You Might Have Been Using the Wrong Tool All Along
After using AI tools for a few months, many people hit the same wall: "My prompts are detailed and thoughtful, but results are inconsistent — sometimes great, sometimes disappointing."
In most cases, that inconsistency doesn't come from poor prompting. It comes from using the wrong tool. It's like trying to tighten a screw with a hammer — the tool itself is fine, but it's fundamentally wrong for the job.
The major AI models — ChatGPT, Claude, and Perplexity — each have distinct design priorities and strengths. Understanding those differences is the single biggest upgrade you can make to your AI workflow.
What Is "Model Selection" — And Why It Matters More Than Prompting?
Model selection means deliberately choosing the right AI model for each task type, rather than defaulting to one tool for everything. It's an active decision, not a habit.
Model selection sits above prompt engineering in the effectiveness hierarchy because it determines your ceiling. Even the most perfectly crafted prompt cannot overcome a model that was designed for a fundamentally different purpose. Conversely, the right model with an average prompt often outperforms the wrong model with a perfect prompt.
Ethan Mollick, AI researcher at Wharton, has noted that the most effective AI users share one characteristic: they don't rely on a single tool. They switch models based on task type, treating their AI toolkit the way a skilled craftsman treats a toolbox.
The Three Main AI Tools: Core Positioning
Here are the three most widely used AI platforms for SMEs in 2026, each with a distinct design philosophy.
Perplexity AI: Find the latest, most accurate information. Perplexity's core design is real-time search plus AI synthesis. It connects to the internet, cites real current sources, and displays links alongside every answer. Use it for: researching market trends, tracking competitor announcements, verifying news or industry statistics, and any question that requires information current as of today. Its weakness is depth — strong at retrieval, weaker at long-form analysis and creative writing.
Claude (Anthropic): Deep analysis, long-form writing, structured thinking. Claude is designed for long-context understanding and complex reasoning. It handles very long documents, maintains consistency across complex conversations, and excels at logical analysis and written output. Use it for: writing business reports, analyzing contract clause differences, summarizing meeting notes into action items, drafting detailed client proposals, and any output requiring clear structure and rigorous logic. Its limitation is that it has a knowledge cutoff and doesn't browse the internet by default — for current facts, combine it with Perplexity.
ChatGPT (OpenAI): Everyday versatility, broad compatibility, rich ecosystem. ChatGPT's primary advantage is its ecosystem. It supports the most plugins and integrations, has the largest user community, and is the basis for the majority of AI tutorials and use cases you'll find online. Use it for: everyday writing tasks, simple data organization, plugin-dependent workflows, and as an entry point when learning from online AI guides. It's slightly behind Claude on deep reasoning and long-document analysis, and behind Perplexity on real-time information, but leads on flexibility and ecosystem breadth.
The Practical Selection Framework: One Question to Decide
When facing a task, ask yourself: "Do I need the latest information, deep analysis, or broad flexible execution?"
If you need current information or fact-checking — use Perplexity. Examples: "What are Hong Kong retail sales figures for Q1 2026?", "What new features did the latest ChatGPT release include?", "What press releases did our competitor publish last week?"
If you need deep analysis, long-form writing, or complex reasoning — use Claude. Examples: "Analyze the risk clauses in this contract," "Write an executive summary from these meeting notes," "Help me structure a client proposal."
If you need quick everyday tasks or specific tool integrations — use ChatGPT. Examples: "Rewrite this email in a more formal tone," "Generate a simple to-do list," "Connect my workflow via Zapier."
Advanced: The Two-Tool Combination Method
When task complexity increases, a single tool often isn't enough. The signature move of high-efficiency AI users is combining two tools to multiply their respective strengths.
Perplexity + Claude is the gold-standard combination for most SME knowledge work. The workflow: first, use Perplexity to gather the latest background data with cited sources. Then paste that researched content into Claude and ask for deep analysis, structural organization, and final writing. This pairing combines real-time information retrieval with high-quality content production — ideal for industry analysis reports, competitor research, and market entry strategy work.
A concrete example: Suppose you need to write a report on AI adoption in Hong Kong retail for a client. Step 1: Use Perplexity to search "Hong Kong retail AI adoption 2026 latest cases and statistics" — get current, cited results. Step 2: Paste Perplexity's output into Claude with this prompt: "Based on the following research data, write an 800-word industry analysis for a retail client, covering current landscape, opportunities, and recommended actions." Two steps, better results than either tool alone.
Common Model Selection Mistakes Hong Kong SMEs Make
Mistake 1: "ChatGPT is the most famous, so I use it for everything." ChatGPT is a great entry point, but it isn't the top performer in every category. For long-document analysis or complex reasoning, Claude is typically more consistent. For real-time market data, Perplexity is significantly more current and reliable.
Mistake 2: "AI gives inaccurate answers so I can't trust it." This often traces back to using a tool without real-time access for questions that require current facts. Switch to Perplexity for factual queries and get in the habit of checking cited sources — accuracy improves dramatically.
Mistake 3: "Longer prompts are better prompts." Prompt quality isn't about length — it's about clarity. A clear task description, desired format, and tone requirement for Claude outperforms a long unfocused background paragraph every time.
A One-Week Experiment: Feel the Difference Yourself
Theory is less convincing than experience. Here's a practical one-week experiment you can start immediately.
Days 1–2: Take three real work tasks from this week and complete each one using both ChatGPT and Claude — compare output quality directly. Days 3–4: Find a question requiring current market information, run it through both ChatGPT and Perplexity, and compare the timeliness and citation quality of the answers. Day 5: Try completing a moderately complex task using the Perplexity + Claude combination, and note how the result compares to a single-tool approach in both time and quality.
After one week, you'll have a much clearer, experience-based picture of which AI tool combinations work best for your specific workflow.
The Right Tool Is a Multiplier, Not Just a Convenience
There's an underappreciated turning point in the AI learning curve: moving from "uses one tool for everything" to "selects the right tool for each task." Before that turning point, most people experience AI as "sometimes useful, sometimes not." After it, AI starts functioning as a genuine multiplier on work output.
The model selection framework doesn't need to be complex: Perplexity for finding information, Claude for analysis and writing, ChatGPT for everyday tasks. Master this basic framework and you're already operating ahead of most AI users.
If you'd like to explore how to systematically bring AI workflows into your business, learn more here:
👉 https://ai.ud.hk/tc/campaign
AI that understands you | UD — AI made human
