Fix Lazy AI Answers With Three Lines Added To Every Prompt

Your prompts get lazy answers because the model is missing three things: the context that explains your situation, an example of what good looks like, and a clear line for what counts as done. Add those three lines to any prompt and the answer sharpens immediately, without touching your topic or your tone.

What makes an answer feel lazy?

A lazy AI answer is one that could have been written for anyone, because in a real sense it was. When you type a short instruction with no situation attached, the model fills the gap with the most statistically average response it can produce. Anthropic’s own prompt engineering documentation frames this plainly: the model has no memory of who you are or what you’re trying to do unless you put that information in the prompt itself, so a bare instruction reads as an invitation to hedge (Anthropic, prompt engineering overview). OpenAI’s own help center guide for developers makes a similar recommendation, urging you to state the audience and purpose explicitly rather than leaving it implied (OpenAI Help Center, best practices for prompt engineering). Ask a new hire to write something about your product with no audience, no format and no definition of done, and you would get the same flat, safe draft. The model isn’t malfunctioning. Given how little you gave it, this is its best guess.

The three things every prompt skips

Most prompts skip context, an example, and a success criterion, and any one missing piece is enough to flatten the answer. Context tells the model who you are, who’s reading the output, and what constraint it can’t ignore. An example shows the model what good looks like in your voice, instead of making it guess your style from scratch. A success criterion tells it when to stop optimizing and commit, instead of hedging every sentence to cover every possible reading of a vague ask. Anthropic’s own guide on chain-of-thought prompting recommends stating what the finished answer needs to accomplish before asking the model to reason toward it (Anthropic, chain-of-thought prompting), and OpenAI’s prompt engineering guide makes a similar recommendation, to specify the desired length, format and audience together rather than one at a time (OpenAI, prompt engineering guide). Three missing lines, one generic answer.

How do you add context in one line?

You add context in one line by naming your role, your audience and the constraint the model needs to respect, in a single sentence before the actual ask. Not “write a follow-up email” but “I’m a support lead writing to an enterprise client who already trusts us, so skip the apology and get to the fix.” Anthropic’s documentation on system prompts describes this same move, giving the model a role and situation before the task, as one of the fastest ways to change its default behavior (Anthropic, system prompts). Microsoft’s Azure OpenAI documentation calls this one of the highest-leverage changes you can make to a prompt, because the model uses that sentence to pick a register, a length and a level of formality before it writes a single word of the answer (Microsoft Learn, prompt engineering concepts). This is different from a model’s long context window, which is about how much text it can hold at once, not what it knows about your situation, a distinction covered in Long Context, Tool Use And Computer Use Explained Plainly.

Do examples beat instructions?

Yes, one good example beats ten lines of instruction, because showing the model your format does more work than describing it. Anthropic’s documentation on multishot prompting states that including examples is one of the most effective techniques for getting a precise, consistent output, since an example carries tone, structure and level of detail all at once, things that are hard to spell out in words (Anthropic, multishot prompting). This isn’t new. Brown et al.’s 2020 paper introducing GPT-3 showed that giving a model a handful of examples inside the prompt, without any retraining, sharply improved its performance over giving it the instruction alone (Brown et al., 2020, “Language Models are Few-Shot Learners”). One good example, pasted from your own past work, teaches the model your voice faster than a paragraph of adjectives.

What counts as success criteria?

A success criterion is one sentence that tells the model what “done” looks like, so it can commit to an answer instead of hedging across every possible interpretation. Skip vague goals like “make it good.” Write something specific: success looks like a paragraph a tired manager can approve in one read, no jargon, one clear ask. Anthropic’s guidance on writing clear, direct prompts makes the same case, that the model performs closer to what you want when you tell it exactly what the output needs to accomplish, rather than leaving that judgment to it (Anthropic, be clear and direct). Wei et al.’s 2022 research on chain-of-thought prompting found that asking a model to show its reasoning before landing on an answer, another way of stating what the finished response should look like, measurably improved accuracy on multi-step problems (Wei et al., 2022, chain-of-thought prompting). Without this line, the model guesses when it has done enough, and it usually guesses conservatively. That guess is what a hedged, generic answer looks like.

