Last week, I was helping a friend who had just started a business come up with ideas for an "anti-competitive campus event." She used ChatGPT to write the plan, and all she got were "giving lectures and distributing flyers." When I switched to Claude, it came up with "using jokes about 'playing dumb to fight the competition' to make students laugh out loud." Finally, I used Gemini and directly tagged "the classmate who's a champion of competition"—the three AI answers were worlds apart. But I know it's not that the AI is bad; it's that she didn't "choose the right person."
AI is not a "one-size-fits-all tool," but rather "people with different skills."“Some AI excel at data analysis, some at storytelling, and some at following trending topics. The key to getting AI to truly help you is to "find the right AI for the task," rather than "using AI to do everything."
Here are 3 steps to "selecting an AI formula," along with specific task scenarios and model matching examples to help you accurately find your "personal assistant":
Step 1: First, create a "profile" of the task, clarifying what "skills" it requires.“
Don't directly ask AI to "write a solution." First, break down the task yourself: What capabilities does this task require most from the AI? Is it "clear logic," "emotional resonance," "accurate data," or "strong internet savvy"?
Here are a few common task scenarios; you can see which one applies to you:
Scenario 1: Writing reports/papers/analyses (requiring "rigorous logic + accurate data")
Task ProfileFor example, “writing a ‘2024 Pet Supplies Market Analysis Report’ should include user profiles, competitor comparisons, and future trend predictions” – the core requirements are “complete logical chain”, “reliable data sources”, and “sufficient analysis depth”.
Matching AI: GPT (especially GPT-4).
WhyGPT's strength lies in its "structured output," which helps you break down user profiles into "user personas → pain points → competitor strengths and weaknesses → trend predictions," with each section clearly listed. For example, when writing about "user personas," it will proactively state that "621 women aged 25-35 account for 30% of users, focusing on 'cost-effectiveness' and 'ingredient safety,' and they typically browse Xiaohongshu (Little Red Book) between 8:00 PM and 10:00 PM," providing more detailed data than you could from your own research.
counterexampleClaude's reports tend to be "too verbose." For example, when analyzing competitors, he might start with "the founder's story of XX brand," then move on to "product details," and finally get to "advantages and disadvantages," making it easy for the logical chain to break. Gemini's reports tend to be "too erratic," and he might suddenly insert "the 'loneliness' of pet owners," making the report sound like an essay rather than an analysis.
Scenario 2: Writing emotional copy/stories/brand promotions (requiring "touching details + warmth")
Task ProfileFor example, when writing promotional copy for a "late-night pet hospital," the goal is to make pet owners feel "understood" and willing to bring their sick pets there. The core requirements are "emotional resonance," "detailed descriptions," and "strong sense of immersion."
Matching AIClaude (especially Claude Opus).
WhyClaude is like a "friend who can tell stories," able to connect with pet owners through their real-life experiences. When writing copy for a pet blogger, it directly wrote, "My hands were shaking as I fed my cat medicine at 3 a.m., just like the computer screen I left on after working overtime—don't worry, we're here 24/7." Fans commented, "I cried; this is someone who truly understands us."
counterexampleGPT's copywriting tends to be "too rational," for example, using "24-hour service = fast emergency response," which sounds like an instruction manual. Gemini's copywriting tends to be "too internet-savvy," perhaps writing something like "@Juanwang pet owner, don't panic if your pet is sick, 24-hour online pet hospital will take the blame," which lacks a sense of "care."
Scenario 3: Writing short videos/social media content (requires strong online presence and the ability to capture attention)
Task ProfileFor example, “Write a short video script for ‘Anti-Involution Campus Activities‘, which should make students stop and watch within 15 seconds, and also remember that ’anti-involution = rejecting ineffective comparisons”“—the core requirements are ”eye-catching opening“, ”lively language“, and ”a point of dissemination”.
Matching AIGemini (especially multimodal Gemini Ultra).
WhyGemini is like a "trendy internet celebrity who follows trending topics," quickly grasping the memes of young people. Previously, when writing scripts for content creators, it directly designed a script that started with "@Your Desk Mate, the anti-rolling guide, please keep this handy," followed by the meme of "slacking off anti-rolling" in the middle: "Others roll at 8 am, I roll '7 am to 7 pm and sleep 7 pm'," and ended with "# anti-rolling code, share your anti-rolling declaration in the comments section." Its completion rate was 40% higher than my previous script.
counterexampleGPT's short videos tend to be "too serious," such as "Event theme 'Rejecting Involution, Starting with Adjusting Mindset,' explained in 3 points," which sounds like a lecture; Claude's short videos tend to be "too sentimental," perhaps starting with "Are you also anxious about 'others being better than you'?" which lacks a sense of "lightness."
