Week 5 Deep Research: The New vs. CPO Investigation
-
Topic: Should I Buy a Car Right Now?
Week: Week 1
Rubric version: v1.0
Platforms compared: ChatGPT, Gemini, Claude
Winner: Claude (98.0 / 100)
Runner-up: ChatGPT (85.0 / 100)
Third place: Gemini (62.0 / 100)
Margin of victory: 13.0 points
Tags: ai-comparison, prompt-engineering, chatgpt-vs-claude-vs-gemini, weekly-showdown, ai-quality, rubric, week-1, car-buying, should-i-buy-a-car
Categories: AI Comparison, Prompt Engineering
Estimated reading time: 12 minutes
SEO title: Week 1 AI Showdown: Claude vs. ChatGPT vs. Gemini — Who Wrote the Best Car-Buying Prompt Post?
SEO description: We gave ChatGPT, Claude, and Gemini the same car-buying prompt topic and scored them across 7 dimensions. See which AI wrote the most useful, detailed, and actionable blog post.
Week 2 Deep Research: The New vs. CPO Investigation
The difference between a new vehicle and a certified pre-owned one can mean thousands of dollars in savings—or thousands wasted on the wrong choice. But the decision isn't simple: new vehicles have protective warranties and latest technology, while CPO vehicles offer depreciation relief and surprisingly robust manufacturer coverage. The math changes by credit score, region, vehicle type, and ownership horizon. Most car buyers make this call with a spreadsheet and good intentions. Smart buyers use AI Deep Research to untangle the variables that spreadsheets miss: dealer certification programs that sound manufacturer-backed but aren't, financing rates that can erase CPO sticker savings, warranty terms that run from surprising dates with unexpected exclusions, and market supply constraints that affect both price and availability. This week's Deep Research prompt is designed to hunt down the real data and build a decision framework before emotion takes over.
Why Deep Research?
Deep Research mode is different from a standard chat conversation. Instead of asking an AI a question and getting a quick answer, Deep Research lets you ask an AI to investigate a topic by searching across multiple sources, synthesizing patterns, identifying gaps, and building a structured analysis from the ground up. It's the difference between "What's the best used car to buy?" (which gets a confident but surface-level answer) and "Investigate the financial differences between new and CPO vehicles in my situation, compare manufacturer CPO programs, analyze dealer economics, and identify the real trade-offs" (which gets a rigorous, multi-threaded analysis with source citations).
This week, Deep Research matters because the new-vs-CPO landscape is opaque, dealer incentives are hidden, and the answer depends on dozens of variables: your credit tier, your location, the specific manufacturer's certification program, current financing rates, depreciation curves by segment, supply dynamics, and factors like "Did this car come from a lease or an individual trade-in?" CPO programs vary wildly—Toyota's 160-point inspection is fundamentally different from a Hyundai CPO program, and both are utterly different from "dealer certified" used cars that have no manufacturer backing. Financing rates shift by credit tier and lender; promotional APR rates on new vehicles (sometimes 0-2%) can erase the CPO sticker advantage. Warranty terms run from the vehicle's original in-service date—not from CPO purchase—a distinction that catches most buyers. Only deep multi-source research, structured analysis, and systematic comparison can untangle this enough to move from emotion to evidence.
The Deep Research Prompt
Prompt Breakdown — How AI Reads the Deep Research Prompt
The Deep Research prompt looks intimidating, but it's designed with precision. Every section does specific work. Understanding how AI reads each section will help you adapt this prompt to any complex decision.
"I need you to conduct comprehensive research on the new vs. certified pre-owned vehicle decision for my specific situation, and produce a structured investment-grade analysis..." — This opening sentence does three critical things: it anchors the AI to a real person with real stakes (not generic advice), it specifies the output format expected (structured, investment-grade, not casual), and it signals that depth and rigor matter. Saying "investment-grade analysis" tells the AI to treat this like institutional research, with citations and careful reasoning, not a friendly chat.
Transferable principle: Begin research prompts by anchoring to a specific person's situation and defining the output rigor level. "Generic advice" produces surface-level results. "Investment-grade analysis" produces evidence-based, cited, methodical results.
"CONTEXT — MY SITUATION: [Budget ceiling, vehicle type, ownership duration, credit tier, etc.]" — Instead of asking the AI to ask you follow-up questions, you provide all context upfront. This prevents the AI from inferring wrong assumptions or asking time-consuming follow-ups. The AI now has the complete picture and can deploy research strategically. Vague prompts produce vague results; specific context produces specific analysis.
Transferable principle: For research prompts, provide all context variables upfront in a structured list. Don't make the AI play 20 questions. Specific inputs enable specific outputs.
