Instruct your AI well, it will treat you better

Today, OpenAI released its Model 4.1, which has already demonstrated that it is considerably more adept at following instructions than the previous Model 4.0.

These rapid advances highlight a critical area of AI use: giving effective and precise instructions. Modern AI systems are better than ever at carrying out the commands of their “masters,” yet this means users must provide instructions that are both accurate and comprehensive. This blog post explores why effective prompts (or briefings) matter, and provides practical guidelines for drafting them.

1. Why Good Instructions Matter

An AI model, no matter how sophisticated, still depends on the guidance and context provided by human users. If that guidance is vague, incomplete, or contradictory, the model’s outputs will often reflect those flaws. On the other hand, when instructions are clear and consistent, the AI can reach far more reliable and insightful conclusions.

For instance, you might be seeking an in-depth analysis of a complex topic or a concise, bullet-point summary. If you fail to specify the depth or style of response, you could end up with an overlong dissertation or an overly brief outline—neither of which meets your needs. Good prompts help AI to understand precisely what you want, leading to more effective results.

Good delegation is not passing the task, it is passing the clarity.

2. Top Five Recommendations for Effective AI Prompts

2.1 Be Specific and Clear

When working with data, vague prompts such as “analyse this dataset” will rarely lead to useful outputs. Always be explicit about what you want the AI to focus on.

For example, a revenue growth manager at a beverage company should not ask, “What can you tell me about our promotions?” Instead, they might say: “Please compare the uplift from our most recent promotion on Classic Lager to the previous three promotions. Identify which customers, channels, and pack sizes delivered the highest incremental volume.”

In professional services, an analyst at a consulting firm might avoid generalities like “summarise client performance data” and instead request: “Analyse revenue and margin trends for our top 10 clients over the last 6 months. Highlight any significant changes in project mix or billing rates.”

The more precise the request, the more accurate and relevant the AI’s output will be.

2.2 Provide Adequate Context

AI models are not mind readers. The quality of their analysis depends heavily on the context you provide. This includes the type of business, the nature of the dataset, and the intended use of the output.

An FMCG user might include context such as: “This dataset contains weekly sales figures by retailer and SKU for our World Beer portfolio in the UK. We are preparing a quarterly performance review and need to understand which retailers are over- or under-performing relative to the same period last year.”

A professional services firm might write: “This dataset covers time spent by consultants across client accounts. We are trying to understand whether lower client profitability is due to over-servicing. Please identify accounts with the highest hours-to-revenue ratio.”

By setting the scene, you enable the AI to deliver a response that is aligned with business goals and constraints.

2.3 Structure Your Request

Structured prompts help guide the AI to respond in a clear and logical way. Especially when analysing data, breaking the request into steps allows the model to address each part methodically.

For example, a structured prompt in FMCG might be: “Using the promotional sales dataset provided:

1. Calculate the average uplift per retailer during the last promotion.

2. Compare this to the 3-month pre-promotion baseline.

3. Identify which SKUs underperformed, and suggest possible reasons.”**

In professional services: “Please structure your analysis as follows:

1. Segment clients by industry.

2. Compare average project duration and billing rates across segments.

3. Highlight which segments show the strongest growth.”**

This approach makes the AI’s output easier to understand and ensures that nothing important is missed.

2.4 Limit or Expand the Scope When Necessary

Define the level of detail you want. Do you need a quick snapshot or a deep dive? Should the AI provide raw metrics, insights, or recommendations?

A commercial manager might say: “In no more than 300 words, summarise how sales of Concentrated Detergents evolved in Q1 across Tesco, Asda, and Sainsbury’s. Focus on volume changes, not value.”

Or, when more depth is needed: “Generate a detailed report (approx. 1,000 words) comparing the effectiveness of price promotions by brand and region. Include charts and call out statistically significant differences.”

Setting clear boundaries helps the AI know when to stop and how deep to go.

