My parents think AI is a search engine.
They are in their mid-to-late seventies. When I asked them what AI actually is, the best they could describe was “the thing that answers questions on the computer.” Their entire mental model is ChatGPT and the little Gemini sparkle that shows up when they search. No training data. No models. No agents. No automation. Just: type a question, get an answer, trust it.
Here is what surprised me. They are not outliers. They are the majority.
The “Ask AI” Shift
A few years ago, my parents said “ask Google” for everything. Recipe, weather, directions, whether some supplement actually works. Google was their oracle, and they still glanced at the links, the sources, the top results.
Now they say “ask AI.” For everything. And they do not look at anything anymore.
No links. No sources. No cross-checking. They type the question, read the first answer, and move on with their day. The idea of verifying the answer never crosses their mind. Why would it? The AI sounds confident, it is always there, and it never says “I don’t know.”
It is annoying, honestly. I hear “ask AI” and I want to explain how it works. I want to say: it is not a person, it is a prediction engine, it can be wrong, it can make things up, you need to check the sources.
But here is the thing. I cannot fully explain it either. I know more than they do, sure. I know about training runs and context windows and tokens. But the capabilities move so fast that my understanding is a moving target too. Last year it summarized text. This year it writes code, builds websites, and runs agents that do things while I sleep.
They Are Not Unusual. They Are the Average.
When I looked at the research, the pattern was unmistakable. This is not a story about two elderly people who are behind the times. It is a story about how the mainstream actually uses AI.
- Half of American adults now use AI chatbots. Pew’s February 2026 survey of 5,119 adults found ChatGPT alone reaches 44% of adults, up from 34% a year earlier and more than double the share in 2023.
- 77% of Americans who use ChatGPT treat it as a search engine. Adobe’s survey found 54% use it because it summarizes complex topics quickly, and a third like that it needs fewer clicks than Google.
- 45% of people think ChatGPT looks up an exact answer in a database. That is the mental model, quantified: they believe it retrieves a fact, not that it generates text from patterns.
- Older adults are the fastest-growing users, not the laggards. A University of Michigan poll found 55% of people 50 and older have used AI tools, and a majority of them use it for everyday questions, the exact “ask AI” behavior my parents have.
The people I assumed were behind the curve are actually riding the same wave as everyone else. The difference is not adoption. It is understanding.
The Trust Gap
Here is the part that concerns me.
AI use is up, and AI trust is down. Polling firm 5W found a 99-point favorability gap between daily AI users (+57) and everyone else (-42), the widest behavioral divide in American public opinion. Half of America uses a technology it does not trust. The fix starts with the basics: our plain-English guide to API keys explains how AI tools actually connect and what they can and cannot do.
Except for one group: the people who trust it the most are often the ones who understand it the least. Edelman’s trust research found confidence rises sharply with use, and hesitation is rooted in unfamiliarity. The more people use AI, the more they believe it. My parents use it every single day. Their confidence is absolute.
And yet, the 2026 AI literacy research found that while half of American adults use an AI chatbot, fewer than one in five feel confident using it. The people who trust it most cannot explain what it is, what it can do, or when it is wrong.
From Gatekeeper to Oracle
Here is what actually changed, and it is bigger than the tools.
Google was a gatekeeper you could audit. It gave you links, sources, multiple results, a trail you could follow. If a result looked wrong, you could see where it came from and decide for yourself. The system was imperfect, but it was transparent enough to check.
AI is an oracle you cannot audit. Yes, it shows citations now, and it does it often. But the citations changed nothing. Users do not click them. They are a decoration on a confident answer, a way to look accountable without asking anyone to check. The user does not see how the answer was formed, or whether it is even a fact versus a confident guess. The 45% who think ChatGPT looks up answers in a database do not know that no database exists. There is nothing to check, and nobody told them.
This is not a small difference in interface design. It is a change in who defines what is true. For most of my parents’ lives, truth came with a paper trail. You could look it up, compare sources, argue with it. The oracle has no paper trail. It just answers, and the answer feels true because it sounds confident.