"RESEARCH MISSION: I need you to investigate and synthesize findings on eight core research threads..." — This segment defines the research architecture. Instead of "research new vs. CPO," it says "research eight specific, named threads, and for each thread, search multiple sources, note conflicts, flag assumptions." The AI now understands it's not doing a quick keyword search; it's doing deep, multi-sourced synthesis. The eight threads are non-obvious (Thread 2 on dealer economics helps you understand why dealers push CPO, which shapes what you'll hear on the lot; Thread 5 on supply/demand explains regional price variation).
Transferable principle: Define research architecture explicitly. Don't say "research X." Say "research these 8 specific dimensions of X. For each, search multiple sources, synthesize, note conflicts, and flag assumptions." Explicit structure produces organized, rigorous output.
"RESEARCH THREAD 1 — MANUFACTURER CPO PROGRAM COMPARISON: Compare the CPO programs for the top 3 vehicle manufacturers... For each manufacturer's CPO program, research and document: [Age limits, inspection rigor, warranty terms, in-service date rule, exclusions, additional benefits, source]" — Each research thread specifies exactly what to research and how deep to go. For Thread 1, it's not "compare CPO programs" (which could mean anything). It's "Research these seven specific CPO program attributes for each OEM, and use official OEM documentation as your source." The specificity forces the AI to be comprehensive and prevents it from stopping early.
Transferable principle: Break each research thread into sub-dimensions with specific questions. The more specific your questions, the more complete and organized the research output. Vague threads produce vague findings.
"ANALYSIS & DELIVERABLES: After researching these eight threads, synthesize findings into four deliverables: [Executive summary, research findings by thread, comparison table, decision framework]" — Rather than asking "provide your findings," you specify exactly how findings should be organized. Deliverable 3 (Comparison Table) is a template you provide, which means the AI will fill it in precisely, not improvise a different format. Deliverable 4 (Decision Framework) is a structured checklist for real-world use at a dealership. This transforms research into action.
Transferable principle: Specify output structure before the AI starts. Don't ask "summarize what you find." Ask "Provide findings in these four formats: [specific formats]." Templates and structured output force clarity and actionability.
"CONSTRAINTS FOR THIS RESEARCH: Search across multiple sources; if sources conflict, note the conflict... Every financial claim must be attributed to a specific source... Flag any data that is estimated or illustrative vs. measured..." — These constraints tell the AI to be intellectually honest. Many research prompts fail because they allow the AI to confidently state uncertain information. These constraints flip that: the AI must distinguish measured data from estimates, must cite sources, must note conflicts. This is the difference between "here's what I found" and "here's what I found, here's my source, here's what I'm uncertain about."
Transferable principle: Always include a constraints section that explicitly requires attribution, distinguishes measured vs. estimated data, and demands transparency about uncertainty. Honest research doesn't hide assumptions; it surfaces them.
"TONE & STRUCTURE: Write in clear, direct language... Lead with numbers and evidence... End each section with 'So what does this mean for my decision?' to connect findings to action." — This final section is a style guide that shapes how findings are presented. It's not enough to have good research; it must be readable and actionable. Asking the AI to end each section with "So what does this mean?" forces it to connect research to your real decision, not just state facts.
Transferable principle: Always include a tone/structure section in research prompts. It determines how findings are presented and how actionable they become. Good research presented poorly is wasted research.
What to Expect from Deep Research
Output Length: Expect 8,000-15,000 words of output, depending on the depth of research available and the specificity of your inputs. The Executive Summary alone will be 300-500 words. Each research thread section will be 800-1,200 words with subsections, findings, and implications. The comparison table will fill a full page. The decision framework will be 400-600 words of actionable checklists.
Completion Time: Deep Research on ChatGPT, Claude, or Gemini typically takes 2-5 minutes to execute, depending on server load and the complexity of your inputs. The research phase runs invisibly (the AI is searching across sources), and then the output is compiled and presented. You don't see the searching; you only see the final synthesis.
Structure: The output will be organized by deliverable, with clear headers, subheaders, and section breaks. Each research thread will have bullet points for findings, source attribution in parentheses, and a closing question or implication. The comparison table will be easy to scan. The decision framework will be a numbered or bulleted checklist you can print and bring to the dealership.
Quality Signals: High-quality Deep Research output will include specific numbers tied to sources (not round figures), conflicting data points with explanations of why they differ, assumptions stated explicitly, and clear links between findings and recommendations. If the AI produces output without citations, or if it hedges every finding with "it depends," the quality is lower—ask follow-up questions or refine your inputs.
Key Research Questions the Prompt Answers
1. What is the actual financial advantage of buying CPO instead of new for my specific situation? The comparison table will show total cost of ownership over your ownership horizon, accounting for acquisition cost, financing rates at your credit tier, depreciation, insurance, and maintenance. This is your primary financial decision metric.
2. Is this dealership selling "manufacturer CPO" or "dealer certified," and what's the difference? The decision framework provides three pre-visit research tasks and five verification questions that will help you distinguish on the lot. Most buyers don't know these are different; this research ensures you do.
