This article covers what makes a strong prompt, the habits that get better results, ready-to-use example prompts for the things Sintra AI does most (posts, images, video, websites, emails, data, documents, Brain AI and automations), and what to do when a result isn't right.
Why your prompt matters
AI employees don't look things up the way a search engine does. They work out an answer from what you tell them and from what they find in Brain AI, so the quality of the result depends mostly on the quality of your request. A vague prompt like "write me an email" leaves the AI employee guessing, and guesses come out generic. A clear prompt with context gets you something you can use straight away.
Your first prompt sets the expectation. This matters most when you give an AI employee a new task, or when you want a different angle on a task it has done before. Even if it remembers how you did it last time, say clearly what should be different this time. If you don't set the expectation, the result can go in many directions.
For tasks you repeat, save the workflow as a skill. A skill holds your rules, expectations and steps, so even a short prompt like "write me an email" follows the way you want it done. See "Save repeat work as a skill" below.
What AI employees already know, and what they don't
Before you write a prompt, it helps to know where AI employees get their context from.
They already know:
Their job. Each AI employee has its own role and skills. Soshie knows social media, Penn knows copywriting, Dexter knows data. You don't need to tell them "you are an expert". Just pick the right AI employee, or use Team chat and the request is sent to the right specialist.
Your Brain AI. Your Brand Kit (logo, colours, brand voice), business profile, and the documents, images and videos you've saved. A complete Brain AI is the single biggest factor in output quality.
The current chat. Everything you've said and attached in this conversation.
They don't know:
What happened in other chats. Each new chat starts fresh. Instructions and files from an earlier chat don't carry over unless they're saved to Brain AI, as a rule or as a skill.
What's in your head. Your goal for this task, who it's for, what you've already tried, what a good result looks like, and which direction you want to take.
What's in your connected apps. Connecting Google Drive or another app doesn't copy its files into Brain AI by default. Upload or save what should guide your AI employees, or tell them which connected app to use and what to look for there.
Good to know: your feedback shapes the current chat only. If you want a correction to apply every time, say "Save this as a rule" or "Remember this is how to do it every time".
The four parts of a strong prompt
A strong prompt answers three questions: what do I want, why do I want it, and what does a good result look like? In practice, that means four parts:
A clear task. Say exactly what you want done, not just the topic. Not "email", but "a follow-up email to a customer who didn't reply to our quote".
Context. Who it's for, what the situation is, what's already happened, and what it's for.
Format and limits. Length, tone, structure, language, platform, number of items.
An example, when you have one. A past post you liked, an email that worked, a competitor's page you admire. An example shows tone and structure far better than a description.
You don't always need all four. A quick edit can be one line. But the bigger or more important the task, the more of these parts you should include.
A quick before and after
Weak: "Write a product description."
Strong: "Write a product description for our handmade soy candle 'Autumn Fig' for our online store. Our customers are women aged 30 to 50 who buy gifts. Keep the tone warm and simple, not flowery. Mention it burns for 40 hours, is made in small batches and comes in recyclable glass. Keep it under 80 words and end with a one-line gift suggestion."
The second prompt gives a task, an audience, a tone, three facts to use, a length and an ending. There's nothing left to guess.
Tip: not sure what information to give? Add "Ask me clarifying questions before you start." The AI employee will ask for the context you might have missed.
Step by step: writing your first strong prompt
Step 1: Pick who should do it
Open a new chat with the AI employee whose role fits the task, or type your request in Team chat if you're not sure. Not sure what each AI employee does? See What are Sintra AI employees and how do they work?
Step 2: Start with the task
Lead with what you want done. Skip greetings like "Hi, I'm back". They can make the AI employee reply socially instead of getting to work.
For example: "Write three Instagram captions for our new autumn menu."
Step 3: Add the context
Say who it's for, why, and what the AI employee should use. Point to Brain AI items by their exact title.
For example: "It's for our regular customers in [city]. Use the 'Autumn menu 2026' document and our brand voice from Brain AI."
Step 4: Set the format and limits
Say how long, which tone, which language, and how many.
For example: "Each caption under 50 words, friendly and a little playful, in English, with three hashtags. No emojis."
Step 5: Review and give specific feedback
Read the result and say exactly what to change, and what to keep. Then save anything you want every time.
For example: "Caption 2 is great. Make 1 and 3 shorter and less formal. Save 'no emojis in captions' as a rule."
General best practices
Give real tasks, not general questions
AI employees act. They write, create, build and send (with your approval). A real task gets you finished work, while a general question gets you advice.
For example: "Write five Instagram captions for our autumn menu" gets better results than "What makes a good caption?"
