The AI chat tools most people use today are built on large language models. They learn patterns from huge amounts of text, then write a reply by predicting, one piece at a time, what text is likely to come next. That makes them very good at drafting, summarizing and rewording. It does not make them a database of facts, a search engine you can fully trust, or a person who understands your business.
Checklist
I know that chat tools write by predicting likely text, not by looking up facts.
I use them for drafting, summarizing, rewording, organizing and brainstorming.
I give the tool the facts about my business instead of expecting it to know them.
I check every fact, number and date before I use it.
I keep decisions that affect customers, money or people with a person.
Tips
Think of the tool as a fast first-draft writer, not an expert.
Give it the facts you want used, every time.
Ask for two or three versions so you have options to choose from.
Read every reply as a draft that still needs your judgment.
Key questions
What does a chat tool's language model do when it writes a reply?
It predicts a likely next piece of text, over and over, based on patterns from its training data.
What is generative AI?
AI that creates new content, such as text, images, audio or video, after learning from existing content.
What does NIST call it when AI confidently states something false?
Confabulation, often called "hallucination."
Does a chat tool know your prices and hours?
No, not unless you give it that information.
Who should make decisions that affect customers or money?
Tell the tool who it should act as, give it the facts it needs, say exactly what you want, describe the format of the answer, and show an example if you have one. Then treat the first reply as a draft and ask for changes. A clear request with your real details almost always beats a short, vague one.
Checklist
I said what role the tool should take.
I gave it the facts it needs and told it not to add others.
I stated the goal of the answer.
I described the format and length.
I included an example when I had one.
I asked for at least one round of changes before using the result.
Tips
Write prompts the way you would brief a capable new hire.
Tell the tool to use only the facts you gave it.
Ask for two or three options when you are not sure what you want.
Keep a document of prompts that work and reuse them.
Key questions
What are the five parts of a useful prompt?
Role, facts, goal, format and example.
Why give the tool your facts?
It cannot know your business details, and it may fill gaps with something that sounds right but is wrong.
What is "few-shot" prompting?
Including one or more examples in your prompt so the tool can follow the pattern.
What should you do if the first reply is not right?
Ask for specific changes instead of starting over.
How should you handle a big, multi-part task?
Break it into smaller prompts, one step at a time.
AI chat tools write by predicting likely text, and likely text is not always true. So they sometimes state wrong facts, invented sources or made-up numbers with complete confidence. This is commonly called a "hallucination." The fix is a habit: treat every fact, figure, quote and source in an AI answer as unconfirmed until you have checked it against a reliable original.
Checklist
I marked every number, name, date, rule and source in the answer.
I checked each one against the original, authoritative source.
I opened every link and confirmed it says what was claimed.
I removed anything I could not confirm.
Where possible, I gave the tool my own facts to work from.
Tips
Treat AI facts as leads to check, not answers to publish.
Supply your own facts and tell the tool to use only those.
Always open the source; never trust a link you have not clicked.
When in doubt, leave it out.
Key questions
What is an AI "hallucination"?
AI stating false information confidently; NIST calls it confabulation.
Why do hallucinations happen?
The tool predicts likely text rather than looking up each fact, so likely sounding text can be wrong.
Is a confident answer more likely to be correct?
No. Confidence tells you nothing about accuracy.
How should you check a fact from AI?
Against the original, authoritative source, not another AI tool.
What should you do with a claim you cannot verify?
Never paste passwords, account numbers, Social Security numbers, card details or other secrets into an AI tool. Be careful with customer and employee personal information and with anything confidential to your business. Before you use any tool for work, read its data settings so you know whether your chats are stored, reviewed by people or used to improve the provider's models. When in doubt, remove names and details first.
Checklist
I have read the data settings for every AI tool I use for work.
I never paste passwords, account numbers, ID numbers, keys or card details.
I remove names, contact details and figures before pasting when they are not needed.
I know whether my plan's chats are stored, reviewed or used for training.
Anyone else using AI in my business knows the approved tools and the never-paste list.
Tips
If you would not email it to a stranger, do not paste it into an AI chat.
Replace names and numbers with placeholders like "the customer" or "[amount]."
Check settings again whenever you change plans or tools.