3. What does my credit score actually mean for financing rates, and how much will a 2-3 point difference in APR cost me? The prompt asks you to specify your credit tier and calculates the interest differential on your specific loan amount. Seeing "$3,300 in additional interest because my credit is good instead of excellent" makes the cost concrete.
4. How much does the manufacturer's specific CPO program matter compared to others? Research Thread 1 breaks down each major OEM's CPO program by inspection rigor, warranty duration, in-service date rules, and exclusions. You'll see that Toyota and GM's programs are substantially more rigorous than some competitors.
5. Why are CPO vehicles in my region harder to find and more expensive than they were last year? Research Thread 5 explains supply dynamics—pandemic-era production cuts, low lease return volumes, and current inventory constraints. Understanding why prices are what they are helps you negotiate smarter.
6. What's the warranty actually covering, and what won't it cover if something breaks? Research Thread 1 details warranty term, transferability, and common exclusions. Many buyers are shocked to learn that "7-year warranty" excludes infotainment, sunroof, and ADAS electronics—the expensive stuff.
7. If I own this vehicle for 5 years, will a 3-year-old CPO or a new vehicle cost me less? Research Thread 4 calculates the depreciation and TCO crossover point. Sometimes new wins; sometimes CPO wins. Your inputs determine which.
8. What should I be worried about when I walk onto the dealership lot, and what specific questions will protect me? The decision framework provides six red flags to watch for and a post-visit decision checklist. This is your defense against dealer tactics and against your own emotional decision-making.
Platform-Specific Tips for Accessing Deep Research
ChatGPT (ChatGPT Plus with GPT-4 or GPT-4 Turbo): ChatGPT doesn't have a formal "Deep Research" mode, but GPT-4 Turbo can perform multi-source research if you enable web browsing. When pasting the prompt, check that web browsing is enabled in your settings. GPT-4 will search the web for current financing rates, CPO program documentation, depreciation data, and market reports. Expect results to be strong for research threads 3, 5, 6, and 7 (data-driven topics) and good for threads 1, 2, 4, and 8 (requiring synthesis and interpretation). Output typically arrives in 3-5 minutes.
Claude (Claude 3.5 Sonnet): Claude does not have integrated web search (as of April 2026), so you'll get research based on Claude's training data cutoff (April 2024). This is a limitation for real-time data like current financing rates and 2026 market reports. However, Claude excels at synthesis and structured analysis—it will produce clearer reasoning, more transparent assumptions, and more rigorous frameworks than other models. Use Claude's output for the structural analysis (comparison tables, decision frameworks, assumption documentation) and supplement with ChatGPT or Gemini for current-data threads. Output typically arrives in 2-3 minutes.
Gemini (Google Gemini with Google Search): Gemini has native deep research capabilities and integrated Google Search. When you paste this prompt into Gemini, check that "Google Search" is enabled. Gemini will search for current CPO program details, financing rates, market reports, and supplier data across Google's index. Expect strong results across all eight research threads. Gemini's output is often more journalistic and reader-friendly than Claude's, which can be a strength for clarity but sometimes less rigorous on assumptions. Output typically arrives in 4-7 minutes because Gemini searches more systematically.
Pro Tip — Multi-Platform Workflow: For maximum research rigor, run the prompt on two platforms: Start with Gemini or ChatGPT to gather current data and market reports (threads 3, 5, 6). Then ask Claude to take that data and build the comparison table and decision framework with maximum analytical rigor. Claude will catch assumptions the other platforms missed and produce cleaner structured output. Total time investment: 10-15 minutes. Value delivered: institutional-grade analysis.
How This Connects to the Weekly Posts
This Deep Research prompt is the investigation layer of Week 2. The three platform-specific posts (ChatGPT, Claude, Gemini) teach you three different prompt variations for making the new-vs-CPO decision at beginner, intermediate, and advanced levels. Those prompts use conversational reasoning and quick analysis. This Deep Research prompt goes deeper: it's designed for buyers who've moved beyond "should I buy new or CPO?" and now need the research-backed analysis to defend their decision and avoid getting trapped by dealer tactics.
Week 2 builds on Week 1's confirmed budget and total cost of ownership analysis. If you completed Week 1's prompts, you have a financial ceiling and a sense of what you can afford. This week's Deep Research prompt takes that confirmed budget and deploys it strategically across the new-vs-CPO landscape. The weekly posts (available on Ketelsen.ai) provide the prompt variations for your specific platform. This document provides the research-intensive variation for maximum depth.
