Using an AI assistant for everyday business tasks

You can ask an AI assistant to turn your offer details into a social post, summarize customer inquiries, or help fit marketing into the time you have. You type what you need in ordinary language, give it the relevant facts, and review the answer before using it.

The examples below use Sjena, a fictional product photography studio in Zagreb, Croatia. Its offer and inquiry records were invented for these exercises. They show how you can ask for help and where an answer still needs your attention.

Turn offer details into a social post

Start by supplying the details the post must preserve. For a studio like Sjena, you could type:

Write a short social post for Sjena, a product photography studio in Zagreb. The package includes eight photos, two backgrounds, and one round of revisions. Ask readers to send a list of the products they need photographed. Don't add a price, delivery date, customer review, or guarantee.

In the writing exercise, the same offer details were given to three AI models, the systems that generate an assistant's answers. One reply included this sentence, translated into English from the original Croatian:

The package includes eight photos, two backgrounds of your choice, and one round of revisions according to your wishes.

The Claude model had added a small promise: the customer could choose the backgrounds. The supplied package specified two backgrounds without offering that choice. The correction is to remove “of your choice” and keep “two backgrounds.” The models from OpenAI and Google preserved that part of the offer. None of the three invented a price or deadline.

That extra choice could matter when a customer books. If your package uses two fixed backgrounds, the post has offered something you don't provide. You now have to correct the copy or explain the difference to the customer.

When checking your own draft, compare every promise with your source information. Look beyond prices: quantities, revisions, delivery times, included services, and words such as “custom” can all change what a customer expects.

Give the assistant an explicit boundary: use only the supplied offer details and ask about anything missing. Then check the answer against those details anyway. A fluent sentence still needs your approval before it speaks for your business.

Summarize inquiries without counting the same one twice

An assistant can also work through a small table. This exercise used five rows of fictional inquiries. Each letter identifies one inquiry, so the second appearance of A is a duplicate.

An instruction for this table would be:

Count each inquiry once and add up accepted quote values by source. Show what you counted and excluded. Payment hasn't been confirmed, so don't describe accepted quotes as money received.

The result should show four unique inquiries. Accepted quotes total €500: €200 from Instagram and €300 from a referral. The rejected and pending quotes don't belong in that total.

All three answers handled this correctly. They also recognized that an accepted quote doesn't confirm payment. The table can't tell you how much money reached the studio's bank account or whether a marketing channel paid for itself.

Keep the original records beside the answer. Comparing the rows with the calculation makes it easier to find a duplicated entry or a pending sale presented as completed business.

This was a small calculation with an explicit instruction to count A once. Getting it right is useful, but it doesn't establish reliability with a large or messy spreadsheet.

Check what a plan expects you to have ready

The planning task gave the studio two hours a week for marketing and asked for a two-week plan to attract relevant inquiries from small online shops. Every answer stayed within that time budget.

The assumptions needed more attention. All three plans expected the studio to have work ready to show in a portfolio and to be able to contact online shops. Google's Gemini model also proposed reviewing email open rates, which assumes the studio can track them.

If you still need to photograph portfolio samples or prepare a contact list, those jobs belong in the schedule. Otherwise, a two-hour plan can leave you with preparation it never counted. If you don't track email opens, ask for a check you can perform, such as recording replies and relevant inquiries.

For your own plan, list what you have ready and add an instruction like this:

Allow two hours per week, including preparation. Give each activity a time allowance. Flag anything that needs photos, contacts, or tracking that hasn't been confirmed. Include time to count replies and relevant customer inquiries, and suggest an activity to postpone.

Read each activity as something you would actually have to do next week. No campaign was run for this exercise, so these answers show proposed work, not evidence that the plans would bring in customers.

Decide whether the help is worth paying for

Choose a task you do regularly and know well enough to check. Use the language you normally work in. If you're comparing assistants, give them the same facts and instructions.

Use fictional information for your first trial. Before entering real customer details, check the privacy terms of the service you're using.

Record the time you spend preparing instructions, checking the answer, and correcting it. Try the same type of task again with different source material before treating a good first answer as dependable.

Pay when you can name the recurring job the assistant helps you finish and the improvement is worth the cost to your business. If checking and fixing its answers takes as much effort as doing the job yourself, keep looking or choose a different task for it.

About the examples

The examples use nine AI answers collected in Croatian on September 17, 2026. The quoted answer is translated into English. App features and subscriptions weren't compared, and this small set of answers doesn't establish which tool is best or how reliably it will handle your work.