Better AI Results Start With a Better Prompt

Key Takeaways

• Ambiguity Gap: your AI assistant fills unclear requests with its own best guess
• Goal Comes First: a clear goal is the only required part of a prompt
• Four Places to Check: Goal, Context, Source, and Expectations catch what's missing
• Final Check Matters: a good prompt still earns a human review before the output is used
• Consistency Is the Skill: knowing the framework differs from applying it every time

One person in marketing wraps up a stakeholder update, built from a campaign report and a few meeting notes, before lunch. Someone in a different department, working from a similar set of source material, is on a third rewrite of their own stakeholder update, and it's still not usable. Different departments, same kind of task, same deadline, but with completely different outcomes.

Two people can run the same type of request through their AI assistant and end up with completely different results. One gets a draft they can send with a light edit. The other gets something that technically answers the prompt but misses what the situation called for, and starts over. The gap between those two outcomes shows up across teams and tasks.

One consequential difference is how they ask their AI assistant for what they want. This difference follows a structure, and structure is something anyone can learn. Four elements give you a repeatable way to turn an underspecified request into one the assistant can act on more precisely.

Same AI Assistant, Different Results

Your AI assistant can only produce work based on the task instructions you provide. When a prompt doesn't include key information, the model infers what you're asking of it, and it can be wrong. When it returns an inconsistent result, the model's capability isn't always the issue.

Sometimes a request's vagueness leaves too much for the model to infer. Incorrect inferences can read as a tool problem when it's really a prompting one.

Research on the "jagged technological frontier" from Harvard and BCG found that having access to AI tools doesn't automatically translate into better performance.

Every prompt either answers these questions upfront or leaves the model to guess at them:

• What is the desired outcome?
• What situation or constraints matter?
• What information should the response draw from?
• What should the finished response look like?

Answer the ones that matter for the task, and your AI assistant has less left to infer on its own. Leave enough of them undefined, and it fills the gaps with its own best guess.

Prompts aren't the only variable that contributes to what an AI assistant produces. Source material quality and domain knowledge play a role too. But prompting is the variable someone has the most direct control over.

How someone uses AI matters just as much as having access to it in the first place. The four elements above reduce the ambiguity the model resolves on its own. Advice like "just talk to it naturally" sounds simple, but it skips over the structure that helps any AI assistant understand the request more precisely.

Talk Naturally, But Add Structure

The "just talk to it naturally" advice is incomplete, not wrong. Most AI assistants do understand plain, conversational language, and they aren't search engines that need keywords. But being conversational doesn't automatically mean being specific, and it's easy to assume otherwise.

Even Microsoft's own guidance draws this line, and OpenAI, Anthropic, and Google make a nearly identical case in their own documentation. A detailed, well-structured prompt gives any AI assistant more to work with than a vague one, even when both are phrased conversationally.

A prompt comparison illustrates this gap. "Write a blog post about sustainable practices in agriculture" is natural language. So is "Craft a 1,500-word blog post for a general audience interested in sustainability, focusing on the significance and benefits of sustainable agricultural practices like organic farming and agroforestry."

Only one of those prompts tells the assistant what to build. The other reads just as naturally, but the task inside it stays undefined. The gap comes down to a pattern simple enough to learn and repeat, independent of tone or politeness.

Build Every Prompt Around Four Elements: Goal, Context, Source, Expectations

Two prompts asking for the same outcome can still land in very different places, depending on how much an AI assistant fills in on its own. Four elements give you a way to identify what a request may be missing.

Microsoft's guidance for Copilot groups a prompt's components under four categories: Goal, Context, Source, and Expectations, or GCSE. The same four categories show up under different names in OpenAI's, Anthropic's, and Google's own guidance, making GCSE a useful shorthand no matter which AI assistant you use. A clear goal is the one non-negotiable part. The other three add specificity, each one sharpening a different part of the response when a task needs it.

Goal is the outcome you want, stated plainly. Stating it directs the assistant to the real problem, instead of a technically competent response to a question you weren't asking.

Context is the situation surrounding the request: who the response is for, and what's already in play, from prior events to active constraints. Including it grounds the response in the moment it needs to fit, rather than one that's correct in general but wrong for the situation.

Source anchors the response in specific information, whether that's a file, an email thread, or a dataset. Specifying it points the assistant to the evidence that matters most, instead of leaving it to guess and risk pulling from the wrong material.

Expectations describe what a usable response looks like, covering length, format, tone, and structure. Setting them upfront gives the assistant clearer direction on the form the response should take, cutting down on reshaping later.

Here's how the four elements together create a detailed prompt for a Q2 campaign performance request:

"Using the Q2 campaign report and the customer survey results (Source), identify the two channels that underperformed against target (Goal). Summarize the likely cause for each in a short paragraph, and format the whole response as a table with columns for channel, performance gap, and likely cause (Expectations). Keep it under 300 words, written for a marketing director who has five minutes before a leadership meeting (Context)."

