SEO Content Writing: How to Write Pages That Rank and Get Cited (2026)

I have spent twenty years watching people write content for keywords instead of people. They open a tool, find a phrase with volume, and start typing. The page reads fine. It ranks for nothing. The problem started before they wrote a word, because they never asked what the person searching actually needed to get done.

A keyword is a clue, not an instruction. When someone types "best running shoes for flat feet," the keyword is the surface. The job underneath is "help me not buy the wrong shoes and end up in pain again." Those are different problems. One gets you a thin listicle. The other gets you a page people finish reading and act on.

The fastest way to learn the job is to look at what already ranks, but not to copy it. Read the top ten results the way the searcher would. What question do they all answer first? What do they assume you already know? Where do they leave you hanging? That gap is usually the job nobody is doing well, and it is where your page wins.

Read intent as a verb, not a noun

I sort every query by the action behind it. Is the person trying to learn, decide, compare, buy, fix, or do? "How to deglaze a pan" is a do query and wants steps you can follow with greasy hands. "Le Creuset vs Staub" is a decide query and wants a clear verdict, not a fence-sit. When you write the wrong shape for the verb, no amount of word count saves you.

• Learn: definitions, mental models, context. The reader wants to understand.
• Decide: comparisons, criteria, a recommendation they can act on.
• Do: numbered steps, screenshots, the exact sequence in order.
• Buy: specs, price, availability, trust signals, a clear next step.
• Fix: the cause, the solution, and how to confirm it worked.

Bad content is usually a planning failure wearing a writing costume. By the time you are fighting a sentence, the real damage was done earlier, when nobody decided what the page was for, who it was for, and what it had to prove. A real brief prevents that. It is the cheapest quality control you will ever run.

My brief is short on purpose. One sentence on the job to be done. One on the searcher and what they already know. One on the single thing this page must do better than everything ranking now. If I cannot fill those three lines, the page is not ready, and writing it anyway just means I find that out after wasting an afternoon.

Build the outline from questions, not headings

Headings come last. First I list every real question the searcher has, in the order they would naturally ask them. "What is it? How does it work? How do I know if I need it? What does it cost? What goes wrong? What do I do next?" Then I turn the questions into a logical sequence and only then write the headings. An outline built from questions answers things in the order a human would ask them, which is also the order an AI engine can follow.

1. 1Write the one-sentence job the page completes.
2. 2List the searcher's questions in the order they would surface.
3. 3Group and sequence them into a logical flow.
4. 4Decide the one piece of information gain this page adds.
5. 5Convert the question flow into H2s and H3s.
6. 6Mark which sections need data, a screenshot, an example, or a quote.

Here is the contrarian part. The web does not need another summary of what already ranks. If your page is a competent restatement of the top five results, you have built a worse version of pages that already exist, and search engines can see that. Information gain is the difference between your page and the consensus. No gain, no reason to rank you.

Information gain does not mean being weird for its own sake. It means adding signal: a number nobody else published, a step everybody skips, a failure mode you actually hit, a clearer model for something everyone explains badly. Even one genuine addition changes the page from derivative to worth citing.

Where gain actually comes from

• First-hand results: what happened when you actually did the thing.
• Original data: a small survey, your own test, numbers from your own work.
• A sharper framework: explaining a confusing topic in a way that finally clicks.
• Specificity: exact settings, exact prices, exact sequences others keep vague.
• Counter-consensus: a defensible take on where the popular advice is wrong.

You do not need original gain in every paragraph. You need it to be unmistakable somewhere. One section that no competitor could have written, because it came from your own work, is worth more than 2,000 words of careful synthesis. That one section is also the part AI engines are most likely to quote, because it is the part they cannot assemble from everywhere else.

The bar is not "is this accurate." The bar is "does this exist anywhere else." Accurate and unoriginal still loses.Shmul

Two audiences read your content and neither reads it the way you wrote it. Humans skim, hunting for the part that answers their question. AI engines extract, pulling self-contained chunks to quote in an answer. The good news is that the same structure serves both. Clear hierarchy, answer-first sections, and standalone passages win twice.

