How Specificity and Conciseness Affect Prompt Quality
Two Qualities That Seem to Conflict Specificity means a prompt states exactly what is needed: the task, the limits, the audience, the format, and how to handle unusual cases.
Conciseness means a prompt contains no unnecessary words: every sentence contributes something the model needs.
At first, these seem to pull in opposite directions. Being specific adds detail; being concise removes text. In practice, they work together:
Specificity is about including the right information. Conciseness is about removing everything else.
A prompt can be long and still concise if every part carries necessary information. A prompt can be short and still vague if it leaves important decisions to the model. The goal is a prompt that is as specific as the task requires and no longer than that.
The Four Combinations Concise Wordy Specific ✅ Ideal: every word carries needed information ⚠️ Works, but important rules are buried in filler Vague ❌ Short but leaves key decisions to the model ❌ Long and still unclear Example of Each Vague and concise:
Write a bio. Vague and wordy:
Hello! I hope you're doing well. I was wondering if you could possibly help me out with something. I need to write a bio, and I'd really like it to be good, you know, something that represents me well and looks nice. Thanks so much in advance for your help, I really appreciate it! Specific and wordy:
Hello! So I need a bio, and I want it to be for my LinkedIn profile, which is a professional networking site, so it should be professional. I'm a Django developer, I've been doing that for about 3 years, and I have built e-commerce sites. I'd like it to be written in first person, like "I am...", and it shouldn't be too long, maybe around 80 words or so, roughly, if possible. Specific and concise:
Write an 80-word LinkedIn bio in first person.
About me: Django developer, 3 years of experience, builds e-commerce websites. Tone: professional and approachable. Example Explanation The last two prompts contain the same essential information: platform, role, experience, specialization, point of view, length, and tone. The concise version expresses it in about a third of the words, using labels instead of sentences. Nothing useful was lost; only filler was removed.
Why Specificity Improves Results It Reduces the Space of Acceptable Answers Every unspecified detail is a decision the model makes for you. "Write a bio" leaves open the platform, length, point of view, tone, and content. Each added detail removes many possible outputs that would not suit you.
It Makes Results Consistent When a prompt leaves decisions open, the model may decide differently on different runs (Lesson 2). Specific prompts produce more consistent outputs because fewer decisions remain.
It Makes Results Checkable "Short" cannot be verified. "Under 80 words" can. Specific requirements give you clear criteria for evaluating the result (Lesson 12).
What to Be Specific About Not all details are equally important. Focus specificity on the dimensions that most affect the result for your task.
Dimension Vague Specific Task Help with my resume. Rewrite the three bullet points under "Experience" to start with action verbs and include a measurable result. Audience For people. For hiring managers at web development agencies. Scope Explain Docker. Explain only Docker images and containers, not networking or volumes. Length Not too long. Under 150 words. Format Organized. A numbered list with one command per item. Quality bar Good code. Code that handles empty input and follows PEP 8. Environment Recent Python. Python 3.12. Edge cases (not mentioned) If no date is given, write "Not stated." How to Choose Ask: "If the model guessed this wrong, would the output be unusable?"
If yes, specify it. If any reasonable choice would be acceptable, leave it open. For example, in a bio for LinkedIn, the point of view (first or third person) matters because it must match the rest of your profile. The exact sentence structure usually does not matter, so specifying it would only add constraints without value.
Why Conciseness Improves Results Important Instructions Stay Visible In a long, wordy prompt, critical requirements compete with filler for attention. A rule hidden in the middle of a long paragraph is more likely to be overlooked than the same rule written on its own line.
Fewer Opportunities for Contradiction Every sentence is an opportunity to say something that conflicts with another sentence. Wordy prompts often restate the same requirement in different ways, such as "around 80 words" in one place and "keep it brief" in another, which creates ambiguity.
Lower Cost and Faster Responses Prompts are processed in tokens (Lesson 2), and longer prompts cost more and take longer to process. For a prompt sent once, this hardly matters. For a prompt sent 50,000 times a month in an application, unnecessary text has a real cost.