Say what you want, not only what you don't want
Instructions like "don't create a new logo" or "don't use stock people" often backfire, especially with images. The AI employee picks up the words "create" and "stock people", and the "don't" gets lost. Describe the result you want instead. For text, a short "avoid" list next to a clear request works well.
For example: instead of "Don't make up people," write "Use only the photos from my 'Team photos' folder in Brain AI." For text: "Keep it conversational. Avoid jargon and don't mention prices."
One task per message, one topic per chat
If you ask for four things in one message, you may only get two, or four rushed ones. Big jobs work best when they're broken into steps: research first, then the draft, then formatting. Keep each topic in its own chat. Answers stay focused and each message uses fewer credits. The AI employee can break a big job into steps by itself, but working out those steps uses more credits. You set the expectation and the depth you want; the AI employee focuses on getting the task done.
For example: "First, list 10 blog topic ideas for our bakery based on what's in Brain AI. Wait for me to pick 3 before writing anything."
Or, to let the AI employee plan it: "From what you know about me and how I like tasks done, research this in depth and work out the steps yourself. If you need to check my expectations, ask all your questions up front, then work without checking in until you're done."
Use numbers, not vague words
"Short", "a few" and "engaging" mean different things to different people. Numbers and concrete rules don't.
For example: instead of "Make it short and engaging," write "Keep it under 100 words, use two short paragraphs and end with a question."
Show an example
If you have something you like, share it. Paste it in, attach a screenshot, or point to it in Brain AI by its exact title.
For example: "Write our November newsletter in the same style and structure as the 'October newsletter' document in Brain AI."
Explain why a rule matters
When the AI employee knows the reason behind a rule, it applies it better in situations you didn't spell out.
For example: "Never promise delivery dates in replies, because our supplier times change every week."
Ask for step-by-step thinking on analysis
For analysis, comparisons or decisions, ask the AI employee to reason through it before answering. That way it works through the problem rather than jumping to the first answer that sounds right.
For example: "Compare these three supplier offers. Think through price, delivery time and minimum order step by step, then recommend one and explain why."
Iterate, don't start over
If the first result is close, say exactly what missed the mark and what to keep. Targeted feedback gets you there faster than rewriting the whole prompt.
For example: "Good structure, but the opening is too long. Cut it to one sentence and keep everything else as it is."
Start a fresh chat when a thread gets long
After many rounds of corrections in one chat, earlier instructions can get lost and fixes can start undoing each other. Long chats also use more credits per message. When that happens, open a new chat and give one clear, complete prompt with only the latest version of your file.
For example: "Here is the latest version of my price list presentation. Change only slide 4: replace the table with the new prices below. Leave every other slide exactly as it is."
Save what should apply every time
Anything you'd otherwise repeat in every chat belongs in Brain AI or in a rule: your tone, your spelling preferences, things to never say.
For example: "Save this as a rule: always use UK spelling and never mention competitor names."
Save repeat work as a skill
When a task works the way you like, ask the AI employee to save it as a skill. A skill keeps your steps, rules and expectations, so next time a short prompt is enough and the task runs the same way every time.
For example: "Save what we just did as a skill called 'Monthly newsletter', including the structure, the tone and the checks you ran."
Prompts per use case
The prompts below are ready to copy. Replace the parts in square brackets with your own details.
Social media posts and captions
Say which platform, what the post is about, who it's for, and which image or video to use. AI employees can create, schedule and publish posts to your connected social accounts, with your approval.
For example: "Create an Instagram post announcing our weekend brunch menu, using the 'Brunch table – overhead shot' image from Brain AI. Caption under 60 words, friendly tone, end with 'Book your table via the link in bio', and add three local hashtags."
For a batch: "Plan 8 LinkedIn posts for December: 2 per week, Tuesdays and Thursdays at 9:00 Vilnius time. Mix tips for small business owners, one customer story and one behind-the-scenes post. Use our brand voice from Brain AI. Show me the plan before writing the posts."
Tip: always give the posting time with your time zone, and say the language if you post in more than one.
Using your own images from Brain AI
AI employees find images in Brain AI by their title, not by what's in the picture. To use a specific image, type its exact title in quotes in your request. The title must match exactly as it appears in Brain AI: no typos, extra spaces or missing words.
For example: "Create a post with the 'Summer Sale Banner' image from Brain AI."
To use a whole folder: "Use only photos from my 'Shop interior' folder in Brain AI for this week's posts. Don't generate new images."
Tip: give your images descriptive titles when you upload them, like "Summer menu – iced latte – front view". A title like "IMG_4471.jpg" tells the AI employee nothing. Make sure the image is saved in All Knowledge in the workspace you're working in.
