Three hikers planned a Mount Shasta trip with Google Gemini and ended up needing rescue last week. Before you roll your eyes at them — read the rest of this article, because the same mistake is happening to small business owners, freelancers, and weekend project planners every single day. They use AI to plan a launch, draft a client proposal, or summarize a vendor contract, follow the AI’s confident answer, and discover too late that AI can be confidently wrong in ways that hurt real businesses. Not in dramatic ways. In quiet ones: bad pricing, missed deadlines, contracts they shouldn’t have signed.
If you use AI to help run your business — and most beginners do now — you need a simple rulebook for when to trust it and when to double-check. Three hikers on Mount Shasta just paid the price for not having one.
What happened on Mount Shasta
The basics: three young men started hiking California’s Mount Shasta at 3 AM. The standard rule on that mountain is turn around by noon if you haven’t reached the summit. They reached the top at 7 PM. Then they tried to descend in the dark, called the sheriff’s office for directions, and spent the night in Mud Creek Canyon before Forest Service rangers pulled them out the next morning.
The Siskiyou County sheriff’s report said the hikers “were advised by Gemini to bring far less food and water than their group required, especially when their planned 8-hour ascent became a multiday ordeal.” That detail is the lesson. Gemini didn’t send them up the wrong trail. It confidently under-estimated how much water a group of three would burn on a 16-hour push, and the hikers followed the answer instead of packing for the worst case.
The official line from the sheriff’s office: “It is always advisable to call the local USFS Mount Shasta ranger station ahead of your trip to ensure you have the most accurate information, and to never rely solely on AI for your trip planning.” The full TechCrunch report is here.
Why Gemini (and every other AI) gets it wrong like this
The hallucination problem in plain English
AI chatbots sound smart because they are trained to sound smart. They generate text by predicting the next plausible word, not by checking facts. That’s an oversimplification, but the practical takeaway is right: AI doesn’t “know” anything in the way you do. It produces text that statistically fits the question.
This is why AI is great at brainstorming, summarizing, and rephrasing. It’s also why AI is dangerous when you treat it like a source of facts about specific real-world things: trail difficulty, current prices, the latest tax rule, your doctor’s dosage advice.
For a deeper look at why long AI chats drift toward worse answers, see our guide on why chatbots get less reliable the longer you talk to them. The mechanics are the same — pattern-matching loses ground to actual knowledge when the topic gets niche.
Why trail-specific details break AI
Mount Shasta route details — water needs at altitude, snow conditions in early September, the noon-turnaround rule — are exactly the kind of information AI is worst at. It’s specific to one mountain, one season, one ranger district. AI trained on general web text might have read the noon-turnaround rule somewhere, but it doesn’t have the live, local, this-weekend context. So it blends what it half-remembers with what sounds plausible, and produces a confident answer that’s just wrong enough to get you in trouble.
This happens with Gemini, ChatGPT, and Claude. It happens with paid and free versions. It’s not a brand problem. It’s a property of how these models work.
The five questions to ask before you follow AI advice
Here’s the framework. Before you act on any AI answer that touches the real world — your money, your safety, your schedule, your client’s expectations — run it through these five checks.
Where did this answer come from?
Ask the AI: “What’s your source for this?” If it gives you a real, specific source, you can verify it. If it gives you a vague “based on common knowledge” or invents a citation that doesn’t exist, that’s a red flag. AI sometimes hallucinates URLs, article titles, and statistics that look real.
What would I lose if this is wrong?
Low stakes = trust the AI. Brainstorming a blog post outline, getting a second opinion on a recipe, asking for synonyms — go ahead. AI is genuinely useful here.
Medium stakes = verify. Pricing for a service, a contract clause, a feature comparison — confirm with the vendor or your lawyer before committing.
High stakes = don’t trust AI at all. Medical symptoms, legal advice, anything involving physical safety, big financial decisions, anything you’d sue someone over. Call a human expert.
Is there a real person who knows more?
For the Mount Shasta hikers, the answer was: yes, the local ranger station. For your business, the answer is usually: yes, your accountant knows more than AI about your taxes. Your lawyer knows more about that contract. Your supplier knows more about delivery times. AI is a starting point, not the final word.
When was this answer last updated?
AI models have a training cutoff. Anything that happened after that cutoff — the latest tax law, a new product feature, a price change, a regulatory update — is invisible to the AI. If you’re asking about anything recent, that’s a flag.
Can I verify this against an official source?
If you can’t find an authoritative source to confirm the AI’s answer, that’s the answer itself: you can’t trust it. AI is most dangerous when it gives you specific facts you can’t easily check. A confident, specific, uncheckable answer is the worst combination.
Where AI is actually safe to trust (and where it isn’t)
Safe: brainstorming, drafting, summarizing
AI is excellent at:
- Brainstorming ten angles for a marketing campaign
- Drafting a first version of an email or proposal you’ll edit
- Summarizing a long document you don’t have time to read cover-to-cover
- Rewriting your rough notes into something polished
- Generating code snippets you’ll review and test
These are high-volume, low-stakes tasks. The cost of an error is small, and you can review the output.
Risky: real-world planning with safety stakes
AI is unreliable for:
- Trip planning (route difficulty, water needs, weather windows)
- Medical information and symptom checks — we covered this in detail for using ChatGPT and Gemini safely for health questions
- Legal interpretation of contracts or laws
- Tax filing decisions
- Anything involving physical safety
- Specific facts about real businesses, prices, people
These tasks share one feature: the cost of being wrong is high, and AI has no skin in the game. If the AI gives you bad advice, it doesn’t lose anything. You do.
What this looks like in your daily life
Here’s how the framework plays out in practice. You’re using AI to help plan a product launch. You ask for a recommended pricing tier for a SaaS product. AI gives you $29/month. Before you commit:
- Where did this come from? — AI says “based on common SaaS pricing patterns.” Vague.
- What would I lose? — If you’re wrong about pricing by $10/month, you might under-price by 30%. Real money.
- Is there a real person who knows more? — Yes, your accountant or a SaaS pricing consultant.
- When was this last updated? — SaaS pricing patterns have shifted in the last year. AI’s training data may be outdated.
- Can I verify this? — Yes, by looking at competitor pricing pages and your own cost structure.
Verdict: don’t follow the AI’s specific number. Use it as a sanity check, then do the real work.
Same pattern, different stakes: AI tells you your freelance contract should include a particular clause. AI tells you a customer is in a state where you need to collect sales tax. AI tells you a vendor’s product has a feature it doesn’t. For all three: verify, verify, verify.
The takeaway
The Mount Shasta hikers aren’t dumb. They just trusted an AI answer that sounded right in a domain where sounding right and being right are different things. If you use AI for your business — and you should, it’s genuinely useful — pair it with a simple rule: AI is a brilliant first draft and a terrible final authority.
Run the five questions before you follow AI on anything that matters. Brainstorm, draft, summarize with confidence. Plan, prescribe, or promise with caution. When the stakes are real, the right answer is a phone call to a human who knows more — not another prompt.
