AI Made Simple

01

What you will learn

Contrast a search engine with a generative answer.
Pick the right tool for lookup versus drafting.
Define a practical trust rule: trust the draft's shape, verify the claims.
Ask for uncertainty without getting a pile of fake humility.
Leave Level 1 with a personal rule you will actually follow.

02

Concept

A search engine returns documents. You still have to read them, but you can see the source sitting there. ChatGPT returns a composition. That composition may be clearer than any single page — and it may blend a real idea with a detail that never existed.

Search is the better first move when you need the page itself: the official docs, the hours, the form, the wording of a policy, the dataset. ChatGPT is the better first move when you need thinking turned into a shape: an outline, a tighter email, a list of questions, a comparison structure.

Many ChatGPT plans can search as well. When that happens you are using both tools, and you should still open what matters. A link in a chat is a lead, not a finished verification. A link that does not open is not a source.

Trust, in this course, is not a mood. It is a rule. You may trust the model to produce a draft you can react to. You may not trust it as the last word on a claim that would hurt someone if it were wrong.

03

Why it matters

Beginners ask ChatGPT questions they should have searched, then search for things they should have asked a model to structure. The mismatch wastes time and creates fake certainty.

A written trust rule keeps you fast on low-stakes drafts and slow on the sentences that carry consequences.

04

Step-by-step

01Name the object you needA document you can cite, or a draft you will edit? If you need the document, start with search or the official site. If you need the draft, start in the chat — with your material.
02If the chat searched, open the pageSkim the actual page for the sentence you are about to repeat. Check the date. Check that the site is the one you think it is.
03Ask for a claim listSay: 'List factual claims in your answer, each with how sure you are and what I should open to check it. Do not give me sources you cannot actually point to.'
04Separate stakesLow stakes: a birthday-note draft, a brainstorm, a rephrase. High stakes: anything a customer, a boss, a student, or a doctor might act on. High stakes get an outside check. Always.
05Write your rule downOne or two sentences, in your words. Example: 'I can send a rewrite of my own words. I cannot send a statistic, a quote, or a policy claim until I have seen it somewhere other than the chat.'

05

Real-world example

The policy sentence

Luis manages a community workshop. He wants to tell members whether a city permit changed this year. He asks ChatGPT, gets a crisp yes, and almost pastes it into the newsletter.

The answer cites a '2024 municipal update' with no working link.
Luis treats that as a lead, not a fact. He searches the city's own site.
He finds a page that does not say what the chat said. He deletes the sentence.
He uses ChatGPT only after that, to turn the real page — pasted in — into a plain-language note for members, with a link back to the city.

06

Prompt examples

Beginner

I need to know whether [TOPIC] is currently true.
Do not guess.
Tell me what official page a human should open, and what words I should look for on that page.
If you are not sure the page exists, say so.

Improved

Purpose: Help me verify a claim before I publish it. Do not be the source.
Role and reader: Research assistant. I will read the pages myself.
Inputs: Claim: [CLAIM]. Audience who might act on it: [AUDIENCE].
Method: List where a primary source would live. Say what evidence would confirm it and what would refute it. If you search, give the link and the date you think the page carries. If you did not search, say "not retrieved."
Expected form: A checklist of at most six lines. No closing pep talk.
Quality bar: Never invent a URL, a case name, or a quotation.

Advanced

Purpose: Decide whether this task starts in search or in chat.
Role: Editor.
Reader: Me.
Inputs: Task: [TASK]. What I will do with the result: [USE].
Method: Choose Search first, Chat first, or Both. Justify in three sentences. If Both, say what the search must return before the chat is allowed to draft.
Expected form: Choice, justification, and the exact first query or first prompt I should run.
Quality bar: If someone could be harmed by a wrong answer, Both is the minimum, and I remain the checker.

