How to Get What You Want Out of AI with Fewer Back and Forths

I have watched very smart people get very mediocre results from AI, and it almost never has anything to do with the tool. It has to do with the prompt.

Here’s what I mean. Asking AI to “write a blog post about content marketing” is a bit like hiring a brilliant new general contractor and handing them a sticky note that says “build something.” They have what they need to start kind of. But what they produce is going to reflect the ambiguity of the instructions, not the ceiling of their capabilities. The output will be generic, because the input was generic.

We like AI at Wheels Up. The team uses it a lot. Parts of our workflow are built around it, we recommend it to clients, and we’ve done a lot of in-the-trenches evaluation of what does and doesn’t work. One thing we’ve learned? The people who get the most out of it aren’t the people with the most technical AI knowledge. They’re the people who have gotten good at communicating with AI tools. 

The good news is that it’s a learnable skill, and most people are closer to it than they think.

Here’s what more effective AI use looks like in practice, with the specific techniques that can cut your back-and-forths in half.

Think of AI as a Very Smart Contractor Who Just Started Today

The single most useful mental model for working with AI is this: Imagine you just hired the most capable, well-read, fast-working contractor you’ve ever met, and today is their first day. They know an enormous amount about the world. But they know nothing about your world.

They don’t know your company, your audience, your competitors, your voice, your previous work, your preferences, or the context behind your request. They want to do a great job. They are genuinely trying to help. But without the right information, they will default to producing something reasonable and general, because that’s the safest interpretation of a vague brief.

This reframe matters because it changes what you do before you type anything. The question to ask yourself is: What would I need to tell a smart, brand-new colleague? The answer to that question is your prompt.

Give It a Role, a Reader, and a Reason

Most prompts that underdeliver are missing at least one of three things: Who the AI is writing as, who the AI is writing to, what the AI is writing for. When you include all three, the quality of what comes back improves substantially.

Give it a role. Telling the AI what perspective to take calibrates the register, vocabulary, and depth of the response. “You are a senior B2B content strategist with 15 years of experience working with tech startups” produces a different output than leaving the AI to decide what kind of expert it should be. 

Define the reader. AI will write for the most general possible audience unless you tell it otherwise. “The reader is a VP of Operations at a Series B SaaS company who is skeptical of new tools and short on time” will shape the tone, the examples, the assumed context, and the level of sophistication in ways that “business audience” just won’t.

Give it a reason. What is this piece of content supposed to do? Is it building awareness? Convincing someone who is already evaluating options? Opening a sales conversation? Giving the AI a clear job to accomplish, not just a format to fill, means the output will be structured around a purpose rather than around just covering the topic.

A prompt that includes all three might look like this:

“You are a senior B2B content strategist. Write a 600-word blog introduction for a VP of Operations at a mid-market SaaS company who is evaluating workflow automation tools for the first time. The goal is to make her feel understood and curious enough to keep reading.” That prompt will produce something meaningfully different from “Write a blog intro about workflow automation.”

Include Context (It’s the Variable That Changes Everything)

If there’s one thing that separates the people who get great AI output from the people who spend twenty minutes editing mediocre output, it’s context. The more relevant background you give, the more precisely the AI can target its response.

What counts as useful context depends on the task, but some categories are almost always worth including: 

• The URL for your company website

• A link to your product’s landing page 

• A thorough description of your audience including their level of familiarity with the topic

• Your messaging & positioning framework

• A description of your brand voice (ideally, with an example)

• A list of words or phrases that are on-brand

• A list of words and phrases that are off-brand

• Examples of work you want the output to resemble

• Your company’s positioning on the topic or the specific argument you’re trying to make

• Any constraints on length or format

• Any prior work the AI should build on rather than reinvent

The most powerful form of context is an example. If you have a piece of content that represents what “good” looks like for your brand, upload or paste it in and say “write in this style.” AI will pattern-match from an example far more accurately than it will interpret a description. “Conversational but professional, not too casual, avoid jargon” is genuinely hard for a model to calibrate on its own. A single well-chosen example makes that calibration almost automatic.

