediate to verify the standard
We’ve examined this rigorously throughout many use circumstances and AI fashions for hours. However you’ll wish to check it your self.
Prepared on your first chat along with your new Immediate-Writing Assistant? Listed here are just a few dialog starters you possibly can attempt as check prompts. The use circumstances are infinite.
- “Write me a immediate that writes electronic mail topic traces and physique textual content when given a draft article.”
- “Write me a immediate that generates content material briefs when given a subject.”
- “Write me a immediate that edits articles for model voice and tone.”
- “Write me a immediate that helps me enhance underperforming touchdown pages.”
- “Write me a immediate that writes case examine intros.”
- “Write me a immediate that creates inside linking methods for an article.”
- “Write me a immediate that audits a competitor’s homepage for messaging weaknesses.”
Attempt one in every of these. The assistant will draft you a recent immediate. Evaluation it first or copy it straight into a brand new dialog to check the output. Right here’s what a check appears like.
I used the final instance from the record above (audit a competitor’s homepage) and generated two prompts: one with the Immediate-Writing Assistant, one with a plain unguided request to the identical AI. Then I ran each aspect by aspect. Evaluate:
On this check, the distinction is dramatic. The Immediate-Writing Assistant generated a immediate that had rather more element and insights.
- It began by asking me questions (“What determination are you driving towards?”)
- It advised inputs (“Are you able to present a screenshot?”)
- It created a scoring system (“Ten-second check outcomes: 2.5 out of 5”)
- It critiqued its personal output (“The weakest judgment on this audit was scoring the hero primarily based on a single screenshot of a rotating carousel.”)
After all, they’re each higher than a one-shot immediate written by hand “Audit this homepage for messaging weaknesses.” I attempted that too. It was a listing of generic suggestions. Ineffective.
What do you do subsequent?
- The output is sweet…
Save your new immediate! (extra on this under) - The output is just not nice…
Repair the problems within the new immediate. Edit it then hearth the check immediate once more. - The output is approach off…
There’s one thing structurally mistaken, ineffective or irrelevant. Coaching time. Return and inform your immediate assistant what’s not proper. Ask it to create a revision to its personal directions. Then copy the revised directions, paste them in, begin a brand new dialog and hearth the check immediate once more.
This extra diligent method to dialing in your Immediate-Writing Assistant includes checking it in a number of levels.
It’s an odd and summary approach to work. You’re utilizing an AI assistant to generate new prompts that you simply’re testing with check prompts in one other dialog, typically with the assistant auditing its personal output. Layers of language. It’s thoughts bending work with a number of ranges of modifying. It’s like that Leonardo DiCaprio film ‘Inception’. Nice film.
Preserve iterating till you might have an assistant you belief and a ultimate immediate value sharing. This method is how we’ve made lots of our best prompts.
Use your Immediate-Writing Assistant to enhance your previous prompts
You might have already got just a little repository of prompts you’re keen on. They’re excellent. However might they be higher? May the evaluation prompts present even deeper insights? May the content material prompts generate much more full articles?
There are two methods to make use of your new Immediate-Writing Assistant to audit and enhance them
- Give your unique immediate on to the Immediate-Writing Assistant
“Right here’s a immediate I’ve been working frequently. Audit it towards the ideas in your directions. Counsel particular enhancements that will make it sharper, extra reusable, or extra decision-driving. Don’t change its goal or the form of its output. Inform me which ideas every advised change is grounded in.” - Have the Immediate Writing-Assistant make a brand new immediate, then give each prompts again to the assistant
“I’m supplying you with two prompts that resolve the identical downside: one you generated, one I’ve been utilizing. Evaluate them throughout the ideas in your directions. The place does every one outperform the opposite? Then write a 3rd immediate that mixes the strongest parts of each.” - Run each prompts and examine the precise outputs.
“I ran each prompts on the identical enter. Right here’s what every one produced. Which output is stronger, and why? What does the weaker immediate have to match or beat the stronger one?”
The identical method works to audit the directions behind Customized GPTs, Claude Initiatives, Gemini Gems, and Claude Abilities. Paste within the directions, ask the Immediate-Writing Assistant to audit them towards its ideas, and you might spot gaps which were quietly limiting each dialog you’ve had with that assistant.