Expert

Purpose: Write my personal trust rule as a card I will reuse.
Role: Instructor who has seen people skip checks when they are busy.
Reader: Future me, on a deadline.
Inputs: My work: [ROLE AND WORK]. Typical high-stakes outputs: [OUTPUTS].
Method: Draft a rule of at most 40 words. Then give two examples from my work: one I may send after a light edit, one I may not send until I open a source. Then give the VERIFY questions in my own vocabulary, not as a slogan.
Expected form: Rule / May send / May not send / Three check questions.
Quality bar: The rule must be specific enough that a tired version of me can follow it. No generic 'be careful.'

07

Before and after

Weak

Did the city change the permit rule this year? Answer yes or no and write my newsletter paragraph.

Strong

Do not answer yes or no from memory. Tell me which official source should settle whether [PERMIT RULE] changed, and what phrase I should look for. After I paste the real paragraph from that page, rewrite it for workshop members at a ninth-grade reading level and include the link I give you. If I have not pasted a source, refuse to state what the rule is.

The first prompt forces a yes or no and a publishable paragraph, which is how fictional policy spreads. The second refuses the verdict until a source is on the table, then uses the model for the job it is good at: plain language.

08

Common mistakes

Asking for a yes-or-no on a live fact and publishing the answer.
Treating a generated bibliography as if you had read it.
Using search results inside the chat and never opening them.
Having no personal rule, so every task feels equally trustworthy.

09

Pro tips

If you cannot say where a fact came from, it is not ready.
'I don't know' from the model is a gift. Do not punish it by demanding a guess.
Official documentation beats a blog that quotes a blog. The model will not always rank them that way unless you say so.
Your trust rule should mention your actual outputs: the newsletter, the client email, the slide — not 'content' in the abstract.

10

Practice

Write your trust rule in 40 words or fewer. Then take one factual sentence you were about to use from a chat this month. Either open a source that supports it, or delete it. Record which you did.

11

Challenge

Find a claim the model states confidently about a public policy, a product limit, or a definition in your field. Check it against a primary page. Bring back one of these: the claim held, the claim was partly wrong, or the source does not exist. That third outcome matters most.

12

Knowledge check

13

Key takeaways

Need the page? Search or go to the source. Need a shape for thinking? Use the chat.
A citation you have not opened is not a citation yet.
Ask for claims and uncertainty, then check outside.
Write a trust rule tied to your real outputs.
Low-stakes drafts can move fast. High-stakes claims cannot.

14

Visual briefs

Production notes for diagrams. These are not screenshots of ChatGPT. If a brief asks for the product window, capture it live on your own account.

Page versus composition

Purpose. Make the search/chat choice obvious.

Show. Left: a list of documents. Right: one smooth paragraph with a question mark on each factual clause.

Design. Two-panel diagram. No imitation of a commercial search page or ChatGPT UI. Label the right panel 'composed answer.'

Trust card

Purpose. A personal rule learners can photograph.

Show. Blank lines: I may send… / I may not send until… / I will open…

Design. Index-card layout, copper rule on the left edge.

15

Film outline

The player above runs this cut. Captions are the lesson, not a fake recording of ChatGPT.

Opening hook: The newsletter almost shipped a permit rule that was never on the city site.
Teaching: Search retrieves, chat composes, and trust is a rule.
Demonstration: A yes-or-no policy question, then the refusal prompt, then a rewrite of a real pasted paragraph.
Practice: Learners write a 40-word trust rule.
Challenge: They fact-check one confident claim against a primary page.
Recap: Open the page. Then let the model help you speak clearly.
Next: Level 2 is hands-on: how to run a conversation so the first answer is actually usable.

Instructor script

When you need a document, go get the document. When you need a clearer shape for thoughts you already have, use the chat. The trouble starts when a composed paragraph pretends to be the document. If a link appears, open it. If a quote appears, find it. Write yourself a short rule you can follow on a busy afternoon: what you may send after a light edit, and what you may not send until you have seen it somewhere else. That rule will do more for you than any model name I could put on this screen.