Adaptability Tips: Using This Prompt for Other Decisions
1. New vs. Refurbished Equipment (Business Context): Replace vehicle-specific research threads with equipment-specific ones: manufacturer refurbishment standards, certified refurb programs vs. dealer refurb, financing availability for used equipment, depreciation curves for your equipment category, supply constraints in your industry, regional equipment pricing, the refurb vs. "seller's warranty" distinction, and emerging trends. The prompt architecture remains identical; only the research dimensions change. This works for medical equipment, manufacturing machinery, restaurant kitchen gear, professional audio/video equipment—anything with a new-vs-refurb decision.
2. Build vs. Buy Software (Technology Context): Apply the same structure to the build-vs-buy software decision: compare third-party solutions' feature sets against a custom build's capabilities, analyze vendor lock-in risks (the equivalent of warranty exclusions), research implementation costs and timelines, analyze TCO including maintenance and upgrades, investigate vendor stability and product roadmap (the equivalent of manufacturer depreciation), research regional/industry variations in software pricing, understand the SaaS vs. perpetual license distinction (equivalent to dealer-certified vs. manufacturer CPO), and track emerging alternatives and consolidation in the software category. The eight-thread structure maps directly.
3. Lease vs. Own Office Space (Business Context): Adapt the prompt to real estate: research lease vs. purchase financial comparison, investigate landlord/tenant protections vs. outright ownership, analyze financing options for commercial real estate, model depreciation (or appreciation) of office property in your market, study supply and demand dynamics in your region, analyze regional commercial real estate pricing, distinguish between lease terms that lock you in vs. flexible exit clauses, and track emerging trends (remote-first work, co-working alternatives, workspace-as-a-service). Same structure, different domain.
4. Franchise vs. Independent Business (Entrepreneurship Context): Use the research framework to compare franchises vs. independent business launches: franchise program comparison (initial fees, ongoing royalties, support, restrictions), franchisor economics and your visibility into their incentives, financing availability for franchises vs. startups, success rate and depreciation of franchise value, supply/demand for franchises in your category, regional variations in franchise costs and support, franchisor-branded vs. independent business distinction and legal protections, and emerging trends in franchising (micro-franchises, equity crowdfunding models).
Follow-Up Prompts
Follow-Up 1 — "Refine the Comparison Table:" If the Deep Research output produces a comparison table but you want to dig deeper on one vehicle or one variable, ask: "Take the comparison table from the Deep Research and expand the depreciation calculation, breaking it into year-by-year residual values for the top 2 vehicles. Also add a row for 'breakeven point' — at what ownership duration does CPO become cheaper than new for my situation?" This refines the table without requiring a full re-research.
Follow-Up 2 — "Build a Dealer Lot Decision Script:" Once you have the research, ask the AI: "Using all the findings from the Deep Research, create a word-for-word script for conversations I should have with a dealer when evaluating CPO vehicles. Include: (1) five questions I should ask before looking at any vehicle, (2) three questions I should ask specifically about the CPO certification, (3) four red flags to watch for during the test drive, (4) three final questions before I leave the lot." This turns research into real-world conversation prep.
Follow-Up 3 — "Stress-Test Your Recommendation:" Ask: "The Deep Research recommends [new/CPO]. Now assume you were wrong about one key assumption (e.g., my credit score improves by 100 points, or financing rates drop 2%, or supply of CPO vehicles tightens further). How would each of these changes affect the recommendation? At what point does the recommendation flip?" This helps you understand how sensitive your decision is to variables that might change.
Metadata
Topic: New vs. Certified Pre-Owned: Let AI Make the Case — The new-vs-CPO decision framework using Deep Research methodology
Week: Week 2 of 7 ("AI at the Dealership: 7 Weeks of Prompts That Could Save You Thousands")
Series: AI at the Dealership
Content Type: Deep Research methodology + prompt breakdown + follow-up prompts
Platform Compatibility: ChatGPT (GPT-4 with web search), Claude 3.5 Sonnet, Google Gemini (with Google Search)
Prerequisite: Week 1 — "Should I Buy a Car Right Now?" (recommended to have confirmed budget and TCO analysis from Week 1)
Tags: Deep Research, new-vs-CPO, vehicle financing, depreciation analysis, dealer tactics, CPO programs, total cost of ownership, financial decision-making
Categories: Car Buying, Financial Planning, AI Research Methodology
Difficulty Levels of Related Posts: Beginner (Week 2 ChatGPT variation), Intermediate (Week 2 Claude variation), Advanced (Week 2 Gemini variation)
Reading Time: 15-20 minutes to read this post; 8-15 minutes to run the prompt; 30-45 minutes total to work through the output and build your decision framework
SEO Title (under 60 characters): Deep Research: New vs. CPO — AI-Powered Analysis
SEO Description (150-160 characters): Use Deep Research mode to investigate the new-vs-CPO decision across eight research threads. Get a structured, investment-grade analysis with financing, depreciation, and dealer tactics covered.
Publication Date: April 13, 2026
Last Updated: April 13, 2026
