AI writing tips: draft faster with Llama and still sound like you
Keep the decisions for yourself and give Llama the mechanical parts: structuring, trimming, varying and checking. This workflow shows how, step by step.

AI can get you from a blank page to a workable draft in seconds. Left to itself, it also produces the flat, over-polished prose that readers now recognise at a glance. The trick is to keep the decisions for yourself (what to say, for whom, in what voice) and hand Llama the parts that are mechanical: structuring, trimming, varying, checking. Here is a workflow that does exactly that.
Step 1: write a brief before you write a prompt
Two minutes of thinking saves ten rounds of rewrites. Answer these in a few words each:
- Purpose: what should the reader do or feel after reading?
- Reader: who are they, what do they already know, what do they care about?
- Key points: the three to five things that must be in it.
- Proof: the facts, examples or stories you will use. Only you have these.
- Length and place: a 120-word email, an 800-word blog post, a headline that fits on one line.
Paste the brief into the chat. From now on, everything Llama writes will be about your content rather than generic filler.
Step 2: outline together, then draft
Weak: Write a blog post about saving energy at home.
Strong: Here is my brief: [paste]. Suggest two different outlines for this post, each with 4-6 section headings and one line on what each section says. Don't write the post yet.
Choosing between outlines is where you shape the piece. Once you have picked one, ask for one section at a time, or write the draft yourself and use Llama only to edit. Working section by section keeps each part focused and lets you change course early.
Step 3: give Llama your voice
Describing your style rarely works; showing it does. Paste two or three paragraphs you wrote and liked, and add:
Study the sample below: sentence length, word choice, how often I use questions and humour. Write the next section in the same voice. Avoid words I never use, such as "moreover" and "in conclusion". Sample: [paste]
Then read the result aloud. Anything you would not say yourself, cut or change. You can try the voice prompt in the chat with a sample of your own.
A worked example: from rough notes to a paragraph
Here is how the first three steps look on a real task, a short note for a neighbourhood newsletter.
Rough notes: community garden - open Saturdays 9-12 - need volunteers - tools provided - kids welcome - we grew 40 kg tomatoes last year?? check - sign up at the board by the gate
Prompt: Turn my notes into one friendly paragraph for a neighbourhood newsletter, about 90 words. Readers: busy families who walk past the garden but have never come in. Goal: get them to sign up. Keep every fact from the notes, leave out anything marked "check", and don't add facts of your own. End with how to sign up.
Notice what the prompt does. It names the reader and the goal, sets a length, protects the facts and tells Llama to drop the one detail you were not sure about. The first draft will still need your touch, perhaps a name or a local joke, but it starts from your content instead of a generic "join us" text. If the draft sounds stiff, reply: "Warmer and simpler, as if I were saying it at the school gate."
Step 4: edit in separate passes
A single prompt that asks to "improve" a text changes everything at once, and you lose track of what happened. Ask for one kind of edit per pass:
- Structure: "Does the order make sense? Suggest moves, but don't rewrite."
- Clarity: "Mark sentences that are hard to follow on first reading and suggest a plainer version of each."
- Length: "Cut about a quarter without losing any facts. Show me what you removed."
- Tone: "Make it warmer for a customer who is annoyed, and apologise only once."
- Correctness: run the final version through the grammar checker.
Asking Llama to show what it changed keeps you in control, and over time it teaches you your own habits.
Remove the "AI sound"
Machine-drafted text has recognisable tics. Look for these and fix them:
- Openers that announce instead of saying anything: "In this article, we will explore…"
- Pairs and triplets of adjectives everywhere ("clear and concise", "fast, simple and reliable").
- Every paragraph ending in a neat little moral.
- Praise words with no evidence behind them.
- A conclusion that repeats the introduction.
A useful follow-up: "Rewrite this, replacing every general claim with a specific example from my notes. If there is no example, delete the claim." Specific detail is the fastest cure for generic text.
Check facts and originality
Llama can state wrong facts fluently, and in our chat it cannot look anything up. Every name, date, number and quote in a draft needs checking against a real source, especially anything after the model's knowledge cut-off. Do not ask it to invent statistics, reviews or quotes, and delete any it adds on its own. If you publish under your name, you are responsible for what the text says. Our guide on fact-checking AI answers has a routine for this.
For school or work, follow the rules about disclosing AI help; some publications and clients ask for it.
The right tool for each job
| Task | Where to do it |
|---|---|
| First drafts of posts, pages and descriptions | AI writer |
| Rewording a passage in a different tone | Paraphraser |
| Grammar, spelling and punctuation | Grammar checker |
| Structured essays | Essay writer |
| Emails that need the right tone | Email writer |
| Fiction and creative ideas | Story generator |
| Short posts for social networks | Social media post generator |
| Long back-and-forth editing | The chat |
For ready-made starting points, see the writing prompts and creative prompts, or read Llama for writers. If you are new to prompting, start with how to write prompts.
Which model for writing?
The default, Llama 4 Maverick, is a good place to start. For a second take on the same draft, try Llama 3.3 70B: two different models give you more varied options to choose from. For quick rewording of a sentence or a headline, Llama 3.1 8B is usually the quickest. Whichever model you use, the final read-through is yours.
FAQ
Will AI-written text hurt my search rankings?
Google's published guidance says it judges content by its quality and usefulness rather than by how it was made, and that generating pages mainly to game rankings breaks its spam policies. Pages with no real value are a problem whatever made them. Add your own experience, facts and examples.
Can Llama write in my language?
Yes, in many languages. Meta's model cards list 12 officially supported languages for Llama 4 and 8 for Llama 3.3 and 3.1; others often work too, with more mistakes. Write your prompt in the language you want back.
How do I stop it from making the text longer every time?
Give a number: "Keep it under 150 words" or "same length as the original". Models are not precise with word counts, but a number keeps them close.
Who owns the text Llama writes for me?
Our terms explain how you may use what you create here. Copyright rules for AI-assisted writing differ between countries and are still developing, so check local rules for important commercial work. This is general information, not legal advice.
Ask Llama your first question
No account and nothing to install. Open the chat, type a message and see what Llama can do.
Open Llama chat