The prompt reduces ambiguity because the details are explicit.

A well-structured prompt built around the GCSE elements makes it more likely any AI assistant understands what you're asking for. But the user's own judgment still determines whether the output is accurate and ready to use.

For consequential work, that means checking important claims and numbers against the source material and confirming the response still fits the audience and business context, including whether the assistant introduced assumptions the source doesn't support. These elements apply no matter who's writing the prompt, or which AI assistant is on the receiving end.

See the Framework Applied Across Six Different Roles

The specific information behind Source, Goal, Context, and Expectations changes from one department to the next, but the four elements themselves don't. Recognizing the elements across six different job roles makes it easier to apply them to new, job-specific tasks.

Each pair below starts from the same task, contrasting a weak prompt with a strong one built from all four GCSE elements. In practice, a prompt might only need one or two elements spelled out.

Marketing: Quarterly Channel Review

Weak: "Write a summary of our social media performance."

Strong: "Using the March social media report (Source), identify which two platforms drove the most engagement (Goal). Write three bullet points explaining why, for a team meeting Thursday (Context). Keep each bullet under 25 words (Expectations)."

Finance: Explaining Budget Shifts

Weak: "Explain the budget variance for Q1."

Strong: "Using the Q1 actuals and budget file (Source), identify the three line items with the largest variance from plan (Goal). Write a short explanation for each, for the CFO ahead of a board update (Context). Format as a table with columns for line item, variance amount, and cause (Expectations)."

HR: Eliminating Policy Confusion

Weak: "Draft an email about the new PTO policy."

Strong: "Using the updated PTO policy document (Source), draft an email announcing the change to all employees (Goal). This rolls out company-wide next month and past changes have caused confusion about accrual (Context). Keep it under 200 words, plain language, no legal jargon (Expectations)."

Sales: Follow-up Vendor Outreach

Weak: "Write a follow-up email after our call."

Strong: "Using my notes from today's call with Acme Corp (Source), draft a follow-up email recapping their two stated priorities and proposing next steps (Goal). They're evaluating three vendors and want a decision by Friday (Context). Keep it under 150 words, professional but not stiff (Expectations)."

IT: Clarifying Password Reset Steps

Weak: "Write instructions for resetting a password."

Strong: "Using the current password policy (Source), write step-by-step instructions for resetting a network password (Goal). This is for employees with no technical background, several of whom submitted tickets this week (Context). Number each step, five steps max, no jargon (Expectations)."

Leadership: Refocusing and Lessons Learned

Weak: "Summarize this quarter for the team."

Strong: "Using the Q3 performance dashboard (Source), summarize what went well and what needs attention next quarter (Goal). This is for an all-hands where morale has been mixed after a tough launch (Context). Keep it to one page, direct but encouraging in tone (Expectations)."

Across all six roles, making the relevant details explicit cuts down the kind of ambiguity that leads to avoidable rework. Doing that once is straightforward. Doing it consistently across all tasks is a skill worth building.

Why Better Prompting Needs to Become a Repeatable Skill

Getting one great response is useful. Being able to add structure to prompts consistently, across different tasks, is a skill. Building that skill means relying less on trial and error and more on a repeatable process. This distinction matters more as AI assistants move from a single use case into everyday workflows across a team.

This gap between one good result and consistent results shows up across the broader workforce too. According to Docebo's 2026 AI Readiness Gap report, 85% of employees say the training they've received doesn't help them use AI in their role.

A separate DataCamp and YouGov survey found that while 72% of enterprise leaders now consider AI literacy essential, 59% still report a real skills gap inside their organization.

Both numbers point at a broader pattern, not prompting specifically. Knowing a skill matters and being able to apply it under real working conditions are two different things, and GCSE is no exception. Closing the distance requires practice and repetition, not another framework.

Close the Gap Before Your Next AI Prompt

Two people, two departments, the same kind of task, and the same deadline. Access alone didn't explain the difference. One had a repeatable way to tell their AI assistant what good work required. The other was still guessing at it prompt by prompt.

GCSE is simple to explain in a few minutes. The skill is recognizing what each new task needs and making those requirements explicit before the assistant infers them. Applying it consistently gives any AI assistant clearer instructions and reduces the rework caused by ambiguity.

Spelling those requirements out up front can prevent the rounds of rework caused by an underspecified request, even though it won't eliminate iteration entirely.

New Horizons helps organizations build that consistency, giving employees a structured way to practice GCSE across real business scenarios until it becomes automatic.

Would every person on your team reach for that same structure on a request they didn't see coming?

Explore New Horizons' AI prompting training and learn how to build consistency before your next deadline.