How you prompt matters more than which tool you prompt. Whether you are drafting a post in ChatGPT or Claude or generating an image in Adobe Firefly, these five AI prompting strategies separate useful first drafts from generic ones. They line up with what the model makers themselves publish: OpenAI’s original 2023 prompting guide led with “write clear instructions” and “provide reference text” (AI Business summary), and Anthropic’s current guidance starts with “be clear and direct” (Claude docs).
1. Write instructions a new hire could follow
Anthropic offers a useful test: show your prompt to a colleague with minimal context on the task. If they would be confused, the model will be too.
Say who the piece is for, what it should accomplish, the format, the length, and what to avoid. Compare:
- Vague: “Write a blog post about rebranding.”
- Clear: “Write a 900-word post for owners of B2B service firms with 10 to 50 employees who are wondering whether to rebrand. Explain three signs a rebrand is warranted and two signs it is not. Short paragraphs, no clichéd opening, no ‘In conclusion’ header.”
The second prompt takes a minute longer to write and saves several rounds of revision.
2. Show examples of what good looks like
Anthropic calls examples “one of the most reliable ways to steer Claude’s output format, tone, and structure,” and says a few well-crafted ones improve accuracy and consistency (Claude docs). Paste two or three paragraphs you have written and liked, and say what to take from them: the sentence length, the directness, the way they open. For images, reference the composition, lighting and palette you want rather than adjectives like “stunning.”
3. Give it the facts to work from
A model fills gaps with plausible guesses, and invented statistics and citations come from those guesses. OpenAI’s current prompting guide has a section on including relevant context information in the prompt for exactly this reason. Paste your notes, the source article, your service details or the client’s actual numbers, and tell it to use only that material and to say when something is missing.
Then check every number, quote and link yourself before anything is published.
4. Break big jobs into smaller steps
OpenAI’s 2023 guide recommended splitting complex tasks into simpler subtasks and giving the model time to work through a problem before answering (AI Business summary). For content, that means asking for an outline first, correcting it, then drafting one section at a time. Each step is easier to check, and a mistake in the outline gets caught before it becomes a mistake in 1,500 words.
5. Save what works as standing instructions
Rating an answer with a thumbs-up or thumbs-down does not change how the model answers your next message. Anthropic, for example, says feedback may be used to train future models but is not combined with your other conversations (Anthropic Privacy Center). What changes the next answer is what you put in the conversation, or in instructions the tool keeps for you: custom instructions in ChatGPT, project instructions and skills in Claude, style references in image tools.
Your voice belongs there. We keep the rules every draft must follow, such as the words to avoid, the sign-off, the punctuation rules and how sources are cited, as saved instructions, so every session starts from them instead of from scratch. When a draft misses, say what was wrong and what you want instead, and if it is a standing preference, add it to the instructions.
One more habit: try a second model
Models differ in tone, length and how closely they follow instructions. Run the same prompt on two of them now and then, and keep the one that needs less correction for that kind of work.
Writing for search as well? How to write website text for SEO covers structure, and LLM visibility covers what gets a brand named in AI answers.
Common questions
Questions, answered.
What are the most effective AI prompting strategies?
Write clear, specific instructions with context; show examples of what you want; give the model the facts to work from; break large jobs into smaller steps; and save the instructions that work so every session starts from them.
Does rating an AI answer with a thumbs-up or thumbs-down change the next answer?
Not in the moment. What changes the next answer is what you put in the conversation or in saved instructions. Say what was wrong and what you want instead, and put standing preferences in the tool’s custom or project instructions.
How do I stop AI from inventing facts?
Give it the source material and tell it to use only that material, then check every number and quote yourself before publishing. A model can produce a confident sentence and a plausible citation for something that is not true.
Should I use more than one AI model?
It is worth testing. Models differ in tone, length and how closely they follow instructions, so run the same prompt on two and keep the one that needs less correction for that kind of work.