The three fixes, ranked by impact

Ranked by how much each one changes the final answer, context comes first, success criteria second, and examples third, based on what each line controls in the response.

1. Context. This changes the substance of the answer, not just its polish. Without it, the model picks a generic audience and a generic angle, and no amount of editing afterward fixes a response built for the wrong reader.
2. Success criteria. This changes how confidently the model commits. A prompt without a stated finish line gets a hedged answer that tries to cover every reading of the question instead of picking one and defending it.
3. Example. This changes tone and format fastest, but it changes least about the substance underneath. It’s the smallest fix of the three, and also the easiest one to skip, since it takes the most effort to prepare.

Does this affect your job?

It affects your job because prompting has become a skill employers test for directly, not just a personal habit for getting better output. A 2024 survey of prompting research cataloged 58 distinct techniques researchers have documented for influencing model output, evidence this has become a defined discipline, not a loose set of tricks (Schulhoff et al., 2024, “The Prompt Report”). I’ve written about why employers now list AI fluency as a baseline requirement rather than a bonus in Why Employers Now Require AI Skills And What To Do, and about how to describe your own AI habits when an interviewer asks in What To Say When An Interviewer Asks How You Use AI. The three-line habit gives you a concrete, demonstrable answer to that question.

Does this work for every prompt?

No. This three-line fix helps most on repeated, specific tasks, like emails, summaries, or reports you write in some form every week, and it helps less on genuinely open brainstorming, where an example can anchor the model on your first idea instead of generating new ones. Google Cloud’s guidance on prompt design makes this same trade-off explicit, describing examples as most useful when you want to constrain the format and style of the output, and least useful when you want the model to explore options you haven’t thought of yet (Google Cloud, prompt design strategies). If you’re brainstorming, drop the example and keep only context and a success criterion. If you’re producing something you’ll send or publish, use all three.

You are a writing coach who specializes in prompt engineering. I'm going to give you a task I want an AI assistant to do for me, and I want you to help me turn it into a strong prompt using three additions: one line of context (my role, my audience, and the constraint I can't ignore), one example of what good output looks like in my own voice if I can provide one, and one sentence stating what success looks like. Here is my task: [paste your rough task here]. If you don't have enough information to write the context line or the success criterion, ask me one clarifying question at a time before you draft anything. Once you have what you need, give me the finished three-line addition, plus the full rewritten prompt ready to paste elsewhere.

Act as a prompt engineering reviewer. I'll paste a prompt I've been using that keeps producing generic or hedging answers. Check it against three things: does it state who I am and who the output is for, does it include or reference an example of the format or tone I want, and does it say what a finished, successful answer looks like. For each of the three, tell me specifically whether it's present, missing, or vague, and if it's missing or vague, write the exact one-sentence addition I should paste in to fix it. Don't rewrite the whole prompt, just give me the three additions. My prompt: [paste it here].

I write [type of content, e.g. client update emails, weekly reports, LinkedIn posts] regularly and want to build a small library of examples I can drop into future prompts. I'll paste two or three pieces I've written that I'm happy with. Read them and tell me what's consistent across them in tone, structure, and length, then write a short one-paragraph example instruction I can reuse that describes this style precisely enough for an AI assistant to match it without needing the full original text pasted in each time. Ask me first if the pieces I paste seem inconsistent with each other, rather than guessing which one to prioritize. Pieces: [paste 2-3 examples].

Act as a blunt hiring manager who has read ten thousand resumes, not a career coach. I will paste my full resume and the job description I want. Rewrite the whole resume for that role, section by section, in this order: summary, experience, skills, education. Rules: every experience line leads with impact, not duty. Use bracketed placeholders like [8 percent] for any number I did not give you, and list at the end every placeholder I need to replace with a real figure. Keep it to one page of text. Plain formatting only, no tables or columns, so screening software can parse it. After the rewrite, tell me the three weakest claims that need evidence before I send this anywhere. My resume: [paste resume]. The role: [paste job description].