![]() |
Andrej Karpathy, OpenAI Cofounder“The most well liked new programming language is English.” |
Give your new reusable immediate a house
A workhorse immediate with no house will get rewritten from scratch by the subsequent one who wants it. Or by you, three weeks from now, when you possibly can’t bear in mind the place you saved it. Put your workhorse immediate within the secure. This might be a number of completely different locations:
- Reserve it in a shared immediate library that your workforce can simply entry
This might be Google Drive, Sharepoint, Notion or no matter else you employ to share docs. Then hyperlink to it out of your workforce coaching docs and SOPs (“customary working procedures”) - Make the brand new immediate its personal customized GPT, Claude Undertaking or Gemini Gem
If it’s used on a regular basis, no want to repeat and paste it each time. Simply make it one other new assistant. Give it a reputation “Orbit Electronic mail Editor” or “Cletus the Copywriting Helper.” So long as you might have a workforce AI account, everybody can entry it. - Claude Customers Solely: Make it a “Talent”
Inform your Immediate-Writing Assistant to adapt it and make it a Claude Talent. Abilities get invoked mechanically when the request matches the request. The consumer doesn’t have to go to the assistant.
Make to your Assistant ever-smarter (add data sources)
What would make your assistant smarter? What might you add to it that it might reference when generative prompts?
That relies on its goal. If it’s producing prompts that do modifying, add some writing requirements that it could possibly incorporate. If it’s producing prompts that make studies for purchasers, add your organization model guides or a previous report. If it’s producing prompts that do audits, add your present methodologies.
Our web page auditor has a number of attention-grabbing data sources that it could possibly reference.
Anytime, wherever. In case you spot one thing which may make any of your AI assistants smarter, put it aside, convert it to a light-weight markdown (.md) format, and add it to the data sources part of your assistant. Your assistant could also be a bit smarter eternally after.
I’ve many marketer mates who bristle at this type of AI delegation. They’ve legitimate causes, from labor market impression to environmental prices. However one of many strongest objections is nearly the way forward for work and its intrinsic worth to human goal. What in regards to the easy pleasure of creating issues? What about that?
The enjoyment of creativity is alive and nicely
In December of 1999, I stop my recruiting job to develop into an internet designer. I needed to construct web sites greater than anything on the earth. By 2001, my cofounder and I had began Orbit Media as an internet growth firm and we by no means appeared again. I haven’t personally designed an internet site in a few years, however I nonetheless bear in mind the enjoyment of creating issues.
And I nonetheless really feel that inventive pleasure, nevertheless it’s moved. I really feel one other form of pleasure, The enjoyment of constructing little instruments and seeing them work. It’s nonetheless inventive, nevertheless it’s a special form of creativity. Possibly it’s extra like the enjoyment {that a} programmer feels when their code does a great job.
And naturally, the best pleasure is efficiency. Understanding that you simply made a distinction for a model.
Advertising and marketing is necessary work. It fills the pipeline that drives offers that drive the income. That income creates job safety for the people that work at these manufacturers.
With this singular goal, we should contemplate any instrument that drives outcomes. AI is one in every of them. However the level was by no means AI or every other instrument. The purpose is the impression that the work has on the enterprise.
And now that we are able to make our personal little instruments, we are able to make one shift that secures the longer term for all of us.
Deal with outcomes, not outputs.
We’re entrepreneurs, not artists. We take satisfaction in our work, however we should additionally take satisfaction within the efficiency of our work. As a result of advertising issues.
Wait, extra sensible insights? Sure, please!
There may be extra the place this got here from…
The most effective content material from this weblog can be found multi function place – our ebook. Now on its seventh version.
Content material Chemistry, The Illustrated Handbook for Content material Advertising and marketing, is full of sensible suggestions, real-world examples, and skilled insights. A must-read for anybody seeking to construct a content material technique that drives actual enterprise impression. Try the opinions on Amazon.
The put up Cease writing prompts. Right here’s an AI assistant that does it for you. appeared first on Orbit Media Studios.