Creating and editing images
Describe what should be in the image, the style, the format and where it will be used. For a series, describe the shared style in every request, because each new image doesn't remember the previous one.
For example: "Create a square image for an Instagram post: a flat-lay of our three candles on a light oak table, soft morning light, plenty of space at the top for a headline. Use our brand colours #2F4B3A and #F2E8DC."
To edit an image, attach it and say what to change and what must stay the same: "Using the attached photo, change the background to plain white. Keep the product exactly as it is: same shape, label and colours."
To combine two images: "Using the provided images, place the mug from image 1 on the café table in image 2. Keep the table and the room in image 2 completely unchanged. The mug should look naturally lit, like it was there when the photo was taken."
Tip: text inside generated images can come out misspelled. For headlines or prices, ask for the image without text and add the text yourself afterwards, or check every word before you use it.
Adding your logo
Your logo comes from your Brand Kit. Attach the logo file in the chat as well, and describe where it should go in plain words rather than in pixels.
For example: "Add my logo from the Brand Kit to this image, small, in the bottom-right corner, on a clear area so it's easy to read. Keep the rest of the image unchanged."
If you need your exact logo, unchanged, ask for it in the chat. The AI employee then places your original logo file on the image instead of drawing it in.
For example: "Use my exact logo file from the Brand Kit, unchanged, small in the bottom-right corner of this image."
Product photos from phone snaps
Upload a few photos of the product from different angles. Lighting and background don't matter. Then say what you need.
For example: "Using these 3 photos of our ceramic vase, create clean e-commerce product images on a plain white background: one front view, one side view, one close-up of the glaze. Keep the shape, colour and pattern exactly like the real vase."
AI videos
Describe how you picture each scene, what text should appear, where your logo goes and how the video should end. Give a reference photo of the person if someone should appear in it. Say the format (9:16 for Reels, TikTok and Shorts; 16:9 for widescreen; 1:1 for square) and the length (each AI clip is 5 to 15 seconds). Before anything is charged, you'll see a review card with the plan and the credit cost.
For example: "Create a 10-second vertical ad for our 'Autumn Fig' candle. Scene 1: hands lighting the candle in a cosy living room. Scene 2: close-up of the flame with the text 'Hand-poured in small batches'. End: fade to black, then our logo from the Brand Kit. Use the product photos in my 'Candles' folder in Brain AI."
Tip: check the spelling of names and brand words in your prompt and in Brain AI before you generate. Keep the brief under about 2,000 characters and focused on scenes, text, logo and ending. Text and web addresses can come out wrong in AI videos, so check them before you post, or add them afterwards as an edit.
Pro tip: for the best results, create 5-second clips first, check each one, then ask for them to be joined into one longer video.
Editing your own videos
Send your video (up to 30 MB) in the chat and list the edits in the order you want them.
For example: "Trim the first 3 seconds, reframe this video to 9:16 for Instagram Reels with blurred bars instead of cropping, add burned-in subtitles timed to the speech, and put my logo small in the top-right corner."
Websites
Commet builds and edits your Sintra AI website. Name the exact page and section you want changed, and ask for one change, or one page, at a time.
For example: "On the homepage, replace the hero headline with 'Fresh sourdough, baked every morning in Kaunas' and change the button text to 'See today's bread'. Don't change anything else."
For a new page: "Add a 'Workshops' page with three workshop cards (title, date, price, short description) using the details in the 'Workshops 2026' document in Brain AI, and add it to the main menu."
Tip: for big changes, build in stages and ask for a credit estimate first.
If you see a "Couldn't save file" message, reply: "Break this request into smaller files and complete them one by one, step by step, instead of applying all changes in one update." If it keeps happening, start a new chat and add that line to the end of your request.
Emails, replies and outreach
Say who the email is for, what's already happened, what you want the reader to do, and the tone and length. With Gmail or Outlook connected, AI employees can draft and send emails, always with your approval.
For example: "Write a friendly follow-up to a customer who asked for a quote last week and hasn't replied. Mention the quote is valid until the end of the month, keep it under 120 words, and end with a clear yes/no question."
For customer replies: "Draft replies to today's unread emails that ask about opening hours or prices. Use our 'Prices and opening hours' document from Brain AI, reply in the customer's language, and put the drafts in my drafts folder. Don't send anything."
For a newsletter: "Draft our monthly newsletter for existing customers: one product update, one customer tip and one upcoming event. Use headings, keep it under 300 words, and end with a call to action to book online."
