A competitive analysis that used to cost $10,000 and take three weeks just took 35 seconds. Paul Roetzer, founder of SmarterX, proved it on a livestream — one prompt, two frontier models, a full strategic breakdown of a rival company in under a minute. No elaborate prompt engineering. No multi-agent orchestration. Just a clear question and two capable models.
If your marketing team is still outsourcing competitive intel at premium rates, or skipping it entirely because the budget won’t stretch, this changes the math completely. Here’s what actually happened, why it works, and the workflow you should steal today.
What Roetzer Actually Did
Roetzer ran a single SWOT prompt against two of 2026’s strongest models — OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Fable 5. He asked each to analyze a specific competitor: strengths, weaknesses, opportunities, threats, differentiation, and strategic positioning.
Both delivered. Detailed strategic output, not generic filler. Named specific products, called out real weaknesses, identified actual market gaps. The kind of analysis a junior strategist would bill a week to produce.
The prompt itself wasn’t clever. It was direct: name the competitor, name the analysis framework, define the scope, ask for the output. That’s the uncomfortable truth for anyone still hoarding “secret” prompt libraries — clarity beats cleverness. A well-scoped request to a capable model outperforms a tortured 400-word prompt template every time.
Why This Works Now (And Didn’t Two Years Ago)
Two things changed between 2024 and 2026 that make this viable:
1. Models Actually Know Your Industry
Frontier models in 2026 have ingested enough earnings transcripts, product documentation, press coverage, and analyst reports that they genuinely understand competitive landscapes. Ask about a mid-market SaaS company and you get specifics — funding rounds, product gaps, customer complaints scraped from G2 and Reddit, recent leadership changes. This isn’t hallucination; it’s retrieval from a training corpus that now includes most publicly available business information.
2. Multi-Model Cross-Checking Is Trivial
Running the same prompt across two or three models and diffing the output is now a 90-second task. Where they agree, you have signal. Where they disagree, you have a question worth investigating. This wasn’t practical when each query cost dollars and took minutes. At 2026 pricing, it’s essentially free.
The 35-Second Workflow
Here’s the exact approach, adapted from Roetzer’s demonstration into something you can run today.
Step 1: Define the Competitor and Frame the Ask
Pick one competitor. Be specific — company name, primary product line, and the market segment you both compete in. Vague prompts produce vague analysis.
Bad: “Analyze Acme Corp.”
Good: “Run a SWOT analysis on Acme Corp’s project management software division, focused on the mid-market enterprise segment. Include differentiation analysis against [Your Company].”
Step 2: Run It Across Two Models
Take that prompt and paste it into two different models — ideally from different labs (OpenAI + Anthropic, or Google + xAI). You’re not looking for consensus for its own sake. You’re looking for the intersection where both models independently surface the same insight, which is where you should trust the signal most.
Step 3: Read Critically, Not Cynically
This is the step most teams get wrong. There are two failure modes:
- Blind trust: Copy-paste the output into a slide deck and present it as finished analysis. Don’t. Models hallucinate financial figures, misattribute quotes, and invent product features that sound plausible.
- Total dismissal: “AI can’t really understand strategy, so this is useless.” Also wrong. The output is a strong first draft that surfaces angles you’d miss and saves 80% of the grunt work.
The right posture: treat it as a junior analyst’s first draft. Good enough to build on, not good enough to ship unchecked.
Step 4: Verify, Then Decide
Spend ten minutes — not ten hours — verifying the three or four claims that would most change your strategy if true. Check the competitor’s actual pricing page. Skim their last two press releases. Look at their recent job postings (they reveal where a company is investing). Then make your call.
The Honest Frame: This Is a Draft, Not a Deliverable
Roetzer’s most important move wasn’t the prompt. It was how he presented the output to his team: as an unedited AI draft that needed verification. No pretending a human wrote it. No laundering AI output as expert opinion. Just transparency.
That honesty is what separates teams that get real value from AI strategic tools and teams that either get burned by hallucinations or never adopt the workflow at all. If you present AI-generated competitive intel as finished expert analysis and it contains an error, you lose credibility permanently. If you present it as a fast first draft that accelerates your thinking, you become the team that moves faster than everyone else.
What To Do Right Now
- Pick one competitor you’ve been meaning to analyze but haven’t had the budget or bandwidth for.
- Write a one-paragraph prompt using the template above. Name the company, name the framework, define the segment.
- Run it on two models today. GPT-5.6 Sol and Claude Fable 5, or whatever frontier pair you have access to.
- Block 15 minutes to verify the three most consequential claims.
- Share it with your team as a draft. Label it clearly. Ask where they’d push back.
You’ll have a competitive analysis that would have cost $10K and three weeks, produced in under an hour, for less than a dollar in API costs. The strategic advantage in 2026 isn’t access to AI — everyone has that. It’s the discipline to use it as a acceleration tool rather than a replacement for judgment.
The marketers who win this year aren’t the ones who trust AI the most or the least. They’re the ones who treat it as a very fast, occasionally wrong junior strategist — and build their workflow around that reality.