Add your personal details back in your own email or document, not in the chat.
Key questions
Name three things you should never paste into an AI tool.
Passwords, account or card numbers, and Social Security or other ID numbers.
Can chats be reviewed by people at the provider?
For some tools and settings, yes. Google says human reviewers see some Gemini app chats.
What is the simplest way to protect customer privacy when using AI?
Remove names and contact details before pasting.
Do business and personal plans always handle data the same way?
No. Terms often differ by product and plan, so check each one.
What should a one-page AI rule for your business include?
Approved tools, required settings and the never-paste list.
Let AI do the preparation: drafting, sorting, summarizing and suggesting. Keep a person in charge of decisions that affect customers, employees, money, legal commitments or your reputation. A simple test works well: if a mistake would be hard to undo or would affect someone's livelihood, a person makes the call, and AI is only an extra opinion.
Checklist
For each AI task, I have decided whether AI helps or decides.
Decisions about people's jobs, money or rights stay with a person.
AI does not make promises or quote terms to customers on its own.
Reviewers actually read AI output before approving it.
I can explain any decision to the person it affects.
Tips
Use AI to prepare decisions, not to make the important ones.
Watch for automation bias: polished output still needs a real read.
Keep anything that commits your business in writing with a person.
Revisit your list every few months as your experience grows.
Key questions
What is automation bias?
The tendency to defer too much to automated systems or to over-rate their output.
What kinds of decisions should stay with a person?
Ones that affect someone's job, money or rights, or that are hard to undo.
Does using AI change who is responsible for a decision?
No. The business remains accountable for what it does and says, whether a person or a tool produced the first draft.
What can AI do to help with a decision?
Prepare it: summarize, sort, draft options and list questions.
What is a quick test for whether a person should decide?
Ask whether the mistake could be undone easily and whether it affects someone's money, job or rights.
A well set-up AI receptionist or website chat assistant does four jobs: it answers common questions from information you give it, captures the caller's or visitor's details, helps book an appointment or next step, and hands the conversation to a person when it should. It should say clearly that it is an AI assistant. It works best on the routine questions you answer every day, with firm limits on what it is allowed to say and promise.
Checklist
The assistant introduces itself as an AI assistant.
It answers only from approved information I wrote.
It has written limits on prices, promises and advice.
It hands off to a person on request and for sensitive topics.
Captured details go somewhere a person checks every day.
I review transcripts and update its information regularly.
I get legal advice before using an AI voice for outbound calls.
Tips
Start the conversation by saying it is an AI assistant and how to reach a person.
Keep its answers short and tied to your approved information.
Pass the full conversation along at hand-off so customers do not repeat themselves.
Update the assistant whenever your hours, services or policies change.
Key questions
What are the four jobs of an AI receptionist?
Answer, capture, book and hand off.
Should an AI assistant tell people it is AI?
Yes. Being upfront protects trust.
What did the FCC rule in February 2024?
That AI-generated voices are "artificial" voices under the Telephone Consumer Protection Act.
When should the assistant hand off to a person?
When asked, when the question is outside its approved information, or when the topic is sensitive or involves money.
Where should an assistant's answers come from?
Approved information you write and keep up to date.
Pick a task you repeat the same way many times a week, write down its steps, and set up an automation: a trigger that starts it, a series of steps that do the work, and a review point where a person checks anything important before it goes out. AI fits inside those steps where judgment about language is needed, such as reading, sorting or drafting. Start with one small task, test it carefully, and add more once it runs well.
Checklist
I picked one task that repeats often and follows the same steps.
I wrote out the trigger, the steps and the review point.
I tested it on past examples before using it live.
Anything going to customers or affecting money is reviewed by a person.
Exceptions go to a person, not into a guess.
I spot-check results regularly, even after it runs smoothly.
Tips
Write the process on paper before you build anything.
Automate one step at a time and test each one.
Send anything unusual to a person by default.
Put a reminder on your calendar to check each automation monthly.
Key questions
What are the three parts of an automation?
A trigger, a series of steps and a review point.
What does AI add to automation?
Steps that read and write language, such as sorting, extracting details and drafting.
What makes a good first task to automate?
It repeats often, follows the same steps, has easy-to-fix mistakes and takes real time.