Measuring Prompt Length in Tokens You can measure the effect of making a prompt concise with OpenAI's tiktoken library (pip install tiktoken):
import tiktoken
encoding = tiktoken.get_encoding("o200k_base")
wordy = ( "Hello! So I need a bio, and I want it to be for my LinkedIn profile, which " "is a professional networking site, so it should be professional. I'm a Django " "developer, I've been doing that for about 3 years, and I have built e-commerce " "sites. I'd like it to be written in first person, like \"I am...\", and it " "shouldn't be too long, maybe around 80 words or so, roughly, if possible." )
concise = ( "Write an 80-word LinkedIn bio in first person.\n\n" "About me: Django developer, 3 years of experience, builds e-commerce websites.\n" "Tone: professional and approachable." )
for name, prompt in [("wordy", wordy), ("concise", concise)]: print(f"{name}: {len(encoding.encode(prompt))} tokens") Code Explanation get_encoding("o200k_base") loads the tokenizer used by recent OpenAI models. Other providers use different tokenizers, so counts differ between models, but the relative difference between two prompts is usually similar. encoding.encode(prompt) converts each prompt into token IDs, and len() counts them. The concise version uses substantially fewer tokens while carrying the same information. Techniques for Writing Concise Prompts
- Remove Filler and Pleasantries Phrases such as "I was wondering if you could possibly", "I hope you're doing well", and "thanks so much in advance" do not affect what the model produces in most cases. They are harmless in casual chat, but in a reusable prompt they add tokens without adding information.
Wordy Concise I was wondering if you could help me write... Write... It would be great if the answer could be... The answer must be... Please make sure that you don't forget to include... Include...
Use Labels Instead of Sentences Wordy: The person reading this is a beginner who doesn't know programming yet. Concise: Audience: beginners with no programming experience. Labels such as Audience:, Tone:, Length:, and Format: communicate the same meaning in fewer words and make each requirement easy to find.
Use Lists for Multiple Requirements Paragraph:
The summary should be short, it should not include any names because of privacy, and it should also focus on decisions, and finally it needs to be in bullets. List:
Summary requirements:
- 3 to 5 bullet points
- Decisions only
- No personal names
- State Each Requirement Once Wordy prompts often repeat requirements for emphasis: "keep it short", then later "remember, brevity is important", then "don't make it too long". Say it once, precisely: Maximum 50 words.
If a requirement is critical and often ignored, making it more specific or more prominent (its own line, or the final line before the input) works better than repeating it vaguely.
- Remove Instructions the Model Already Follows Instructions such as "use correct grammar" or "make sure your answer makes sense" describe behavior that capable models show by default. They add length without changing the output. Keep instructions that change behavior, and remove those that describe the default.
Exception: If testing shows the model does not follow a behavior by default for your task, the instruction is necessary. Decide based on observed results, not assumptions.
- Hedge Words Weaken Requirements Words such as "maybe", "roughly", "if possible", and "try to" make requirements optional.
Weak: Try to keep it around 100 words if possible. Strong: Maximum 100 words. Use soft wording only when a requirement truly is flexible, and then say so clearly: Aim for about 100 words; slightly longer is acceptable if needed to explain the steps.
The Risk of Too Much Specificity Specificity has limits. Over-specifying causes its own problems.
Problem 1: Rigid, Unnatural Output Write a product description of exactly 4 sentences. Sentence 1 must be 12 words. Sentence 2 must start with "This". Sentence 3 must contain the word "innovative" and the word "quality". Sentence 4 must end with an exclamation mark. The model may satisfy every rule and still produce awkward, mechanical text. When you control every detail, the model cannot use its strengths in natural writing.
Better:
Write a 4-sentence product description. Mention the product's main feature in the first sentence and end with a short call to action. Problem 2: Conflicting Constraints The more rules a prompt has, the more likely two of them conflict, such as "include all five features" and "maximum 30 words". When rules conflict, the model must choose, and the choice can vary between runs.
Problem 3: Rules for Problems That Never Happen Rules added "just in case" increase length and the chance of conflicts without improving results. Add constraints in response to observed failures (Lesson 12).
Problem 4: Overly Narrow Rules A rule written for one specific input may not apply well to others. For example, "Do not mention the coupon GOCOURSE10" only covers one coupon. "Do not mention coupon codes" covers the actual intent (Lesson 12, overfitting).
