You will be able to set a checking level for any task based on who will see the output and what an error would cost.
Ravi does marketing for a group of six cafes. On a normal Tuesday he might use an assistant a dozen times: brainstorming names for a new seasonal drink, tidying his notes from a supplier meeting, drafting an Instagram caption, summarising customer reviews for his manager, and checking what the rules say about advertising a promotion as "free". If he checked every one of those answers as carefully as he could, he would get nothing else done. If he checked none of them, sooner or later one would embarrass him in public.
The way out is to decide how hard to check before you start reading, based on what is at stake. This module is about checking, and it starts with that decision.
A simple scale covers most work. There are three levels, and each one adds to the one before.
A light read is enough for drafts only you will see. Brainstorming, notes for yourself, a first pass at ideas you will develop further. You read the output for usefulness and for anything that looks obviously off, and you move on. Ravi's drink names get a light read. If one of them turns out to be a competitor's product name, he will find out long before it reaches a menu, because a light-read output is never the last step.
A full read with a fact check is needed for anything that leaves your hands: sent to a client, a manager, a colleague in another team or the public. You read every sentence as the recipient will, and you check every fact you could be asked about: names, dates, figures, claims about what someone said or what a product does. Ravi's review summary for his manager gets a full read, and he checks each quoted complaint against the actual review.
A primary source check is needed for anything involving money, law, health or someone's job. You do not stop at the assistant's answer, or at a website that repeats it. You check the claim against the source that actually decides it: the law itself, the government agency's own page, the contract, the official policy. Ravi's question about advertising a promotion as free gets this level, because a wrong answer could mislead customers and put the business in breach of the rules.
Two reasons put these four areas at the top. The first is the cost of an error. A wrong figure in a caption is embarrassing. A wrong figure in a staff member's pay calculation, a wrong clause in a contract, or a wrong statement about someone's medical entitlement causes real harm to a real person.
The second is that these are exactly the areas where assistants are most likely to be wrong. AI fundamentals lesson 6.2, Where invented answers are most likely, explained that risk rises with specific details such as exact numbers, dates and legal clauses, and with niche, local or recent topics. Singapore rules on employment, tax, CPF and consumer protection are all specific, local and regularly updated. An answer about them can be fluent, confident and a year out of date.
So for these areas, treat the assistant as the start of your research. It can explain a concept, suggest what to look for or help you read a document you have found. The answer you rely on comes from the primary source: the Ministry of Manpower for employment rules, IRAS for tax, the CPF Board for CPF, the law itself on Singapore Statutes Online. AI fundamentals lesson 6.3, A five-minute routine for checking any answer, gives a quick method for doing this.
Whatever level you choose, you are accountable for what you send. A client who finds a wrong figure in your proposal holds you responsible, and when an email to staff gets a policy wrong, it is your name at the bottom of it. Nobody will accept that the assistant wrote it as a reason the mistake happened, any more than they would accept that a spellchecker sent the wrong attachment.
This is not a reason to avoid using assistants for important work. It is a reason to match the checking to the stakes, every time, so that the important things get the attention they need and the trivial things do not eat your day.
The habit worth building is to decide the level before you read the output. If you leave it until afterwards, a fluent, confident answer tends to talk you down to a lighter level than it deserves. Ask two quick questions first. Who will see this? And what would a mistake cost, and whom?
Ravi now does this in his head as he types the request, and it takes seconds. Drink names are for him alone and cost nothing if wrong, so they get a light read; the summary leaves his hands and could mislead a decision, so it gets a full read; and the promotion question touches law and money, which sends him to the primary source.
Your own list will look different from Ravi's, but the questions are the same. Think through the past week and pick out the different things you used an assistant for, or might have, from the throwaway to the serious.
List ten things you might use an assistant for and assign each a checking level of light, full or primary source.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).