In this article6 sections
A small business can use artificial intelligence for the text work that eats up the most time: answering messages, putting together quotes, writing product descriptions, reading long documents, summarizing and sorting. In all of these, the AI writes a first draft and a person reviews it. It can't decide for you, it doesn't know anything about your business that nobody gave it, and sometimes it's wrong with complete confidence.
This guide covers uses that a hardware store, a clothing shop, a medical office or a distributor in Ecuador can try this week with the chat tools that already exist. If what you want is for a task to run on its own, connected to your systems, that's covered in the guide on automating tasks with AI (in Spanish).
What it does well and what it does badly
The AI everyone is talking about, the one behind ChatGPT, Gemini and similar tools, is a language model: it predicts text. That's why it's very good at writing, rewriting, summarizing and understanding what someone meant. And that's exactly why it can make things up.
| Does well | Does badly |
|---|---|
| Drafting from the information you give it | Knowing prices, stock or policies you didn't give it |
| Changing the tone of a message | Doing long calculations without review |
| Summarizing a long document | Quoting laws, articles or figures without errors |
| Turning messy text into a table | Deciding what's right for your business |
| Sorting messages by topic | Answering for something you sign |
OpenAI explains that its models can give false answers confidently because the way they are trained and evaluated rewards guessing over saying "I don't know." A better question won't fix that; reviewing the output is how you handle it.
Concrete uses, day to day
Messages to customers
You write the idea in two lines and the AI turns it into a clear, friendly message: a delay notice, a payment reminder, a reply to a negative review. It works better when you give it the real context ("the order arrives Thursday, not Tuesday, because the supplier in Guayaquil fell behind") and ask for a tone ("direct, no over-the-top apologies").
What to avoid: letting it invent a compensation or a date. You decide those.
Quotes and proposals
You paste the items, quantities and prices from your price list and ask it to put the quote together with an introduction, payment terms and how long the quote is valid. The AI saves you the writing; you copy the numbers from your system and check them one by one. A wrong total on a quote is your problem, not the tool's.
Product listings and descriptions
If you sell online, AI turns a dry spec sheet into a description people understand: what it's for, dimensions, materials, care. You give it the manufacturer's data and tell it not to add anything that isn't there. Check dimensions and compatibility: that's where it makes the most mistakes.
It also helps turn a messy inventory ("hammer drill 1/2 650w brand x") into a consistent format with name, brand, power and category.
Reading and summarizing documents
A lease, the terms of a public procurement process, internal company rules, an equipment manual. AI gives you a summary and helps you find the clauses that matter to you: deadlines, penalties, warranties. Use it to know where to look, not as a substitute for reading those clauses yourself, or for a lawyer when money is at stake.
One practical detail: OpenAI warns that ChatGPT may not read values from tables in images or scans correctly. If the numbers matter, upload a spreadsheet or a text file instead.
Sorting and organizing
You paste fifty reviews, emails or customer comments and ask it to group them by topic and tell you which ones repeat. In ten minutes you know that half of the complaints are about delivery time, not the product.
Writing copy for your website and social media
FAQs for your website, the text for a promotion, a business introduction for a profile. You tell the AI what you sell, to whom and what makes you different, and ask for three versions. You keep the one that sounds like you and edit it. Avoid publishing the text exactly as it comes out: it shows, and it tends to fill up with adjectives that say nothing.
Preparing for a difficult conversation
Before negotiating with a supplier or replying to an upset customer, you can ask the AI to play the other side and raise the objections you might hear. It won't tell you what to do, but it helps you show up with your arguments in order.
Summarizing meetings and calls
From a meeting transcript, AI pulls out the agreements, the open items and who is responsible for each. Before you record, ask the people taking part for permission.
Where it fails, in practice
- Facts that look true. An article number in a law, a reference price, a date. Always check them against the source.
- Math. Totals, percentages, VAT. Do them in a spreadsheet or in your system.
- Outdated information. The model may not know about recent changes to regulations or rates.
- The wrong tone. A correct message can sound cold or overdone to your customer. Read it before sending.
- What you didn't tell it. If it doesn't know you don't deliver on Sundays, it will write as if you do.
Protecting your business's and your customers' data
Whatever you type into an AI tool leaves your computer. Before you paste information:
- Check which account you're using. On personal ChatGPT accounts, conversations may be used to improve the models unless you turn that off in data controls. OpenAI states that on its business plans and its API it does not train on your data by default.
- Remember that some conversations are read by a person. Google states in its Gemini privacy hub (in Spanish) that human reviewers read some conversations, and recommends not entering confidential information.
- Remove what isn't needed. To draft a reply to a complaint, you don't need the customer's full name or national ID number.
- Don't paste passwords, card details, your SRI (Ecuador's tax authority) credentials or full bank statements.
In Ecuador, your customers' data is protected by the Ley Orgánica de Protección de Datos Personales (in Spanish), the country's personal data protection law. Using an AI tool doesn't take that responsibility off your hands.
How to start with little risk
- Pick a writing task you do every week that has no consequences if the draft comes out wrong: product descriptions, answers to frequent questions, summaries.
- Write down your fixed information in a document: hours, payment methods, delivery areas, policies. Paste it at the start every time, or save it in the tool's instructions if it allows that.
- Try it with real cases for two weeks. Compare how long the task takes with and without AI.
- Review everything before you send it. If you keep correcting the same thing, improve the instructions.
- Set rules for your team: what can be pasted, what can't, and that nothing goes to a customer without a person reading it.
- Only then add another task.
A person stays in charge
The rule behind everything above is simple: the AI proposes and a person decides. It writes the draft, you sign the message. It summarizes the contract, you read the clauses that matter. It sorts the complaints, you decide what to change.
When a use already works well with review and repeats hundreds of times a month, that's when it's worth thinking about connecting it to your systems, so the order or the reply goes out on its own, with checks in place. That's what the automation service is for, and if the use is serving customers on WhatsApp or on your website, the AI chat service.
Where to go next
Sources
Rules and each provider's terms change: check the source before you decide.