Finding the Right Level Task Type Appropriate Specificity Data for a program (JSON, labels) Very high: exact structure, values, and edge cases Technical instructions or code High: versions, environment, requirements Professional writing Medium: audience, purpose, length, tone, key points Creative writing and brainstorming Lower: goal and constraints, with room for variety A Complete Refinement Example Original Prompt Hi there! I'm working on my website, which is an ed-tech site where we sell courses, and I'm hoping you could help me write some FAQ answers. I think the answers should be helpful and clear and not too long but also complete. Our customers are mainly students. The question I need answered right now is about whether students can download videos. We don't allow downloads actually, but we do have the mobile app where they can watch offline. Please make it sound nice and friendly but also professional. Thank you!! Analysis Fragment Category Action Hi there! I'm hoping you could help me Filler Remove ed-tech site where we sell courses Context Keep, shorten customers are mainly students Context (audience) Keep helpful and clear and not too long but also complete Vague and partly contradictory Replace with a specific length whether students can download videos Input (the question) Keep We don't allow downloads... offline Context (facts) Keep; essential nice and friendly but also professional Vague tone Make concrete Thank you!! Filler Remove Refined Prompt Write an FAQ answer for an online course platform. Readers are students.
Question: Can I download course videos?
Facts:
- Videos cannot be downloaded.
- The mobile app allows offline viewing.
Requirements:
- 40 to 60 words
- Friendly, professional tone: plain language, no slang
- Start with a direct answer, then explain the offline option Possible output:
Course videos can't be downloaded directly. However, you can still learn without an internet connection: our mobile app lets you save lessons for offline viewing. Just open a course in the app, choose the lessons you want, and watch them anytime—even when you're traveling or have limited data. What Changed Filler was removed, reducing the prompt length significantly. "Not too long but also complete" became a specific range (40 to 60 words). "Nice and friendly but also professional" became concrete style rules. The facts were separated into a list, so they cannot be misread. A new requirement was added (start with a direct answer), making the structure more useful for FAQ readers. Notice that the output says "Just open a course in the app, choose the lessons you want". That step was not in the facts. The model inferred a likely process, which may not match your actual app. A prompt for real FAQ content should add: Use only the facts provided; do not describe app steps that are not listed. Specificity includes deciding what the model is allowed to add.
Common Mistakes Mistake 1: Confusing Short With Concise What it is: Removing important information to make a prompt shorter.
Why it happens: The belief that shorter prompts are always better.
What it causes: Vague prompts that produce generic or inconsistent outputs.
How to fix it: Remove filler, not information. Every requirement that changes the output should stay.
Mistake 2: Emphasis Through Repetition What it is: Repeating a requirement several times in different words.
Why it happens: A requirement was ignored once, so it was repeated.
What it causes: Longer prompts and possible contradictions between versions of the same rule.
How to fix it: State the requirement once, specifically and prominently.
Mistake 3: Vague Adjectives Instead of Requirements What it is: Strings of adjectives such as "clear, helpful, engaging, concise, complete".
Why it happens: Adjectives feel like they add quality.
What it causes: Little real guidance, and frequent contradictions such as "concise" and "complete".
How to fix it: Replace adjectives with measurable requirements and a stated priority.
Mistake 4: Over-Constraining Creative Tasks What it is: Specifying sentence counts, exact words, and structure for every part of creative content.
Why it happens: Applying the precision needed for data tasks to writing tasks.
What it causes: Mechanical, unnatural output.
How to fix it: Specify the goal, audience, and essential constraints, and leave the rest open.
Best Practices Specify what would make the output unusable if guessed wrong. Leave open what any reasonable choice would satisfy.
Remove filler, keep information. Pleasantries, hedges, and repetition can go; facts and requirements stay.
Use labels and lists for requirements. They are shorter and easier to scan than paragraphs.
Replace adjectives with measurable criteria. Numbers, explicit lists, and defined terms are clear and checkable.
State each requirement once. Make critical requirements prominent instead of repeating them.
Match specificity to the task type. Data tasks need precision; creative tasks need room.
Measure token length for prompts used at scale. Conciseness directly affects cost.
Specify what the model must not add. Restrict additions to the provided facts when accuracy matters.
Try It Yourself Exercise 1: Make It Concise Rewrite this prompt to be concise without losing any information:
Hey, so I really need your help with something important. I have a Python script and I think there might be some kind of problem with it, I'm not totally sure, but it's not working right. Could you please maybe look at it and tell me what's wrong if you can? It's for my college project which is due soon. I'm using Python 3.11. Thanks a lot! Then identify what is still missing for a useful answer. (Hint: the script and the error.)
Exercise 2: Make It Specific Make this prompt specific enough to produce consistent, usable results:
Write a good email to my students about the exam. Decide which dimensions matter (date, subject, audience, purpose, tone, length) and which can be left open.
Exercise 3: Find Over-Specification Write a deliberately over-specified prompt for a short story (exact sentence counts, required words, fixed structure). Run it, then remove all but the three most important constraints and run it again. Compare the naturalness of the two results.