It is easy to ask a chat tool “what time does the pharmacy on Main Street close” or “what do I need to bring to renew my ID at the county office” and get a specific, confident-sounding answer — a closing time, a list of documents, a phone number. It is much less obvious, in the moment, that the model has no reliable way to know whether that information is current, and in many cases no reliable way to know it was ever correct for that specific location at all. Business hours change, offices move, requirements get updated, and a model’s training data reflects whatever was written about a business somewhere on the internet, at some point, which is a different thing from what is true today.
This matters most exactly where the stakes are highest: a clinic appointment you cannot afford to arrive late for, a government office with a one-shot document requirement, a shop with a return window that is about to close. A wrong answer about a streaming app is a minor annoyance; a wrong answer about whether a clinic accepts walk-ins, or what documents an agency requires, can cost a wasted trip, a missed appointment, or a rejected application.
Never treat a model’s answer about a specific business’s, clinic’s, or agency’s opening hours, required documents, fees, or contact details as fact without verifying it against the organization’s own official page or a direct phone call — especially before travel, an appointment with real consequences, or a one-time errand. Generated local information can be outdated, generic, or in rare cases entirely fabricated for a less well-documented business; the model has no way to flag which of these applies to a specific answer, and it will sound equally confident either way.
Step 1: Understand why this specific kind of answer is unreliable
A model’s knowledge of a specific local business is a mix of whatever text mentioned it during training, general patterns about “how businesses like this usually operate,” and, occasionally, a plausible-sounding guess when the specific business is not well documented anywhere the model saw. None of these sources update in real time, and none of them come with a built-in confidence signal you can see — a well-documented, generic business and an obscure one get the same fluent, certain-sounding tone.
Step 2: Use the model to prepare the question, not to answer it
The model is still useful here — just for the framing and the call script, not the fact itself.
I need to visit [type of business/office, e.g. "the county vehicle
registration office"] for [purpose]. Draft a short list of the
specific questions I should ask when I call or check their official
site, to confirm: current hours, whether an appointment is needed,
what documents or fees are required, and any recent changes to the
process. Do not answer these questions yourself — just help me
prepare what to ask.
Step 3: Verify against the official source or a direct call
Find the organization’s own website (not a listing aggregator, which can also be outdated) or call the number listed on their official site — not a number a chat tool gave you, which could itself be wrong. Log what you actually confirmed, when, and how.
Organization: [name]
Official source checked: [URL or phone number found on their own
site, not from a chat tool]
Date/time checked: [when]
Confirmed: hours [___], appointment needed [yes/no], documents
required [___], fee [___]
Verify contact details themselves against an official source before calling or visiting — a chat tool’s phone number or address for a less common business can be wrong in ways that range from outdated to entirely incorrect, and following bad contact information can waste more time than the original question. This is also a place where scam risk enters: never provide payment details or sensitive personal information to a phone number or website you have not independently verified belongs to the real organization. Recognising AI-enabled scams covers the broader pattern of confident-sounding fake contact points and impersonation.
Step 4: Build the habit around anything with real stakes
Make verification automatic for a specific category of request rather than deciding case by case under time pressure: any appointment, any government office visit, any purchase near a return-window deadline, and anything involving a fee or a document requirement. For low-stakes, easily reversible errands (checking if a corner shop is generally open on Sundays, when you can simply walk over), a lighter touch is reasonable — the discipline should scale with the cost of being wrong.
A useful rule of thumb: if being wrong would cost you more than the ten minutes it takes to verify — a missed appointment slot, a wasted trip across town, a rejected form because you brought the wrong document — verify first. If being wrong just means walking to a second option nearby, the check is optional. Medical appointments, government offices, and anything with a hard deadline or a fee attached almost always belong in the first category, even when the underlying question feels mundane.
Step 5: Treat “I’m not sure” as a valid and useful answer
A well-behaved model will sometimes say it cannot confirm current hours or requirements for a specific, less-documented location — treat that honesty as more useful than a confident guess, not as a failure of the tool. If a model does hedge or flag uncertainty, that is exactly the signal to verify directly rather than pushing for a firmer answer, which risks getting a fabricated one instead.
Common pitfalls
- Trusting a confident tone as a proxy for accuracy. A model sounds exactly as certain when it is right as when it is filling a gap plausibly.
- Using a phone number or address the model provided without checking it against the organization’s own site. This is also the vector for the kind of AI-polished scam contact information covered in recognising AI-enabled scams.
- Verifying once, long before you need it. Hours and requirements can change between when you first checked and when you actually go — a final check closer to the date catches recent changes, the same discipline covered in travel plan verification.
- Skipping verification because the answer sounded specific. Specificity (an exact closing time, an exact document list) is not the same as accuracy; it is just detail.
Validation and fallback
For anything with a real deadline or cost of being wrong, verify a second time close to when you actually need the information — a clinic’s hours can change between last week’s check and today’s appointment. If the official site is unclear or you cannot reach anyone by phone, treat the visit as uncertain and build in a buffer (arrive early, bring extra documents) rather than proceeding on an unverified assumption.
Build the habit this week
Use the local service verification checklist the next time you need to visit an unfamiliar clinic, office, or shop for anything that matters. The same verify-before-you-act discipline applies directly to travel planning, and if a “helpful” contact detail or urgent-sounding message ever feels off, recognising AI-enabled scams covers how to check before you act on it.
