Most SEO prompts you find online are too generic. They produce text that sounds plausible and changes nothing. Here are seven prompts we use ourselves, along with why they work.
And the honest part, up front: At the end, you'll see where ChatGPT consistently hits a wall on SEO. Not because the model is too weak, but because it can't see your website.
First: give context, not just the task. “Write an SEO article about dental implants” gets you average, because average is what the model learned. Tell it who the page is for, what you're offering and which objection you're up against, and you get something you can use.
Second: insist on a source. A model will make up search volume if you ask for it. It will make up rankings, too. Ask for structure, wording and arguments, not numbers.
Third: let it push back. The most valuable addition to any prompt is asking the model to criticize its own suggestion. Models love to agree with you, and you have to train that out of them.
It doesn't know current rankings, real search volume or the details of your competitors. Anything that sounds like it does is a guess. That's not a knock on the tool, it's simply how it works: it deals in language, not in a database.
Every prompt is built so you only have to replace the brackets. They work equally well in ChatGPT, Claude and Gemini.
I want to build a page for the search term "[KEYWORD]". First, analyze the search intent: 1. What does someone typing this really want? To learn, compare, buy, or solve a problem? 2. What situation is this person in? What just happened? 3. Which three questions does the page have to answer so they stop searching? 4. Which page type fits: guide, comparison, product page, calculator, checklist? Finally, tell me honestly whether a page for this term makes sense at all, or whether the intent belongs to a different term.
The most common SEO mistake isn't the copy, it's the wrong page type. Build a product page for a comparison search and it will never rank, no matter how good the copy is.
Topic: [TOPIC]. Audience: [AUDIENCE, e.g. owners of mid-sized businesses with 20 to 100 employees]. Create an outline with H2s and H3s. Rules: - Every heading is a question this audience actually asks, or a statement they actually care about. No filler sections like "Conclusion" or "Introduction". - Order the sections the way someone makes a decision, not by the logic of the topic. - Mark the three sections that carry the buying decision. - Tell me which section is usually missing from existing articles on this topic.
The last line is the trick. It forces the model past the usual, instead of handing you the twelfth version of the same outline.
Write 8 versions of the title tag and meta description. Page: [WHAT IT'S ABOUT] Main keyword: [KEYWORD] What sets us apart: [YOUR DIFFERENCE] Rules: - Title max. 60 characters, description max. 155. - Put the main keyword as early in the title as possible. - No superlatives, no exclamation points, no "now". - Each version must highlight a different benefit. - After each version, add in parentheses which person clicks on it and which person doesn't.
The parentheses at the end turn a list into a basis for a decision. Without them, you just pick the first version, because they all sound equally good.
Here is the copy of an existing page: [PASTE TEXT] Review it in this order: 1. Which statement is an empty phrase with no substance? Quote it word for word. 2. Where do we claim something without backing it up? 3. What question does a reader have here that never gets answered? 4. Which paragraph could go without losing anything? Then rewrite only the three weakest paragraphs, not the whole text. Keep the tone of voice and the technical terms.
A complete rewrite is almost always the worse choice: it loses details that come from real customer experience. Replacing three paragraphs on purpose is more effective and faster.
Here is the list of my pages with title and topic: [PASTE LIST, one line per page: URL, title, topic in 5 words] Tasks: 1. Which page is the topical center and should get links from many other pages? 2. Which pages overlap in content and compete with each other for the same term? 3. Suggest 10 specific internal links, each as: from which page, to which page, with which anchor text. 4. Which page currently has no meaningful internal link pointing to it?
Points 2 and 4 catch two problems that are hard to spot without a tool: two of your own pages taking rankings away from each other, and pages nobody links to.
Offer: [WHAT YOU SELL], price: [PRICE]. Audience: [AUDIENCE]. Write 8 FAQ questions. Rules: - Only questions people really ask before buying, including the uncomfortable ones: too expensive, too complicated, what's the catch. - Phrase them in the audience's language, not ours. - Answers are no more than three sentences and dodge no question. - For at least one question, the honest answer has to be that our offer isn't the right fit for it.
The last rule is the most important one. Nobody believes an FAQ where everything fits. One that admits a limit makes the other seven answers believable.
You just recommended [SUGGESTION] to me. Switch roles: you're a skeptical SEO consultant who thinks this suggestion is a bad idea. Give me: 1. The three strongest arguments against it. 2. What happens if we implement it and it doesn't work. What will we have lost? 3. Which assumption in the suggestion has no evidence behind it? 4. What would be the simpler solution that gets 80 percent of the impact? Don't soften your answer. I want to see the weak spots.
This prompt alone is worth the whole article. Almost every first AI suggestion fails question 4, and the simpler solution is usually the right one.
These seven prompts turn a chat window into a useful tool. Even so, three things are left undone, and they have nothing to do with the quality of the model.
You paste in some copy, but not the load time, not the internal links, not the 40 other pages, not what visitors actually click on. The diagnosis is still on you, and that's the hard part.
In the end, you have better copy sitting in a chat window. Getting it into the CMS, the meta tags, the internal links: that's your job. For one page, that's ten minutes of manual work. For twenty pages, nobody does it anymore.
Every new chat starts from zero. Your tone of voice, your arguments, last week's decision: you paste it all in again. That repetition eats exactly the time the AI was supposed to save.
The same language models you use in ChatGPT, only with access to your website and the job of actually implementing the changes. The app runs on your Mac and works with your own AI account, or with a model that runs locally. You set up your tone of voice once, and it applies to every task.
And when ChatGPT is still the right choice: For a single question, for brainstorming, for rewriting a paragraph. It's already open, it costs nothing extra, and for a single piece of copy it's unbeatably fast. The difference only kicks in when the same work comes back every week.
If you'd rather hand off these three gaps than bridge them yourself every week, here's the answer: nebra AI for SEO, with eleven ready-made workflows and an honest list of what the app isn't good for.
These prompts come from the daily work at konversion.digital and were developed on client projects. They work with any major language model, no particular subscription required.
For research, outlines, titles and copywriting, yes, and it does them well. Not for ranking data, search volume or competitor analysis: there, a language model makes up numbers that look plausible. And it doesn't implement anything. Getting from the chat window into your CMS is still manual work.
For SEO copywriting, the gap between the big models is smaller than the gap between a bad prompt and a good one. All seven prompts on this page work with ChatGPT, Claude and Gemini.
Google judges the result, not how it was made. Thin content that says nothing new fails, no matter who wrote it. Content with real experience, evidence and a clear answer to the search intent works, even if a model helped with the wording.
Not necessarily for these prompts, they also run in the free version. If you use it every day, you'll notice the difference with longer texts and with how much it remembers within a conversation.
It uses the same models, but it has access to your website, knows the tone of voice you've set up, and publishes changes straight to your CMS once you approve them. The difference isn't the quality of the writing. It's that the work actually gets done.
Instead of copying seven prompts one after another: give it a task, walk away, review the result. €99 per month plus VAT, cancel monthly.
Written by
Founder and managing director of konversion.digital, a behavioral economist with 20 years of hands-on conversion work for mid-sized businesses and more than 800 A/B tests run. Author of the German book “Zeiträuber & Eroberer” and a keynote speaker on conversion, AI and the psychology of decision-making. Nebra AI packages his methodology so teams can apply it without an agency.