How to Write a Blog Post That Actually Ranks

Key takeaways

Answer the core question in the first two sentences — not after a definition, a history lesson, or a "why this matters" preamble
Write each section so its best sentence survives being quoted alone, out of context, by a person or a language model
Cut anything that only restates the heading; every paragraph needs a fact, number, step, or example the reader couldn't have predicted

Start with the question, not the topic

Most bad blog posts confuse a topic ("email marketing") with a query ("how often should I send a cold email follow-up"). A topic invites throat-clearing. A query demands an answer.

The tell is the first paragraph. If it opens with "Email marketing has become an essential tool for businesses of all sizes," the writer picked a topic and is stalling before getting to the point. If it opens with "Send three follow-ups spaced two, four, and seven days apart," the writer picked a query and answered it. AI answer engines pull from the second kind almost exclusively, because there's an actual claim to extract in the first sentence. Google's own guidance on helpful content pushes the same direction — it explicitly asks whether a page leaves the reader "feeling like they've learned enough about a topic to help achieve their goal," which is a polite way of saying: does the first screen actually answer the thing they typed? (Google Search Central)

Research the query before you touch the outline

Before writing a word, pull up the top 5-8 ranking pages and any AI Overview or Perplexity answer for the exact query. You're not looking for what to copy — you're looking for what's missing. If every ranking post lists "steps" but none give a time estimate, add one. If none define a term precisely, define it in one sentence and move on.

This is also where most indie hackers waste the most time doing it manually, tab-hopping between ten search results with no system for capturing what's actually different. Running that research through a structured process — even something as simple as feeding SERPs into an AI assistant with a consistent prompt — turns a 45-minute slog into 10 minutes. We've written a full walkthrough on using ChatGPT plugins for this kind of SERP research if you want the specific workflow instead of doing it by hand every time.

Structure for skimmers first, algorithms second

Eye-tracking research from the Nielsen Norman Group found that people read web content in an F-shaped pattern: they read the first couple of lines fully, then scan down the left margin picking up fragments of subsequent lines rather than reading straight through. (Nielsen Norman Group) That's not a reason to dumb down your writing — it's a reason to put the payoff of each paragraph in its first sentence, not its last.

Practically, that means:

One idea per paragraph, three to five sentences max
The strongest, most specific sentence goes first in each section, not as a "wrap-up"
Headings phrased as the actual question a reader has, not a vague label

If you're covering terminology-heavy topics, this is the exact reason glossary-style pages tend to perform well in AI search — the definition-first format is structurally identical to what an F-pattern reader and an AI extraction model both want: term, then answer, then supporting detail.

Write paragraphs that survive being quoted alone

This is the part most SEO advice from 2021 misses, because it predates AI answer engines pulling single paragraphs out of pages and presenting them with no surrounding context. When ChatGPT, Perplexity, or Google's AI Overview cites your post, it usually grabs one paragraph or one sentence — not the whole section. If that paragraph depends on the sentence before it ("As mentioned above, this approach works because..."), the citation reads like nonsense.

Write each paragraph as a self-contained unit: claim, then evidence, then (optionally) a caveat. Avoid pronouns that refer back three sentences — "it," "this," "that" should almost always resolve within the same sentence. A paragraph that starts with "This makes it faster" forces the model to guess what "this" is; a paragraph that starts with "Batching your outline and draft into one AI prompt cuts editing time" doesn't need any prior context to make sense.

We test this internally by literally deleting everything except one paragraph and asking whether it still reads as a complete, accurate answer. If it doesn't, the paragraph gets rewritten — not because a human reader would notice, but because a citation engine will lift exactly that paragraph and nothing else.

Edit out the specific things that break in automated content

If you're using AI to draft the post — and most solo founders now are — the editing pass matters more than the drafting prompt. The failure modes are predictable and specific:

Hedge-everything sentences. "This can vary depending on your situation" appears in a draft where the model doesn't have a strong enough example to commit to. Fix it by giving the model (or yourself) a real number or scenario instead of a general caveat.
Restating the question. A draft that opens a section with "When it comes to writing a blog post, structure is important" is spending 12 words to say nothing. Delete the setup, keep the claim.
Fake specificity. Watch for numbers that sound precise but aren't sourced or falsifiable — "73% of marketers agree" with no citation. If you can't point to where a number came from, cut it and describe the mechanism instead ("shorter subject lines get opened more often because they don't get truncated on mobile inboxes" is verifiable and useful without needing an invented percentage).
Section padding. If a section could be one sentence but is four paragraphs, the model (or the writer) ran out of real information and started restating the heading in different words.

None of this is unique to AI-assisted writing — human writers do all four just as often. The difference is that AI drafts do it at scale, consistently, in every post, which means an unedited AI content pipeline produces a whole site full of the same four problems rather than one bad post buried among good ones.

Treat metadata and internal links as part of the post, not an afterthought

The blog post isn't finished when the body copy is done. Two things determine whether it gets clicked and whether it gets cited alongside the click:

The meta description. AI answer engines and Google both sometimes pull the meta description verbatim as a summary snippet, so it needs to be an actual sentence answering the query, not a keyword-stuffed tagline. We cover the specific format that gets reused by AI engines in our guide to writing meta descriptions.
Structured internal links. A blog post that never links anywhere reads as an orphaned page to crawlers and to readers — there's nowhere to go deeper. Link out to two or three genuinely related pages using anchor text that describes what's on the other end, not "click here." If you're publishing regularly, it's worth deciding your linking pattern in advance rather than improvising it post by post — we go into the specifics of interlinking cornerstone content so your best pages accumulate authority instead of getting buried under newer posts.

Schema markup (FAQPage, Article, HowTo) is a smaller lever than either of these, but it's not nothing — it gives crawlers an unambiguous, machine-readable version of the same structure you already built for human readers. The vocabulary for all of it lives at Schema.org, maintained jointly by Google, Microsoft, Yahoo, and Yandex specifically so structured data works the same way across search engines.

The length question, answered directly

There's no minimum word count that makes Google or an AI engine trust a post more. A 600-word post that answers the query completely will outperform a 2,500-word post that pads three subtopics no one asked about. Length should be a byproduct of how many distinct sub-questions the topic genuinely has — not a target you write toward. If you find yourself adding a section just to hit a number, that section is the first thing to cut.

Frequently Asked Questions

Q: How long should a blog post be to rank well?

There's no fixed word count that helps rankings on its own. Length should match how many genuine sub-questions the topic has — a narrow query might need 600 words, a comparison guide might need 2,000, and padding either one to hit a number usually hurts readability more than it helps ranking.

Q: Should I write the title before or after the body?

Write a working title first to keep the post focused on one query, then revise it after drafting once you know which sentence in the post is the strongest, most specific answer — that sentence often makes a better title than the one you started with.

Q: Do AI answer engines need different formatting than Google?

Mostly the same formatting works for both, because both reward clear headings, self-contained paragraphs, and specific facts. The main difference is that AI engines tend to extract one paragraph in isolation, so every paragraph needs to make sense without relying on the sentence before it.

Q: How many internal links should a blog post have?

Two to four relevant internal links is enough for most posts. More than that starts to dilute which pages you actually want to pass authority to, and a reader can only reasonably follow so many tangents before losing the thread of the current post.

Q: What's the biggest mistake founders make writing their first blog posts?

Opening with background information instead of the answer. A reader who searched a specific question and gets three sentences of setup before the actual answer will bounce, and that bounce is itself a signal search engines pick up on over time.