Pre-launch · Training and the free Foundations exam are open now. Levels 1–5 open soon. Get notified on any level page.
20 min read
Before you can write a good prompt, you need an accurate picture of what happens when you type one. This module gives you that picture in plain terms. It matters the first time you ask an AI chat tool to draft an email, plan a trip, or explain a policy, and it matters every time after that, because most early mistakes come from expecting the tool to know things it was never given.
Most AI chat tools run on a large language model, or LLM. An LLM is trained on a very large amount of text and learns patterns in how words, facts, and formats tend to go together. When you send it something, it generates new text that fits those patterns and whatever you gave it.
Four terms are worth keeping apart:
That last point explains a lot. Because the output is generated, it can be fluent, well organized, and wrong. A prompt steers the answer. It does not check it. You can write a very clear prompt and still get a date that is off by a year or a fee that was never real. Treat the output as a draft from a fast writer who has read a lot and verified nothing.
Two more boundaries trip people up.
A new chat may be a blank start. Many tools only see the current conversation. Some products offer saved memory or project files, but even then you can't assume a detail from last week's chat is available, complete, or still true. If something matters, put it in the current chat.
Training has a cutoff. The model learned from text up to some date. It has no built-in knowledge of what changed after that, such as this month's prices, a new store schedule, or a flight that was rescheduled yesterday. Some tools can search the web or read a file you attach. If yours can't, any answer about current facts is a guess dressed as an answer. For anything time-sensitive, ask for a dated source or check it yourself.
Dana Ruiz, a recruiter at a regional hospital, needs to plan a two-day trip to a nursing job fair in Columbus, Ohio, on March 12–13. Her needs are concrete: she leaves from Pittsburgh, can't leave before 7:00 a.m. on the 12th because of an early call, needs a hotel within a 15-minute walk of the convention center, has a $180 per night lodging limit, and has to be back by 6:00 p.m. on the 13th. She opens a fresh chat.
Before:
Plan my trip to the Columbus job fair like we talked about. Find me
the best train and a good hotel.
That prompt depends on a past conversation the tool may not see. "Best" and "good" have no measurable meaning. And "find me the best train" invites the model to produce a schedule from memory, which may be out of date or invented.
After:
I'm planning a work trip and need a draft plan I can check.
Trip facts:
- From Pittsburgh, PA to Columbus, OH convention center
- Depart no earlier than 7:00 a.m. on March 12
- Back in Pittsburgh by 6:00 p.m. on March 13
- Hotel: within a 15-minute walk of the convention center,
$180 per night or less
Task: Lay out my travel options (driving, bus, or rail) as a short
list, with the questions I should check for each. If you cannot see
current schedules or rates, say so and do not give specific times
or prices. List the hotel requirements as a checklist I can use
when I search.
Why this is stronger. Dana translated each trip need into a requirement the output can be checked against: times, distance, and a dollar limit. She gave the new chat its own brief instead of relying on memory. And she told the tool what to do about freshness, so it won't present an old timetable as today's. The plan she gets back is a starting point, and she still confirms times and rates on the carriers' and hotels' own sites before booking.
Setup. Pick a real or made-up trip: a one-day visit to a client, a weekend at a family event, a training course in another city.
Task. In a text editor, write down your needs as plain notes first: where from, where to, dates, earliest and latest times, budget, and any must-haves. Then turn those notes into a prompt like Dana's After version. Paste it into any chat tool and read the reply.
What a good result looks like. Every need in your notes appears in the prompt as a specific requirement with a number or a clear condition. The prompt doesn't refer to earlier chats. The reply either cites a dated source for times and prices or says it can't confirm them. If the reply gives you a confident train time with no source, you've found exactly the gap this module is about.
A polished answer feels trustworthy. Notice when you're relaxing because the formatting looks professional. Formatting is generated too.