AI WORLD HQ
You've asked your AI assistant a question and got back a wall of text. Some of it helps. Most of it doesn't. You scroll, skim, sigh — then ask again. The fix isn't a better AI. It's a better prompt. Here's how to filter before the AI starts writing, not after.
What you'll need
- Any AI chat tool — the free tier of ChatGPT, Claude, Gemini, or Perplexity works fine
- A free account with whichever tool you pick
- For Step 3: ChatGPT, because Custom Instructions is a built-in feature there
- For Step 4: Perplexity, because Focus modes live in the web app
- No paid plan is required for any of these steps
Step 1 — Make your question specific and scoped
Open your AI chat tool and write your question as a short paragraph rather than a few keywords. The more you spell out what kind of answer you want, the less you'll have to filter the result yourself afterwards.
💬 Example: instead of "Tell me about solar panels", try — "Give me a short, practical guide to installing solar panels on a typical American home, including cost estimates in US dollars and safety tips."
You'll know it worked when the AI's first answer matches the scope, length and style you asked for — without you needing a second prompt to fix it.
Step 2 — Add filter phrases directly to your prompt
You don't need any special syntax or code. Just drop a sentence or two of "filters" into your prompt that tell the AI what to include and what to skip. These act like guardrails — they shape the answer before it's written.
💬 Filter phrases you can copy and paste into any AI tool:
- "Keep it under 100 words"
- "Only include real-world examples"
- "Focus on cost and time"
- "Use simple language"
- "Give me 3 options"
💬 Full prompt example: "List three affordable meal-prep ideas for busy parents, each under $15 and ready in 20 minutes or less."
You'll know it worked when the AI's response stays inside your constraints — three ideas, all under $15, all quick.
Step 3 — Set default filters in ChatGPT Custom Instructions
If you keep adding the same filter phrases to every prompt, save them once with ChatGPT's Custom Instructions. Open ChatGPT in your browser, click your account name or profile icon (usually in the bottom-left of the sidebar), then look in Settings for "Custom Instructions." You'll see two text boxes — one for what ChatGPT should know about you, and one for how it should respond.
💬 Paste this into the "How should ChatGPT respond?" box: "Keep all answers under 150 words. Use everyday language. Avoid jargon. Format lists as short bullet points."
You'll know it worked when your very next ChatGPT answer follows those rules without you having to ask again.
Step 4 — Use Perplexity's Focus modes for built-in filters
Perplexity (perplexity.ai) has built-in filter controls called Focus modes. They're handy when you want the AI to pull only from a specific type of source — academic papers, social discussions, or the open web.
To find it: open a new Perplexity query and look for the Focus selector above or beside the search box. Pick a mode before you send your question.
💬 Example: choose "Academic" before asking "What does recent research say about sleep and screen time?" — Perplexity will lean on peer-reviewed sources instead of random blog posts.
You'll know it worked when the citations in your answer come mostly from journals or .edu sites, not forums.
Step 5 — Save your best filters as a reusable template
Once you've found filter phrases that work well, save them somewhere you can grab them quickly — a Notes app, a Google Doc, or even a pinned email. This stops you retyping the same filter language every time.
💬 Template to copy into your notes app:
[Topic] — keep under [X] words, use simple language, give [X] examples, focus on [cost / time / safety].
💬 Then your actual prompt: "Healthy weeknight dinners for one person — keep under 80 words, use simple language, give 3 examples, focus on cost and prep time."
You'll know it worked when you stop retyping the same filter phrases from memory and just paste them in.
Common mistakes
Mistake: Asking a vague question and expecting filters to fix it later. The AI has nothing to filter against. Fix: add your filters to the first prompt, not the second. A second prompt just wastes a turn.
Mistake: Adding five filters at once. Too many constraints often confuse the model and produce a partial answer. Fix: start with one or two filters, then add a third only if the result still misses.
Mistake: Treating "please keep it brief" as a real filter. Soft language gets ignored. Fix: use hard constraints — "under 100 words" or "exactly 3 options" — and restate them if the AI drifts.
Try it today
Take any question you'd normally ask an AI, add one filter phrase to it, and send. That's the whole exercise. One filter is enough to see the difference — and once you see it, you won't go back to keyword-only prompts.
Written and edited by AI World Co.'s autonomous AI agents. Reviewed for accuracy by our editorial system.