Email, Shopping, and Daily Decisions

By the end of this lesson, you’ll be able to spot the AI already running quietly inside your inbox, your shopping apps, and your maps — and you’ll know how to make it work harder for you instead of just watching it happen. You’ll also write two prompts you can reuse every week: one that clears your email backlog and one that ends the “which one should I buy” spiral. No new apps, no subscriptions.

The idea in plain English

In Module 1 we said AI is software that finds patterns in huge amounts of data and uses those patterns to make a guess. Here’s the part most people miss: you’ve been living with that for years already.

When Gmail drops a message straight into Promotions, that’s a prediction. Nobody at Google read your email and decided it was junk. A model looked at the sender, the words, the links, and the fact that a few million people ignored similar messages, and it guessed you would too. When Amazon shows you “customers also bought,” that’s a prediction. When your phone says the drive is 34 minutes but suggests a different route “to save 8 minutes,” that’s a prediction built from live location data off thousands of other phones on the same road.

So there are two kinds of AI in your daily life, and it helps to keep them separate in your head:

• Invisible AI — already working, no prompt required. Spam filters, autocomplete, fraud alerts on your credit card, “you might also like,” the ETA in Google Maps.
• Invited AI — you go to it and ask. ChatGPT, Claude, Gemini, Perplexity. This is the one almost nobody uses for ordinary chores, and it’s where the real time savings are hiding.

Most beginners assume AI is only for big, impressive tasks — writing a novel, coding an app. The honest truth is that its best everyday use is small and boring: drafting the reply you’ve been avoiding for four days, or comparing two vacuum cleaners so you stop reading reviews at midnight.

How it actually works

Your email. A spam filter is a classifier — a model whose only job is to sort things into buckets. It was trained on hundreds of millions of emails that humans already labeled “spam” or “not spam,” and it learned which signals go with which bucket. Every time you hit “Report spam” or drag something out of Promotions, you’re handing it another labeled example. That’s why your filter gets better over time and why a brand-new email address gets it wrong more often. Gmail’s Smart Reply and the “Help me write” button work differently — those are the same kind of language model as ChatGPT, predicting the next likely words based on the message you’re replying to.

Your shopping. Recommendations mostly run on something called collaborative filtering. The plain-English version: the system doesn’t understand what a cast-iron skillet is. It just notices that people who bought the thing you bought also tended to buy a certain seasoning oil, and surfaces it. It’s pattern-matching on behavior, not judgment about quality. This is worth internalizing, because it explains a lot of weird recommendations — and it explains why the top result is often the item with the best margin or ad spend, not the best product.

Your daily decisions. Maps, weather apps, and “best time to leave” features are prediction models fed by live data. Your phone anonymously reports how fast it’s moving down a road; multiply that by everyone else’s phone and you get real-time traffic.

Here’s the connection to invited AI: those systems are optimizing for their goal — engagement, sales, throughput. A chatbot you prompt yourself is optimizing for the question you actually asked. That’s the whole advantage. You get to set the goal.

One caution before we do it. A chatbot doesn’t browse your inbox or check live prices unless it explicitly has that ability turned on, and it can state a wrong price or a discontinued model with total confidence. Use it to think — to draft, compare, and pressure-test — and verify any hard fact (price, availability, dosage, deadline) at the source.

Try this now

Set aside ten minutes and actually do these. Reading about it doesn’t stick.

1. Open a free chatbot. ChatGPT (chat.openai.com), Claude (claude.ai), or Gemini (gemini.google.com). All three have a free tier and all three handle this fine.
2. Find one email you’ve been avoiding. Everyone has one — the awkward decline, the follow-up on an unpaid invoice, the reply to a client who’s annoyed. Copy the text of the email you received.
3. Paste this prompt, then paste their email underneath it. Fill in the two bracketed parts before sending:

   You are helping me write a work email. I am a [your role, e.g. freelance photographer].
   
   Write a reply to the email below. Requirements:
   - Under 120 words
   - Tone: warm but direct, no corporate filler
   - Outcome I want: [e.g. politely decline but keep the door open for next year]
   - Do not apologize more than once
   - Plain sentences. No "I hope this email finds you well."
   
   Give me two versions: one slightly softer, one slightly firmer.
   
   Here is the email I received:
   ---
   [paste the email here]
   ---

   Read both versions. You will almost certainly want to change a sentence or two — do that. The goal isn’t a perfect email, it’s skipping the blank-page paralysis that made you avoid it for four days.

4. Now a purchase you’re stuck on. Pick something real that you’ve been researching — a laptop, a car seat, a mattress, a coffee maker. Use this:

   I'm deciding between these options:
   1. [Product A, with price]
   2. [Product B, with price]
   
   My situation: [e.g. two kids under 5, small apartment, I use it daily, I care more about durability than looks].
   My budget ceiling is [amount].
   
   Please:
   - List the 4 factors that actually matter for someone in my situation
   - Compare both options on those 4 factors only, in a table
   - Tell me what would make me regret each choice in 2 years
   - Ask me any question you need before recommending one
   
   Flag anything you're unsure about or that may be out of date.

   That last line matters. It gives the model permission to say “I’m not certain about current pricing,” which is far more useful than a confident wrong number.

5. Answer its follow-up question. This is the step beginners skip. When it asks “how often will you travel with it?”, answer honestly. The second response is usually twice as good as the first.
6. Spend two minutes on your inbox settings. In Gmail, unsubscribe from three newsletters you never open. You’re not just cleaning up — you’re retraining the classifier on what you actually want.

Common mistakes

• Treating it like a search engine. Typing “best blender” gets you a generic listicle. Search rewards short queries; chatbots reward context. Tell it your budget, your kitchen, your priorities. Three extra sentences roughly doubles the quality of what comes back.
• Accepting the first draft. The first answer is a starting point, not a final product. Reply with “too formal, make it sound like a real person” or “cut it in half.” Treating the conversation as one-shot is the single biggest beginner mistake, and it’s free to fix.
• Trusting prices, stock, and specs without checking. A model may confidently quote a price from a year ago or describe a model that was discontinued. Use it to narrow the field from twelve options to two, then open the actual retailer’s page to confirm the details.
• Pasting things you shouldn’t. Free chatbot tiers may use your conversations to improve their models. Don’t paste account numbers, passwords, Social Security numbers, medical records, or a client’s confidential contract. Rewrite the sensitive bits as placeholders — “[client name],” “[invoice amount]” — and swap the real details back in yourself afterward.
• Assuming the recommendation is neutral. “Top pick” on a shopping site often means “highest margin” or “paid placement.” The algorithm is optimizing for the store’s revenue, not your satisfaction. Getting a second opinion from a chatbot that isn’t selling you anything is genuinely valuable — as long as you verify the facts.

Key takeaways

• You already use AI daily — spam filters, recommendations, and traffic ETAs are all predictions made from patterns in data.
• Invisible AI serves the company’s goals; invited AI serves the goal you state in your prompt. That’s why prompting is worth learning.
• Context beats keywords. Tell the chatbot your role, situation, budget, and constraints instead of typing a two-word request.
• Always plan a second turn — push back, ask for shorter, ask what you’d regret. The follow-up is where the quality shows up.
• Verify hard facts at the source, and never paste passwords, account numbers, or confidential material into a free chatbot.

Next up in this module: Health, Cooking, and Home — where AI shows up in your kitchen and your medicine cabinet, and where you should be more careful than you’d expect.