Why AI Answers Feel Generic — and How to Fix It in One Prompt
Why AI Answers Feel Generic — and How to Fix It in One Prompt — How to Talk to AI: Prompts That Help You Get the Answer You Need. Where generic answers come from; The smart-newcomer test; A short brief instead of more adjectives; What specificity changes; Compare outputs fairly. Practical assignment: Rewrite one flat prompt as a brief.
Demo edition. The materials are undergoing editorial review.
Topics covered
- Where generic answers come from
- The smart-newcomer test
- A short brief instead of more adjectives
- What specificity changes
- Compare outputs fairly
Where generic answers come from
When you write “Describe our new sofa,” the request contains nothing that separates the needed text from thousands of other product descriptions. The model does not know who will read the page, what buyers worry about, which facts matter, or which language your shop avoids. It has to infer those details from a short request, so it may produce plausible but broadly applicable copy.
We will call the missing information the context gap: the difference between what you know and what you have actually given the model. Closing that gap does not require magic wording. It requires relevant facts, a clear task, an audience, constraints, and signs of an acceptable result. This still does not guarantee accuracy; important facts in the answer need verification.
Common mistake: adding praise words instead of information. “Write an amazing, professional, highly compelling description” does not say what the description must contain. “Open with the sofa’s width and explain whether its covers are removable” gives the model an instruction you can execute and check.
The smart-newcomer test
Imagine giving the request to a capable colleague on their first day. They can write, but they do not know your product, readers, or internal rules. Could they begin without guessing important details? That is the smart-newcomer test.
First, turn a topic into an action. “Inflation” is a topic; “explain inflation to an adult with no economics background” is a task. Use a concrete verb such as explain, compare, shorten, suggest, or turn into a table. Then supply what the model cannot reliably know on its own: the purpose, audience, source facts, constraints, and quality criteria.
If critical information is missing, explicitly invite the model to ask clarifying questions. Do not demand an interview for every small request, though. A simple rewrite may need only the text and desired tone; material that you plan to publish, send, or rely on for a decision deserves a fuller brief.
A short brief instead of more adjectives
Use these fields as a checklist, not as a compulsory magic formula:
- Task: what to do.
- Why: where the result will be used and what it must achieve.
- For whom: who will read or use it, including what they already know.
- Input: facts, numbers, source material, and boundaries the model cannot obtain from the general request.
- What good looks like: two or three criteria you will use to accept the result.
- Format: length, structure, tone, and required fields.
- Unknowns: “If critical information is missing, ask me before starting.”
“Why” helps resolve ambiguity. “Keep it short” does not define a useful length; “this is an in-app note people read while checking out” explains the reason for brevity. Quality criteria should also be observable: “includes the three return conditions” is more useful than “sounds excellent.” Fill only the fields that affect the result.
What specificity changes
Consider this fictional teaching example. Weak request: “Write a description of the Alba sofa.” Improved brief: “Write a description of the Alba sofa for its product page. Readers live in small flats and care about size, delivery, and cleaning. Facts: three seats; 214 cm wide and 92 cm deep; removable machine-washable covers; the back detaches for delivery. Put the dimensions and cleaning method in the first two sentences. Format: 120–150 words, two short paragraphs, then 4 specification bullets. Do not use ‘premium,’ ‘luxurious,’ or ‘elevate.’ If a critical fact is missing, ask.”
The second request leaves less room to invent the audience, priorities, or shape. Its output is testable: either 214 cm appears or it does not; either there are exactly 4 bullets or there are not. The brief does not prove the product facts, however. The product owner must supply and verify them.
The same method works for a meal plan. Replace “Make a weekly meal plan” with 5 weekday dinners for two adults and a 6-year-old, no spicy food, under 30 minutes each, a small oven and no microwave, a fictional budget of about $60, and ingredient reuse. For an inflation explanation, specify an adult beginner who wants to follow the news, one everyday example in dollars, 2–3 common terms, fewer than 300 words, and no formulas.
Compare outputs fairly
To see whether the brief helped, compare the two requests under similar conditions. Run the original request in a new chat and save the answer, then run the improved brief in another new chat. This prevents earlier conversation from giving one version extra context. Because generation varies, one attempt does not prove that a prompt is universally better, but it is enough for an initial practical check of your own task.
Compare observable criteria rather than overall feeling: whether the requested facts appear, whether the length and structure are followed, whether unsupported assumptions were introduced, and how many substantive edits remain. If the answer is weak, identify the missing field and revise the brief.
Do not paste customer names, contact details, company secrets, or other information an external service does not need. Data handling depends on the service and its settings, so check the selected tool’s current privacy and data-use terms before submitting sensitive material.
Practical assignment
Rewrite one flat prompt as a brief
Continue to learningCourse outline
- Why AI Answers Feel Generic — and How to Fix It in One Prompt
- Context: What the AI Can't Know About You
- Show, Don't Tell: Teaching the AI With Examples
- Format, Length and Structure: Get the Answer in the Shape You Need
- A Conversation, Not a Single Shot: Feedback and Interviews
- Roles and Perspectives: When “You Are an Expert” Helps and When It Doesn't
- Big Tasks: Break Them Into a Chain of Steps
- Checking the Result: Rubrics and the Devil's Advocate
- Prompting Myths: What Doesn't Work and How to Test Any Trick
- Your Prompt Library: Reusable Templates You'll Actually Use