Data analysis and reports
Attach the file (or point to the Brain AI database), say what you want to learn, and how you want the answer shown.
For example: "Analyse the attached sales spreadsheet for January to October. Show the 3 best and 3 worst months, the top 5 products by revenue, and one chart of monthly sales. Summarise the findings in 5 bullet points, then suggest 2 things we should change."
Tip: before you send the file, check that it has clear column headers and one row per record. See "How to prepare data for Dexter analysis" in Related articles.
Documents, presentations and research
Say what the document is for, who will read it, its structure, its length and the file type you need.
For example: "Create a one-page PDF price list for our cleaning services using the 'Services and prices' document in Brain AI. Group the services into Home and Office, add our logo at the top, and keep it easy to read on a phone."
For research: "Research the 5 biggest competitors for dog grooming in Riga. For each, list their prices, services, Google rating and what they do better than us. Put it in a table and add 3 ideas we could use. Include links to your sources."
For a presentation: "Using the attached template and the notes below, create an 8-slide pitch deck for a bank loan meeting. One idea per slide, short bullet points, and a final slide with our contact details."
Business strategy and planning
Give your goal, your numbers and your limits. A plan built on your real situation is far more useful than a general one.
For example: "I run a small yoga studio with 120 members and want 40 more by March. My marketing budget is 300 EUR a month and I have 3 hours a week for it. Suggest a 3-month plan with weekly actions, and tell me which 2 actions will have the biggest impact."
Setting up and filling Brain AI
Brain AI is what every AI employee reads before doing anything, so it's worth asking an AI employee to help you fill it.
For example: "Look at everything in my Brain AI and tell me what you know about my business, offer and customers. Then list the 5 most important things that are missing and ask me for them one at a time."
To build your basics: "Interview me one question at a time about what I sell, who I sell to and what sets me apart. Then create and organise the documents in Brain AI."
Tip: asking an AI employee to "save this to Brain AI" saves a summary, not the exact text. For policies, scripts and legal text, upload the file or use the save button next to the message, so the exact wording is kept.
To point to a saved document, give its exact title and a phrase or section name from inside it. For example: "Use the refund rules from the 'Customer policies' document in Brain AI, in the section 'Late cancellations'."
Automations and scheduled tasks
Run the task once in chat and get it right first. Then ask for it to repeat, with a clear schedule in your time zone and a time window (for example "from the last 24 hours") instead of a fixed number of items.
For example: "Every Monday at 8:00 Vilnius time, summarise last week's sales from the 'Sales 2026' Google Sheet, compare them with the week before, and email me the 3 main changes in bullet points."
For an inbox routine: "Every weekday at 9:00, check emails from the last 24 hours, draft replies to customer questions using our FAQ in Brain AI, and leave them as drafts for me to review."
Tip: connect the apps the task needs before you set it up. An automation that needs an app you haven't connected will fail.
When results go wrong: made-up answers and how to handle them
Sometimes an AI employee gives an answer that sounds confident but is wrong or made up. This is called a hallucination. It isn't a sign that something is broken. AI employees predict the most likely answer, and when facts are missing they can fill the gap with something that only sounds right.
When you're most likely to see it:
Missing facts. You ask for an annual plan or a product page, and the details aren't in your prompt or Brain AI, so the AI employee invents them.
Vague requests. "Make it better" leaves the AI employee to guess what "better" means.
Questions about what it can do. An AI employee may say it can do something Sintra AI doesn't support, or say a task is done when it isn't.
Images. A request is misread, details change between versions, or text in the image is misspelled.
What to do:
Give the facts. Put prices, names, dates and figures in your prompt or in Brain AI, and ask the AI employee to use only those.
Let it say "I don't know". Add: "If you don't have the information, say so and ask me instead of guessing."
Ask for sources on research. "Include a link for every fact."
Check before you use it. Review figures, claims, names and image text before anything goes out in your name.
Check what was actually done. If an AI employee says it sent, saved or scheduled something, check it in the app, inbox or Brain AI. Anything that sends, publishes, schedules or spends shows an approval card first.
Rephrase in a new chat if a chat keeps going in the wrong direction.
Not sure whether Sintra AI can do something? Check the help center or ask support before relying on it.
Good to know
Some results are limits, not prompting problems. For example, each new image doesn't remember the last one, text inside images can be misspelled, and AI video clips are 5 to 15 seconds each. If a result keeps missing in the same way after you've given clear instructions, contact support.
Feedback only lasts if you save it. Approving or rejecting results doesn't teach an AI employee your taste on its own. Tell it what you like and why, then say "Save this as a rule".
