A blog post prompt ranks when it encodes the same things a human SEO brief does: the target query, the search intent behind it, the heading structure, the entities the page must mention, and where internal links go.
If your prompt only says "write a blog post about X," you'll get fluent text that answers nothing specific, and Google has no reason to rank it above the ten pages that already answer the query directly.
The mechanism is simple. Search engines match a page to a query by looking at whether the page's headings, opening lines, and body text address the same intent the searcher had. A prompt that names the query and the intent forces the model to build the page around that intent instead of around whatever it finds most statistically common. That's the difference between a post that reads like a Wikipedia summary and one that reads like it was written for the person typing the query.
Here's a full example prompt you can adapt. Notice every line has a job.
"You are an SEO content writer for a small business blog. Write a 1,200-word post targeting the query 'how to clean a cast iron skillet.' The searcher's intent is instructional: they have a dirty pan and want a safe method now.
Open with a two-sentence direct answer, then use these H2 headings: why seasoning matters, the salt-and-oil method, the water-and-scrub method, what never to do, and how to re-season after a mistake. Mention these entities by name: flaxseed oil, polymerized oil, rust, and chainmail scrubber.
Place the exact phrase 'clean a cast iron skillet' in the first 100 words, in one H2, and once in the conclusion. Add one internal link to a related post about seasoning. Write at an 8th-grade reading level. Do not invent studies, temperatures, or product prices."
Break that down. The role line sets tone. The query line anchors the topic. The intent line tells the model what the reader wants, which controls the opening. The H2 list controls structure, so the model can't wander. The entity list forces the specific vocabulary that search engines associate with the topic. The keyword-placement line prevents the model from burying the phrase in paragraph nine. The internal-link line keeps your site architecture in the output. The reading-level and no-invention lines are your quality gates.
A decision rule for building your own prompt: if a line doesn't change what the model writes, cut it. "Make it engaging" changes nothing. "Open with the direct answer in two sentences" changes the first paragraph. "Use these five H2s" changes the whole outline. Every instruction should map to a visible difference in the draft.
For internal links specifically, don't say "add relevant links." Give the model the anchor text and the URL, or at minimum the topic of the page you want linked. According to our AI tool database, which tracks 360 AI tools with pricing and capability snapshots verified as recently as 2026-09-18, most writing tools will happily insert a link if you name the target page, but none will guess your site structure correctly on their own. You have to supply it.
Where this breaks down: a good prompt cannot fix a bad keyword choice. If you target a query that a thousand established sites already answer with more authority than you have, structure won't save you. Prompts also can't verify facts.
The model will still invent a temperature or a study if you don't forbid it, and even when you do, it sometimes slips. Treat the output as a first draft that needs a fact-check pass, not a finished page. And if your target query has no clear intent — some brand-name searches don't — the intent line has nothing useful to say, so skip it rather than forcing it.
One more tip: keep your best prompts in a file. The prompt that produced a ranking post is an asset. Rewriting it from scratch each time is how people end up with inconsistent output and no idea which instruction actually mattered.