The Trust Is the Target
Now add the part that worries me most: what happens when the trust itself becomes the target?
I am not talking about a cartoon villain programming a lying AI. Directly controlling what a model says is genuinely hard. Models are stochastic, open-source versions exist, and no one fully steers them. The puppet-master scenario is mostly science fiction. The scammer needs none of that. They do not need to control the model. They need one person who was trained to trust the confident voice and never check.
The grandparent scam is the clearest example. A clone of a grandchild’s voice, built from thirty seconds of a public video, calls and says there is an emergency and money is needed right now. It used to take a lucky guess and a muffled line. Now it takes a free tool and a real voice. The scale is already public: McAfee found one in four adults has experienced an AI voice-clone scam or knows someone who has, and 77% of victims who engaged with the call lost money.
It does not stop at phone calls. A finance worker in Hong Kong transferred $25 million after a video call where the “CFO” and every “colleague” on screen was a deepfake. Fake customer support calls now sound exactly like the bank. Investment advice from a slick “AI analyst” is often a funnel to a fake platform. Every one of these works the same way: the victim was trained by the interface to believe the confident voice and skip the checking.
My parents are the perfect marks, and they are not alone. They never built a skepticism phase at all. They went from “ask Google” to “ask AI” without ever learning that an answer is not a source. The scammers know this. They know the people who trust AI most are the people who understand it least. That is not a side effect of the design. It is the foundation of an entire scam industry.
Why This Matters
This mismatch is not a curiosity. It is the risk.
My parents do not know AI can code. They do not know it can create. They do not know it is already running parts of the world, making decisions, drafting documents, and acting on their behalf. To them, it is a better search box, one that never argues back.
When AI is turned against people, who do you think gets hurt first? It will not be the builders. It will not be the power users. It will be the people who skipped the skepticism phase entirely, the people who went from “ask Google” to “ask AI” without ever knowing what changed under the hood. The scam does not need to be clever. The confidence is the exploit.
My parents did not stop checking sources because they are naive. They stopped because the interface made checking unnecessary. One answer, no links, no effort, total confidence. That is not a personal failing. It is a design outcome, and it applies to hundreds of millions of people.
What to Do About It
Here is the practical part, because despair is not a strategy.
If someone you love treats AI as a search engine, you do not need to lecture them about models and tokens. You need four small habits.
First, ask for the source and read one. Every major chatbot can cite now. Most people never click. One click is enough to rebuild the reflex that Google trained for twenty years.
Second, ask twice. The same question, rephrased, often produces a different answer. The first answer is not the truth. It is a draft.
Third, cross-check one number. A statistic, a date, a name. Thirty seconds on a search engine. That single habit does more for AI literacy than any explainer video, because it breaks the loop where confidence is mistaken for accuracy.
Fourth, when a voice asks for money, hang up and call back on a number you already have saved. A real bank, a real grandchild, will still be there. And agree on a family word, one word only your family knows, for emergencies. A voice can be cloned. A code word cannot.
If you want to show someone what an agent actually does instead of telling them, our no-code guide to building your first AI agent walks you through it in an afternoon.
None of this requires understanding transformers. It just requires remembering that the confident voice is not a source. It never was.
What I Want Them to Know
I am not writing this to mock my parents. I am writing it because I realized the gap is on all of us who build, write about, and use this technology daily.
If someone you love thinks AI is a search engine, they are not wrong about the interface. They are wrong about everything underneath it. And the industry has done almost nothing to fix that, because the current design is extremely profitable. One answer. A citation nobody clicks. No friction. No understanding required.
My parents trust AI completely, and they are not alone. The people who trust this technology the most are the people who understand it the least, and nothing about the way it is built is trying to change that. The interface was optimized for confidence, not accuracy, and that design is extremely profitable. For the builders, and now for the criminals.
My parents do not need to understand transformers. They need one habit, repeated until it is reflex: the confident voice is not a source, it is not a family member, and it is not the bank. It never was.
About The Author
