We launch dozens of models, throw prompts, wait for miracles… and get chaos in responses, plotless hallucinations of artificial intelligence, and jokes that are funny only to AI. And this even seems “normal” until you see how you can work differently: faster, cleaner, structured.

This article explains how to tame AI to produce logical, accurate results. About principles that save hours and nerves. And about how to make AI a full-fledged tool.

This material is based on a webinar hosted by Sasha, the CEO of newage. You can watch the full recording below:

Use several models, not just one

One model means one way of thinking, one set of biases, and one way of interpreting your words. Because of this, AI quickly becomes “blurred”: it starts repeating the same phrases, giving the same advice, and even missing nuances that other models would notice.

Why work with multiple models?

• More accuracy. Responses can be cross-verified.
• More ideas. Models think differently: one gives a structured plan, another an unconventional approach, and the third something in between.
• More context. Comparing results lets you see the topic more broadly and grasp the real picture faster.

In our team, we have a favorite “test prompt” for checking any model:

“Tell me the football players who, during their careers, played both for Karpaty Lviv and for Shakhtar Donetsk.”

Sounds simple, but it’s a very complex task inside: it requires finding data, comparing, filtering, and delivering the truth, not fantasies.

And what does practice show?

• Gemini, GPT, or Grok usually give normal results: Fedetsky, Kobin, Gladky.
• Other models sometimes “create art”: they add players who never played there, but sound convincing.

That’s why tools like OverallGPT appear. It sends one prompt to multiple models at once and collects responses in one window. And in a minute, you see the results and can compare them.

Personalization of AI: the simplest setup that changes everything

Most people work with AI “by default.” Then they wonder why the response turns out too formal, too creative, or just weird. Personalization is about how AI stops being an abstract “assistant” and starts working as your assistant.

Why do this?

• Tone and style. You immediately set how the model should sound: friendly, professional, with irony, without fluff.
• Level of accuracy. You can demand reliability, sources, dates, and links to documentation.
• Depth of responses. AI can give short summaries or long structures as you wish.
• Context of your role. Marketer, PM, analyst, copywriter. Depending on the role, each gets a different type of response.

After this, AI stops improvising and starts operating within the framework you set yourself.

How to enable personalization in ChatGPT

1. Click Settings.
2. Go to the Personalization section.
3. Fill in the fields:
   • who you are (role, field, level),
   • how you like AI to work with you,
   • writing style,
   • what you don’t like (too long responses, made-up facts, etc.),
   • which sources AI should use.

After this, every AI response is already tailored to you, regardless of which chat you’re writing in. But even with personalization, there are queries where it’s better to specify the context directly in the prompt: when the task is unusual or one-off, when working on a specific project, when exact parameters are needed (budget, audience, market), when AI analyzes analytics or ad data.

The formula is simple:

“My name is X, I’m working on Y. I need N. Use style A, sources B, accuracy level C.”

It takes 10 seconds but saves hours and prevents weird responses. After personalization, AI becomes a conversational partner who knows your style and standards.

Two-way work: AI should ask questions, clarify, and even argue

Most people work with AI like this: “Hi, give me…” followed by a wish formulated in the hope that the system will guess what you meant. The problem? AI is polite. Too polite. If you give a vague task, it will do everything possible to complete it… even if you end up with a set of made-up facts.

But there is a short phrase that literally elevates the level of interaction:

“Ask clarifying questions that you lack to properly complete the task.”

After this, AI stops being a submissive “executor” and turns into a normal colleague who: asks about goals, clarifies context, checks hypotheses, identifies contradictions, and stops you if the task is illogical.

When it’s useful for AI to argue

Sometimes the best result is not an answer, but resistance. Smart, reasoned, respectful. AI can (and should!) object if:

• the task contradicts your goals;
• you chose the wrong channel or tool;
• your request has an obvious logical gap.
• the data is insufficient or incorrect;
• there’s a better path you haven’t thought of.

Such moments save dozens of hours and a lot of nerves. A properly configured AI starts really helping, not submissively producing text for the sake of text.

And here, one simple, almost magical phrase works: “Argue with me if necessary.”

This switches AI into conversation mode. The model starts thinking critically: questions weak ideas, suggests alternatives, catches contradictions, and often literally saves you from bad decisions.

Break large tasks into small steps

There’s a persistent myth: if you give AI a large, detailed, epic prompt spanning three screens, it will produce a genius result. Spoiler: no. For LLMs, this looks like “here’s a dissertation, make me something.” In the end, AI either gets lost or invents extras. Models don’t like overly large queries for several reasons:

• you have multiple tasks in one query right away, and AI simply doesn’t know what’s main;
• part of the important information gets lost under a pile of unnecessary details;
• the system starts “filling in” where you didn’t specify;
• the response becomes either too general or just random.

But the question arises: “how to do this right?” And we have a simple answer: instead of one monolithic prompt, make mini-tasks:

1. Give the first part of the task. For example: “Create a structure for the future presentation.”
2. Get the response. Check the logic, style, and direction.
3. Give clarifications. “OK, detail blocks #2 and #3, focus on e-commerce examples.”
4. Get the next steps.
5. Repeat the cycle until the result is complete.

