AI for writers, marketers and creators
Draft, edit, keep your voice, and stay on the right side of originality.
Learning goal: You produce volume without sounding like everyone else's chatbot.
1 / 9
A plain guide to using AI for volume and speed without losing your voice or your credibility.
Before you start (prerequisites)
- A chatbot or writing tool (ChatGPT, Claude, Gemini, Copilot, or a dedicated marketing tool like Jasper).
- A few samples of your own past writing, ready to paste in - these teach the tool your voice faster than any instruction.
- A clear sense of your platform's or client's disclosure rules (many now require you to say if content is AI-assisted).
- Basic awareness of copyright and image rights - you don't need to be a lawyer, but you need to know it's a live issue.
- No technical skill required, just a willingness to edit, not just generate.
Do this first:
- Paste three of your own past posts or emails into a chatbot and ask it to describe your writing style in a short paragraph.
- Ask it to draft one short piece using that style description, then compare it to how you'd actually have written it.
- Check your platform's or employer's current policy on AI-generated content and disclosure.
Where things stand (as of 2026)
AI is strong at first drafts, restructuring, headline variations, and turning notes into prose. It's weak at genuinely original ideas, accurate niche facts, and consistently matching a specific person's voice without heavy guidance. Image generation is powerful but rights and licensing remain murky and provider terms differ, so always check the specific tool's commercial-use terms before publishing generated images at scale. Detection of "AI-written" text is unreliable in both directions, so treat that whole area with caution - for creators and for anyone judging them. This space shifts fast; assume today's platform policy on disclosure may change within the year.
The mental model
animated · a prompt, layer by layer
Six sentences turn a useless answer into a usable one.
the prompt, growing
the bare ask
Write about our new coffee subscription.
how usable is the output
18%
You'll get a generic blog post that could belong to any company on earth.
Good output here isn't magic phrasing - it's giving the model the ingredients only you have: your voice, your facts, your audience, your goal. Treat the AI as a fast intern who's read everything but knows nothing about your actual brand or your actual reader.
Scenarios that work
Turning rough notes into a blog draft
medium stakeswhat you want
Get a first draft fast from bullet-point notes
what most people type
write a blog post about our new product feature
the briefed version
Turn these bullet notes into a blog post draft for our small business audience (non-technical, time-poor). Notes: [paste notes]. Tone: friendly, practical, no hype words like 'revolutionary'. 400-500 words, end with one clear call to action.
Stake: Publishing an invented product claim can mislead customers and damage trust.
Matching your own voice
low stakeswhat you want
Draft a newsletter that sounds like you, not like generic AI
what most people type
write a newsletter in my voice
the briefed version
Here are three past newsletters I wrote: [paste]. Study the sentence length, humour, and how I open and close. Then draft a new one on this topic: [topic], keeping that same voice. Flag anywhere you deviated from my style.
Stake: Worst case is a slightly flat draft, easily fixed in editing.
Researching a topic before writing
high stakeswhat you want
Get background on an industry trend for a client article
what most people type
tell me about trends in sustainable packaging
the briefed version
Give me 5 current discussion points in sustainable packaging as of [year], each with one sentence of context. Mark clearly which ones you're confident are current fact versus general trend commentary. I will verify each before using it.
Stake: Publishing a false trend or statistic under a client's name damages their reputation and yours.
Generating an image for a campaign
medium stakeswhat you want
Create a hero image for a social post
what most people type
make an image of a coffee shop
the briefed version
Generate a warm, welcoming image of an independent coffee shop interior, morning light, no visible text or logos, suitable for a small business social post. I need to know the licence terms for commercial use of this output before I publish it.
Stake: Using an image that infringes rights, or breaches platform ad policy on AI imagery, can get a campaign pulled or challenged.
Producing volume without sounding robotic
low stakeswhat you want
Write 10 social captions for the week without them all sounding the same
what most people type
write 10 social media captions for my bakery
the briefed version
Write 10 short Instagram captions for my bakery this week, each tied to a different real detail I give you: [list 10 real details - new loaf, weather, a customer story, etc]. Vary sentence structure and length across them. No generic phrases like 'indulge yourself' or excessive exclamation marks.
Stake: Worst case is bland copy, not a disaster, but it does erode audience trust over time.
Where it fails, and what it costs you
The two big traps are invention and homogenisation. Invention means the AI states a fact, statistic, or customer quote that doesn't exist - publish that under your byline and it's your name on the error, not the AI's. Homogenisation means everyone's marketing copy starts sounding the same flavour of upbeat-and-vague, which is a slower but real cost: audiences tune it out.
Plagiarism and originality are also live risks. AI models are trained on huge amounts of existing text and image data, and outputs can sometimes closely echo a specific source without you knowing. For text, run anything you're unsure about through a plagiarism checker before publishing under your name. For images, be aware some outputs can resemble a living artist's style closely enough to cause a legitimate complaint.
Disclosure matters more than people think. Many platforms, publications, and clients now have explicit rules about declaring AI-assisted content. Not disclosing when required isn't just a house-style issue, it can breach a contract or a platform's terms of service.
How to check the answer
- Fact-check every specific claim, statistic, or quote against an independent, dated source before it goes anywhere public.
- Read the whole piece aloud - flat, generic-sounding passages are the clearest sign of unedited AI output.
- Run text through a plagiarism/similarity checker if you have any doubt about originality.
- Check your platform's, client's, or employer's current AI disclosure policy before publishing.
- Get a second human read, especially for anything client-facing or under your byline.
How to read the docs and find the truth
For text tools, check the provider's own usage policy and terms of service page (openai.com/policies, anthropic.com/legal, etc.) for what you're actually allowed to do commercially with the output. For image tools, the licence terms are usually a specific page separate from the general terms - search "[tool name] commercial use rights" on the provider's own site, not a forum post. For disclosure norms, check your specific platform's current policy (many update these yearly) rather than assuming last year's rule still applies. For copyright questions specific to your country, your national copyright office's website is the authoritative source, not a general blog post about "AI and copyright."
Your one-page playbook
PIECE: [what you're creating - blog, caption, email, image] AUDIENCE: [who reads this and what they care about] VOICE SAMPLES: [paste 2-3 examples of your own past writing] REAL FACTS/DETAILS: [only things you know to be true - no gaps for it to fill with invention] BANNED PHRASES/CLICHES: [your personal no-go list] DISCLOSURE NEEDED: [yes/no, per platform policy] VERIFY BEFORE PUBLISHING: [fact-check, plagiarism check, human read]
Check yourself
- Why is pasting your own past writing more effective than describing your style in words?
- Name one risk each for text output and image output that a marketer should specifically check for.
- What's a quick way to tell if AI copy has gone unedited into final publication?
- Why can't you rely on "it sounds unique" as proof that AI-generated text is original?