Use AI images and video to illustrate ideas, not to show things that did not happen. Never put a real person's face or voice into AI content without their clear permission. Follow each platform's AI disclosure rules, which generally ask you to label realistic AI content. And check every AI image closely before posting, especially any words in it, because AI tools can get text, hands and small details wrong.
Checklist
The AI content illustrates an idea; it does not pretend to show real work, products or people.
I have written permission for any real person's likeness or voice.
I followed the platform's disclosure rules for realistic AI content.
I checked every word, hand, face and logo at full size.
I added my own text in a design tool where possible.
Tips
Generate images without text, then add words yourself.
Use stylized or illustrated looks when you do not need realism.
Keep a simple record of which posts used AI images.
When in doubt, add a short "created with AI" note.
Key questions
What is the simplest rule for AI images in business?
Use them to illustrate ideas, never to show things that did not happen.
What does YouTube require creators to disclose?
Realistic AI content, such as a real person appearing to say or do something they did not, or a realistic scene that did not happen.
What should you check most carefully in AI images?
Any text, plus hands, faces, logos and physical details.
What do you need before using a real person's likeness in AI content?
Their clear, written permission.
What are Content Credentials?
An open standard from C2PA that records where digital content came from and how it was edited.
Start with the job you need done, not the tool. Then ask every option the same short list of questions: what exactly it does, how it handles your data, how you review and correct its work, what happens when it is wrong, what it costs in full, and how you leave if it does not work out. Ask for a trial on your own real tasks, and be wary of any vendor that promises results it cannot show you.
Checklist
I wrote down the problem and how I will know it is solved.
I asked every option the same questions and recorded the answers.
I have the vendor's data policy in writing for the exact plan I will use.
I can review the tool's work and set rules for it.
I know the full cost and how to leave.
I tested it on my own real tasks before signing.
Tips
Ask for evidence, not adjectives.
Get data terms in writing for the exact plan you will buy.
Trial on your own real examples, side by side.
Prefer short commitments until a tool has proven itself.
Key questions
What should you write down before looking at any AI tool?
The problem you need solved and how you will know it is solved.
Name two data questions to ask a vendor.
Is my data used to train AI, and can people at the vendor see my conversations?
Why ask what other services a vendor relies on?
AI products often depend on third-party parts, and those parts affect accuracy and data handling.
What is a major warning sign in a vendor's pitch?
Guaranteed results, such as a promised number of new customers.
How should you trial an AI tool?
On your own real tasks, side by side with other options, scored against your goal.
Start with one task, not a company-wide overhaul. Write a few simple rules, measure how the task performs before and after, train the people involved, and review the results after a set period. Keep what works, fix what does not, and only then add the next task. Slow and steady adoption builds skill and trust at the same time.
Checklist
I chose one task and wrote down the goal in one sentence.
I have a one-page AI policy covering tools, never-paste items, checking, decisions and disclosure.
I recorded "before" numbers.
Everyone involved has been shown the task, the prompts and the rules.
I have a review date on my calendar.
I will not add a second task until the first one has been reviewed.
Tips
Pick a low-risk, high-repetition task to start.
Keep the policy to one page so people actually read it.
Record a "before" number, even a rough one.
Treat each review as a real decision: keep, change or stop.
Key questions
What are the four functions in NIST's AI Risk Management Framework?
Govern, Map, Measure and Manage.
What is the best way to start using AI in a business?
With one low-risk, repetitive task.
What should a one-page AI policy cover?
Approved tools, never-paste items, checking facts, which decisions people make, disclosure and who to ask.
Why record "before" numbers?
So you can tell whether the AI task actually improved anything.
When should you add a second AI task?
After the first one has been reviewed and a decision has been made.
This site uses essential browser storage to function. With your consent, Sherwood also uses Microsoft Clarity, an optional analytics tool, to understand how visitors use the Site. Set your preference for this and any other optional technology, or read the Cookie Policy.
Chat with Sherwood
Hi, I'm Sherwood's AI assistant. Ask me anything about what Sherwood builds, or I can set up a free consultation with John, the founder. What brings you in?
You are chatting with an AI assistant. Messages are processed by our AI service provider to respond and are not stored by Sherwood. See our Privacy Policy.
