Is ChatGPT's answer about your business idea enough?
Last updated: November 18, 2026
Written by the Zylar team. How we research and check our guides
ChatGPT can give you a useful first read on a business idea, but it tends to start from yes. When we asked it about a dog walking service for reactive dogs in Manhattan, its first word was "Yes" and its verdict was "Worth pursuing".
What it did well
It named four local businesses already offering the service, with their prices, listed real risks such as bites, insurance and licensing, and suggested a 60-day pilot with five to eight clients. That's a solid start.
Where it falls short for a decision
The verdict came before the evidence. Demand was rated "High" with no number behind it, so there's nothing to compare against your next version or your next idea. It also offered revenue scenarios of $10,000 to $20,000 a month that we hadn't asked for. OpenAI itself rolled back a ChatGPT update in April 2025 for being "overly flattering or agreeable".
Family is kind. Check it anyway. Sam pitches dog walking for reactive dogs at Thanksgiving and everyone is encouraging, but nobody asks what happens when a dog slips its harness until Zylar's Risk agent does. Concept demo with a hypothetical idea.
Same idea, two answers
| ChatGPT, signed out, November 7, 2026 | Zylar's example on our home page | |
|---|---|---|
| First thing you read | "Yes. I think this could be a very good business idea" | Promising, 60 out of 100 |
| Demand | "High" | Market 58 |
| Competition | "Moderate to high", four businesses named | Competitor 42, the weakest area |
| Can it be done | Operational complexity "High" | Feasibility 78 |
| What to do next | A 60-day pilot, aiming for five to eight clients | Prove it: lead with doorman and building referrals |
| Money | Unasked revenue scenarios, $10,000 to $20,000 a month | Five-year numbers only in the business plan, built from assumptions you can see and change |
Asked three more times, it still starts with yes
On November 18 we asked the same question three more times, in a new chat each time. ChatGPT now searches the web and cites its sources, which is better than in our first test. But every answer still began with "Yes", and each gave a different verdict: "Promising business opportunity", "Worth exploring" and "Worth testing". Each also added monthly revenue we hadn't asked for: $5,954, $7,150 and $7,600, which became $7,794 later in the same answer. None put a number on demand, so there's nothing to compare your next version against.
| Run | First word | Verdict | Number for demand | Sources cited | Unasked revenue a month |
|---|---|---|---|---|---|
| November 7 | Yes | Worth pursuing | None ("High") | No | $10,000 to $20,000 |
| November 18, run 1 | Yes | Promising business opportunity | None | Yes | $5,954 |
| November 18, run 2 | Yes | Worth exploring | None | Yes | $7,600, then $7,794 |
| November 18, run 3 | Yes | Worth testing | None | Yes | $7,150 |
How we tested
- Tool: ChatGPT at chatgpt.com, signed out, default settings. ChatGPT doesn't show signed-out users which model answered.
- Web search: on by default. The November 18 answers cited sources; the November 7 answer did not.
- Prompt, word for word: "I want to start a dog walking business for reactive dogs in Manhattan. Is it a good idea?"
- Runs: one on November 7, 2026 and three on November 18, 2026, each in a new chat.
- What we compared: the first answer, the verdict, whether demand got a number, sources, figures nobody asked for, and the next step.
- The Zylar column is the example idea on our home page, shown step by step in the concept demo. It is not a Zylar run of this exact prompt.
How to get a straighter answer from any chatbot
- Ask for the three strongest reasons the idea would fail, before you ask whether it's good.
- Ask what evidence would change its mind.
- Ask for its sources, and open them.
- Ask the same question about your second-best idea and compare the answers.
Zylar's agents each research one part of the idea, cite their sources and give it a score out of 100 with your main gap, so you can compare versions and ideas. It's in early alpha. Join the waitlist.