This is especially true for voice. If you want AI output that sounds like you, the most reliable path is to show it what you sound like, not describe it.

Tell the Tool What Format You Want Before It Decides for You

AI has default preferences, and they are not always your preferences. Left to its own devices, AI tends to produce short, choppy sentences and a lot of bullet point lists. It likes lengthier headers and shorter paragraphs. And in its writing, it tends to cover all the bases rather than making a decisive argument. These are all reasonable defaults for an audience of everyone. They are probably not right for your specific use case.

Specifying format upfront is one of the easiest ways to cut editing time dramatically. A few of the most useful format instructions to have in your back pocket: “Write in paragraphs, not bullet points.” “Keep this under 900 words.” “Use a conversational tone and active voice.” “Avoid em dashes.” “Open with a specific scenario, not a general statement.” “Do not use the word ‘leverage.”

That last category, the specific prohibitions, is worth building a list of for your brand. Most companies and writers have a set of words, sentence structures, or stylistic tics they want to avoid. Telling the AI explicitly is faster than editing them out after the fact. And yes, AI will (mostly) follow those instructions if you give them.

Ask for One Thing at a Time

This is the mistake that creates the most unnecessary back-and-forths, and it’s the one many people don’t realize they’re making.

When you ask AI to “write a landing page headline, three subheads, and a CTA, with a tone that’s confident but not arrogant, aimed at CTOs who are skeptical of vendors, and make sure it connects to our positioning around speed” in a single prompt, you have given it a lot to balance simultaneously. Some of it will land. Some of it won’t. And when you edit, you won’t always know which instruction caused the problem.

The faster workflow is often sequential. Ask for the headline first. Evaluate it. Tell the AI what’s working and what isn’t. Then ask for the subheads with that feedback in mind. Each step in the conversation can build on the last, and the cumulative output is usually much stronger than a single complex prompt could produce.

Think of it as a brief, not a batch order. Briefs have stages. Batch orders get filled all at once and sometimes return the wrong thing.

Tell It When It’s Wrong, and Be Specific About Why

One of the most underleveraged behaviors in AI conversations is giving useful feedback. Most people look at an output that doesn’t quite work, delete it, and start over with a new prompt. That approach throws away the accumulated context of the conversation.

Instead, tell the AI what specifically isn’t working. “This is too formal” is useful. “This sounds like a press release and I want it to sound like a LinkedIn post from a founder” is more useful. “The second paragraph loses the thread” is useful. “This doesn’t address the real objection, which is that our customers have tried tools like this before and been burned” is more useful.

The AI is not starting from zero every time you respond. It has the full conversation in context. A specific, honest critique followed by a clear direction is often the fastest path from a mediocre first draft to something much closer to publishable.

Treat It as a Collaborator, Not a Vending Machine

The final reframe, and maybe the most important one: AI works best when you treat it as a thinking partner rather than a production machine. A vending machine accepts input and dispenses output. You wouldn’t ask a vending machine to reconsider. You wouldn’t give it feedback. You’d push the button and take what came out.

AI is more useful than that. Ask it to argue the other side of a position you’re developing. Ask it what you’re missing. Ask it to evaluate your draft before you edit it. Ask it what a skeptical reader would push back on. These uses require no technical skill. They just require thinking of the tool as something worth having a real conversation with.

The people who get the most out of AI are the ones who engage with it that way, iteratively, specifically, and with genuine curiosity about what it can contribute. The output reflects the quality of the collaboration.

Getting the Most Out of Your Work with AI

Better AI output is almost always downstream of better input. The tool is generally capable of more than most people are asking it for. Closing that gap doesn’t require technical expertise or a prompt engineering certification. It requires a little more intention before you hit send: a clear role, a defined reader, a specific purpose, useful context, and an honest feedback loop when the first draft isn’t right.

If you’re trying to figure out where AI fits in your marketing workflow, or you want a second set of eyes on how your team is using it, that’s exactly the kind of conversation we’re having with clients right now at Wheels Up Collective. Reach out here whenever you’re ready.