Scenario 4: Data calculation/prioritization/planning (requires "accuracy and high efficiency")
Task ProfileFor example, "budgeting for a 'pet supplies promotion' involves calculating venue fees, advertising costs, prize costs, and outlining the timeline from 'procurement' to 'promotion' to 'execution' to ensure budget isn't exceeded"—the core requirements are "accurate data," "clear steps," and "no omissions."“
Matching AIDeepSeek (or a more precise model like GPT-4).
WhyDeepSeek acts like a "meticulous assistant," calculating data down to the "two decimal places." Previously, when calculating promotional budgets for e-commerce operations, it not only listed "venue fee 5000 yuan (including utilities), advertising fee 3000 yuan (short video ads account for 60%), prizes 2000 yuan (100 pet toys)," but also noted "shipping fee 400 yuan (remote areas calculated separately)" and "labor cost 300 yuan (flyer distribution part-time)." It even included a "cost overrun warning": "If advertising costs exceed the budget, the short video ads can be reduced to 'Xiaohongshu reviews' (cost reduction 50%)."
counterexampleClaude's data calculations are "too convoluted," perhaps starting by explaining the "importance of the budget" before listing the data, which can be confusing; Gemini's data calculations are "too simple," for example, simply stating "total budget of 10,000" without explaining where the money was spent.
Step 2: Three "quick filters" to find the most suitable AI.
If you're still unsure how to choose, try these 3 simple filtering steps. I've personally tested them and they helped me avoid the "AI misuse" problem in 90%:
Step 1: Break the task down into smaller questions, and ask each AI to answer only this part.“
For example, when writing an "anti-involution event plan," instead of directly asking "Write an event plan," break it down into three smaller questions and let the AI answer them separately:
- Ask GPT: "How do you break down the three core objectives of this campaign (gaining 1000 followers / achieving a conversion rate of 5% / a repurchase rate of 30%) into specific steps?" (To see if the logic is clear.)
- Ask Claude: "What are the 'emotional resonance points' of this event? Give two specific examples" (to see if the details are touching).
- Gemini was asked: "How should the campaign's promotional topics be designed? Use three memes commonly used by young people" (to see if it has a strong online presence).
Then, by combining the answers from the three AIs, you get a complete solution that includes "logic, emotion, and points of communication".
Step 2: Conduct "low-cost tests" to see if the AI "can be implemented."“
Don't start by using AI to write large-scale plans. First, write a "small paragraph" for testing: for example, write the opening of an "anti-involution campaign," and have 3 AIs write 3 versions respectively. Then, use "target reader preferences" to judge—if the target audience is college students, choose Gemini's "@Wangjuan's Desk Mate" version; if the target audience is working professionals, choose Claude's "The loneliness of working overtime late at night is like a computer that's not turned off" version.
Step 3: Fix your "frequently used AI list," don't chase the "latest models."“
I've seen too many people switch between GPT, Claude, and Gemini models, only to find themselves constantly adapting. There's no need to chase the latest models.Choose 1-2 AIs that you are most comfortable with and pair them up according to the task scenario.：
- My go-to approach: GPT (logic) + Claude (emotion) + Gemini (internet savvy). When writing reports/proposals, I first use GPT to build the framework, then Claude adds the emotional details, and finally Gemini adds the catchy phrases—it's efficient and effective.
Step 3: Use DiffMind"One-click comparison" of the outputs of different AIs
If you still find the process of choosing between different models tedious, try DiffMind—it can open multiple AI windows simultaneously and highlight the differences in output from different models. For example, when writing copy, you can see GPT's logic, Claude's emotion, and Gemini's communicative points at a glance, allowing you to directly "copy the most needed parts," which is 10 times faster than switching platforms yourself.
More AI is not necessarily better; more accurate AI is better.The core of choosing the right AI is to "match the AI's strengths to the needs of the task"—just as you wouldn't ask a "storyteller" to calculate financial statements, nor would you ask a "meticulous accountant" to write emotional copy.
The next time you use AI, pause for 10 seconds and think: "What problem does this task most need it to solve?" Only by choosing the right "person" can AI truly become your "efficiency accelerator".