2.5 Test and Refine

The first instructions you try out are rarely the best. Review them critically. Did the AI address your priorities? Did it misinterpret any part of your question?

If not, revise your prompt. A vague request like: “Show me insights from this chart” can be rephrased as: “Based on this chart showing weekly sales of flavoured water in the UK, identify the top three weeks in terms of volume and explain what might have caused the spikes.”

This refinement process is part of the workflow at MetaMarketing. We encourage users to experiment and iterate until they reach a level of clarity that consistently delivers high-quality results.

3. Working with Quantitative Data

At MetaMarketing, we specialise in generating insights from quantitative data. When working with numbers, the clarity of instructions is arguably even more important than when generating text. With numerical data, it is difficult to verify whether the AI has identified all relevant insights. Vague prompts such as “analyse the data” are unlikely to produce satisfactory results. Instructions need to be precise—clearly stating what to analyse and which causal relationships to explore. For example: “Analyse how sales are evolving, and identify which brands and retailers are gaining or losing market share.”

Large Language Models (LLMs) have their limitations. Analysing causality often requires different types of models. For instance, answering a question like “Are my sales affected by promotions and pricing?” calls for an AI model trained on algorithms that can assess causal relationships between sales and influencing variables such as price and promotional activity.

Style matters too. When prompting an LLM, it is important to define the desired output format and writing style. Disappointment with AI-generated text often stems not from poor content, but from a mismatch in tone, structure, or level of formality.

Do’s and Do Not’s

• Do include essential keywords and details.

• Do re-check your prompt for consistency before submitting it.

• Do keep prompts concise and targeted.

• Do Not rely on excessively broad questions, such as “Tell me about marketing.”

• Do Not expect AI to guess or fill in critical information you have left out.

• Do Not bury your crucial requirements under paragraphs of unnecessary text.

4. Preset Analyses: Reducing Guesswork and Raising Standards

At MetaMarketing, we recognise that not every user has the time—or the confidence—to experiment with prompts. That is why our solution includes preset analyses: pre-defined instructions that have been fine-tuned, tested, and optimised for common analytical needs in FMCG and professional services.

Preset analyses allow users to run high-quality analyses on their own data in just a few clicks, without needing to formulate or refine complex prompts. For example, a revenue growth manager might select “Promotion ROI Analysis” or “Price Sensitivity by Retailer” from a preset list, instantly generating a reliable, standardised report.

This approach offers two key advantages:

4.1 Faster, More Reliable Results

Because preset prompts are designed and tested by experts, they avoid the pitfalls of vague or poorly phrased instructions. The output is consistent, actionable, and aligned with best practices—without the user having to trial different formulations.

4.2 Internal Standardisation

Preset analyses help organisations establish a consistent analytical language and reporting structure across teams. Instead of each team member defining their own analysis style, they can rely on standardised instructions that reflect company-wide methods. This improves comparability of reports, simplifies quality control, and accelerates decision-making.

For instance, a professional services firm might use preset analyses for “Client Profitability Breakdown” or “Time Spent vs. Budget by Project Phase,” ensuring that all analysts approach these topics in the same way, regardless of their individual background or preferences.

Preset analyses reduce friction, increase quality, and empower every user—regardless of experience level—to get the most from their data.

5. Conclusion

The rollout of powerful AI models such as OpenAI’s 4.1 further underscores the significance of providing clear, well-structured, and complete instructions. Although these systems are evolving rapidly to become more intuitive and flexible, their success still depends on humans crafting purposeful prompts that guide them towards useful results.

By putting the top five recommendations into practice—being specific, supplying context, structuring your requests, controlling scope, and iterating on responses—you will improve both the efficiency and accuracy of AI outputs. When dealing with numerical data, take extra care to specify precisely what the AI should examine, as demonstrated by MetaMarketing’s approach. As you experiment with this new generation of models, do not be afraid to share your prompts and findings with others; together, we can continue to refine and evolve our approaches to AI-driven communication.