Lesson 4 / 18Module 2. Output ControlDetailed lesson
Academy/Prompt Engineering/Lesson 4. Constraints and quality criteria
Intermediate17 min

Lesson 4. Constraints and quality criteria

Constraints don't make the answer poorer — they bring it closer to the result you actually need. In this lesson you'll see which constraints are useful and which ones stifle the prompt.

Topic breakdown

When users tell the model 'answer freely', they often get output that's too long, too generic, or uncontrollable. This happens because the model wasn't given boundaries to work within.

Constraints are the prompt's internal control system. Length, tone, required blocks, restrictions, and quality criteria make the result noticeably more useful.

This is especially important for ads, support responses, SEO descriptions, and any repeatable content where you need not just text, but a controllable quality standard.

What you'll learn

  • control length and tone
  • properly add restrictions to a prompt
  • check results using quality criteria
  • separate required elements from undesirable patterns

Lesson plan

Why constraints are needed

They don't slow down the model — they help it work within the right range: shorter, more precise, and closer to your task.

Required blocks and restrictions

A strong prompt separately describes what must be included and what absolutely should not appear in the result.

Internal review by criteria

If you ask the model not only to write text but also to check it against criteria, quality becomes more consistent.

When constraints become too many

If constraints contradict each other or don't affect the result, they only confuse the model and reduce output quality.

Weak vs strong prompt

Weak prompt

Write a short ad copy for our service. Make it really good.

Strong prompt

You are a copywriter. Write a Telegram ad for an online accounting service. Audience: small business owners. Text — up to 120 words, tone simple and confident. Required blocks: headline, 2 benefits, CTA. Restrictions: clickbait, jargon, unsubstantiated promises. At the end, check the response for clarity, brevity, and CTA strength.

The strong prompt sets boundaries, structure, and verification criteria, so the result is closer to a working standard rather than a random text.

Ready prompt template

Copy and adapt
You are an experienced copywriter. Product: [product or service]. Audience: [segment]. Goal: [outcome]. Write the text in simple language, no longer than 140 words. Avoid overly formal or artificial phrasing. Required blocks: headline, main benefit, CTA. Restrictions: empty promises, unsubstantiated superlatives, mixing with jargon. At the end, self-check your answer against 4 criteria: clarity, brevity, audience fit, CTA strength.

Why it works

A length constraint keeps the answer from sprawling and forces the model to concentrate on what's important.

Tone helps adapt the text to the audience: simple, confident, friendly, or expert.

Restrictions cut off undesirable directions: clickbait, empty promises, unnecessary jargon.

Quality criteria transform the response from just text into a verifiable work product.

Practice

  • Take one prompt for a Telegram post or ad.
  • Add 3 constraints: length, tone, and a ban on undesirable wording.
  • Write 3 quality criteria: clarity, CTA strength, audience fit.
  • After the model responds, check it against those criteria and improve the prompt once.

Mini-project

Mini-project: controlled ad prompt

Create a repeatable prompt for one product or service that consistently produces a short and verifiable ad text.

Tasks

  • Define the product, audience, and goal.
  • Add 3 required blocks and 3 restrictions.
  • Write 4 quality criteria.
  • Run the prompt and evaluate the result against criteria.

Deliverables

  • 1 controlled prompt template
  • 1 AI response
  • brief evaluation by criteria and an improved version

Checklist

Is there a length constraint?
Is the tone explicitly set?
Are required blocks specified separately?
Are prohibited elements listed?
Are there criteria for checking the result?

Common mistakes

  • adding too many constraints and stifling the prompt
  • setting contradictory criteria
  • listing only restrictions without required blocks
  • using overly vague criteria like 'make it powerful'

Lesson FAQ

Will too many constraints make the result worse?

Not necessarily. If constraints are useful and don't contradict each other, the result usually improves. The issue isn't quantity but quality of constraints.

Should criteria be written at the end of the prompt?

Not mandatory, but a separate criteria block makes the prompt clearer for both the model and the human, especially if you want to build in a review step.

Next step

How to set constraints and criteria in a prompt? Lesson 4 | Prompter Academy