This is faster, more accurate, and safer. This principle is simple but works flawlessly: AI thinks best when you think with it in small steps.

Work in workspaces

Every time you open a new dialog and write: “Hi, make me a campaign structure…”, the AI thinks: “I see you for the first time, but I’ll generate something.” Instructions, context, and favorite style disappear. That’s why response quality fluctuates, and you waste time repeating what you’ve already said.

Document environments like NotebookLM, ChatGPT Projects, or any other workspace systems give AI long-term context that it doesn’t forget after every message. It works like your own “notebook” where everything lives:

• your files, briefs;
• rules for tone, structure, format;
• task history and interim conclusions;
• your previous edits (extremely valuable for the model);
• links to sources you trust.

As a result, AI doesn’t just respond—it remembers who it’s working for and why. It keeps context, your style, and previous tasks in memory and builds responses not from scratch, but based on the entire project. This saves hours and increases response accuracy.

Prompt Optimization

There’s a temptation to write every query “from scratch.” Then we wonder why it turns out genius one time, and another time, AI responds completely off-topic. The solution is simple and very pleasant: don’t invent prompts, optimize them.

And for this, there are prompt optimizers. It’s like an editor that takes your thoughts and turns them into a clear task that AI can complete without unnecessary surprises. They are needed to:

• remove ambiguity;
• add clarifications you didn’t even think of;
• structure the task;
• increase the accuracy and reproducibility of results.

And the best part—they save time and effort. Because instead of AI “guessing” the logic, you give it a perfectly formulated task.

Where to get good templates

There are two types of sources:

1. newage. Prompt Base. This is our “battle-tested set” of prompts that really work in projects: analytics, creatives, audiences, research, strategy, and technical queries.
2. Third-party libraries. Prompt Perfect, FlowGPT, Awesome GPT Prompts and others.

AI can also write a prompt for you. And yes, it works better than it seems. The focus is very simple: you write the context to AI → AI writes you the perfect prompt with all clarifications, questions, structure, and requirements. Example formulation:

“Please create a prompt that will help me:
– prepare a presentation structure;
– show key cases;
– collect risks and recommendations.
Add questions you need for the maximum result.”

And what happens next? AI does the rough work for you: it forms a detailed instruction that you then take to a separate chat and get the result.

Content and Emails

AI excels at structuring, polishing, shortening, and making text clearer, but when it writes everything from scratch, it comes out somewhere between an official speech and an obituary. So the rule is simple: you first, then AI. A live voice is always needed for the text to sound natural. Give AI your draft text, and get a clean, logical, and human version.

In this process, AI is not the author or copywriter, but a polisher. It removes fluff, aligns structure, and eliminates complex sentences, but doesn’t invent your intonation for you.

There’s also a small trick that saves more time than it seems: ask AI to pick 2–3 relevant emojis for a post or email. Not “fire-fire-rocket,” but truly appropriate symbols that support the tone. A micro-decoration that makes the text visually warmer.

Privacy and security: don’t give away the extra

AI is polite as long as you don’t give it the extra. But upload a client dataset or campaign budget to the chat once, and you’re playing the lottery. That’s why the basic rule is: never give AI anything you wouldn’t want to see in the public domain.

Even if the model promises privacy, don’t trust it unconditionally. There are parts of work that are safer to anonymize. Simple methods suffice here:

• remove brand names, replacing them with neutral words or just letters;
• scale budgets or KPIs (e.g., multiply all numbers by 3.14 or 354);
• delete unique identifiers that could link data to a specific project or person.

In most cases, this is enough to get a useful analysis without risk.

Additional resources and final AI rules

To grow faster and keep your AI skills sharp, regularly feed your mind with quality content. The newage. team recommends two resources:

• Flow podcast about the AI,
• a course from Fractal Group, which covers solid theory and systematizes practice.

Also, maintain your own “AI shelf” in bookmarks: tools for analysis, prompt bases, generators, and automation services. The AI world grows faster than any industry, so information hygiene becomes essential.

Finally, AI is a process. Its effectiveness depends not on which button you press, but on how you work with the model: provide context, clarify, argue, break tasks into steps, and check every fact. In tandem with a human, AI fully reveals itself: accelerates, structures, suggests, and sometimes saves from mistakes.

FAQ

Do you need to study prompt engineering to work effectively with AI?

Yes and no. Basic skills suffice: clearly formulate the task, provide context, specify response format. This significantly improves results.

Which model is best for work?

None. And that’s great news. Combining models works much better as it covers each other’s blind spots. The key is comparing results to minimize biases and errors.

Can you rely on AI for analytics?

Yes, but only with verified data: demand sources, update dates, and quotes. AI should not invent information or metrics.

Why does AI sometimes give inaccurate or incomplete responses?

Because it lacks context. Provide more details and get a more accurate result.

Should you use AI for emails and content?

Yes, but as an editor, not the main author. AI helps structure text, improve clarity, and avoid errors, but content and tone are best shaped yourself.