Files attached in one chat aren't saved to Brain AI. Upload files you'll need again to Brain AI, so every AI employee can use them in future chats.
Credits
Every message uses credits, and some tasks use more than others. A clear prompt usually gets a good result in fewer rounds, which keeps your credit use down.
Clear briefs cost less. An image uses about 0.5 to 1.8 credits depending on quality. A vague brief with many rounds of feedback costs more.
Long chats cost more per message. Start a new chat for each new task.
Video is the heaviest task. It uses about 3 credits per second of video, and you see the cost on the review card before you click Generate.
Ask before big jobs. For a website build or a large batch, ask: "Give me a credit estimate before you start."
See Workspace credits for more.
Tips
Fill your Brand Kit and Brain AI first. Generic output almost always means Brain AI is missing information.
Refer to Brain AI items by their exact title, in quotes.
Train one AI employee well before you add others.
Use voice input if typing long prompts feels slow. Talking it through works just as well.
Use real numbers: word counts, dates, quantities, prices.
Keep one prompt per task, and one topic per chat.
When something works well, save the prompt or ask the AI employee to turn it into a skill or an automation so you don't have to write it again.
FAQ
Why does the AI employee say it understood my instructions and then not follow them?
Why does the AI employee say it understood my instructions and then not follow them?
Usually because the request was long, held several tasks at once, or came after many rounds of corrections in the same chat. Split the request into one task per message, and if the chat is long, start a new one with a single clear prompt. Describe what you want instead of only what you don't want.
Do I have to repeat myself in every chat?
Do I have to repeat myself in every chat?
No. Anything you want every time belongs in Brain AI or in a rule. Save your tone, brand details and key documents in Brain AI, turn repeated corrections into rules by saying "Save this as a rule", and save tasks you repeat as skills. Each new chat doesn't see your earlier chats, but it does see Brain AI.
When will my AI employee learn what I like?
When will my AI employee learn what I like?
Once you tell it and save it. Approving or rejecting results doesn't train it by itself. Say what you liked or didn't, and why, and then "Save this as a rule". Good examples saved in Brain AI help a lot too.
How do I get it to use my own photos instead of AI images?
How do I get it to use my own photos instead of AI images?
Upload your photos to Brain AI with descriptive titles, then name the exact title in quotes in your prompt, for example "Create a post with the 'Summer Sale Banner' image from Brain AI." To use a group of photos, put them in a folder and ask: "Use only photos from my 'Shop interior' folder."
Will it find an image by its file name?
Will it find an image by its file name?
It finds images by the title stored in Brain AI. Give each image a clear, descriptive title and use that exact title in your prompt.
How do I get it to use my real logo?
How do I get it to use my real logo?
Add your logo to the Brand Kit, attach it in the chat as well, and say where it should go in plain words. If you need it exactly as it is, ask in the chat for your exact logo file, unchanged, and the AI employee places the original file on the image.
How long should my prompt be?
How long should my prompt be?
As long as it needs to be to cover the task, the context, the format and an example, and no longer. Most everyday prompts are a few sentences. Long-standing information like your brand voice belongs in Brain AI, not in every prompt.
I asked for four things and only got two. Why?
I asked for four things and only got two. Why?
Several tasks in one message compete for attention. Ask for one thing at a time, or ask the AI employee to make a plan first and then work through it step by step.
Should I keep correcting in the same chat or start over?
Should I keep correcting in the same chat or start over?
For one or two rounds of feedback, stay in the chat and say exactly what to change. If the chat has gone through many corrections and results are getting worse, start a new chat with one clear prompt and only the latest version of your file.
Why does it make things up?
Why does it make things up?
AI employees predict the most likely answer. When facts are missing, they can fill the gap with something that sounds right but isn't. Give the facts, ask it to say when it doesn't know, and check important details before you use them. See "When results go wrong" above.
Does being polite help?
Does being polite help?
Being polite doesn't hurt, but it doesn't improve results either. Clear, specific instructions do. Start with your request rather than a greeting.
Do I need to tell it to act as an expert?
Do I need to tell it to act as an expert?
No. Each AI employee already has its own role. Choose the AI employee that fits the task, or use Team chat. You can still say who the content is for and what tone to use.
How do I avoid wasting credits?
How do I avoid wasting credits?
Write one clear prompt instead of many quick ones, keep one topic per chat, and ask for a credit estimate before big jobs like websites or videos. See the Credits section above.
I'm still not getting good results. What else can I do?
I'm still not getting good results. What else can I do?
Ask the AI employee: "What information do you need from me to do this well?" If a certain type of task keeps failing even with clear prompts, contact support through Help in the app or at [email protected].