Lead each section with the answer, then support it. The old instinct is to build up to the point. Online that loses, because the skimmer leaves before you arrive and the AI engine grabs the first clear sentence under the heading. Put the conclusion in the first line of the section, then explain, qualify, and prove it underneath.

Make sections self-contained

An AI engine often quotes a single paragraph out of context. If that paragraph relies on "as I mentioned above" or an unstated antecedent, it is useless when lifted out. Write key passages so they stand on their own. Name the subject instead of saying "it." Restate the condition instead of assuming the heading carries it. This feels slightly redundant when you read top to bottom, and it is exactly what makes the chunk quotable.

• Descriptive headings that state the answer or the question, not clever labels.
• Short opening sentence per section that delivers the point immediately.
• Lists for parallel items, steps for sequences, tables for comparisons.
• One idea per paragraph, so a skimmer can scan first lines and still follow.
• Self-contained key passages that survive being quoted in isolation.

Formatting is not decoration. A list of five parallel items reads faster as a list than as a comma-jammed sentence, and an AI engine parses it more reliably too. A sequence of actions belongs in numbered steps. A head-to-head belongs in a table. Choosing the right structure for the content is half of readability, and it is doing real work for extraction at the same time.

E-E-A-T gets talked about like a checkbox. It is not. The extra E, experience, is the one that matters most right now, precisely because it is the hardest thing to manufacture at scale. Anyone can summarize. Far fewer people can tell you what actually happened when they did the thing, and that gap is your advantage.

Experience shows up in the details that only doing the thing produces. The setting that was not in the manual. The step where it went wrong. The number you measured yourself. You cannot fake these convincingly, which is the entire point. When a reader hits a detail like that, trust jumps, because the page clearly came from someone who has been there.

Demonstrate, do not declare

Saying "I am an expert" is worthless. Showing your reasoning, your tradeoffs, and the edge cases you know about does the job without the claim. Expertise leaks out in what you choose to warn about, what you tell people to skip, and where you say the popular advice breaks down. Authority is the reputation that builds on top of repeatedly demonstrating that, page after page.

• Experience: "When I ran this on a live site, here is what broke."
• Expertise: "Most guides skip this setting; here is why it matters."
• Authoritativeness: a track record others reference and link to.
• Trust: clear sourcing, honest tradeoffs, no overclaiming.

I use AI to write. I also publish almost nothing it produces unedited. The mistake people make is treating the model as the author instead of the assistant. Used well, AI removes the blank-page problem and speeds up the boring parts. Used lazily, it produces the exact generic mush that search engines and readers are learning to ignore.

The reason raw AI content reads as slop is that the model writes toward the average of everything it has seen. The average is, by definition, the consensus, which is the opposite of information gain. So if you let it run unsupervised, you get a competent restatement of what already ranks, which is the one thing you cannot win with.

How I actually use it

1. 1Feed it my brief, my outline, and my own raw notes, not just a topic.
2. 2Have it draft sections, not whole articles, so I keep control of structure.
3. 3Inject my own data, examples, and first-hand details by hand; the model cannot.
4. 4Use it to tighten and rephrase my writing, not to invent claims.
5. 5Fact-check every specific it produces, because it will state wrong things confidently.
6. 6Rewrite the opening and the key passages myself so the voice is mine.

The non-negotiable input is your own material. The model can structure, phrase, and accelerate, but it cannot supply your first-hand experience, your test results, or your point of view. Those are the parts that create information gain, and they have to come from you. If you have nothing of your own to add, AI will not save the page, it will just help you publish the duplicate faster.

AI did not lower the bar for content. It raised it, by making mediocre content infinite. The only thing that stands out now is the part a model could not have written.Shmul

Writing is generating raw material. Editing is making it good. Most people skip the second part, which is why so much content reads padded, hedged, and forgettable. The draft is not the deliverable. The draft is the thing you carve the deliverable out of, and the carving is where the quality comes from.

My first editing pass is subtraction only. I delete every sentence that does not move the reader toward finishing the job. Throat-clearing intros, restated points, hedging qualifiers, words that exist to hit a count. A page that says the same thing in 1,200 tight words beats one that says it in 2,500 loose ones, for the reader and for the machine reading density.

The passes I run

• Cut: remove anything that does not serve the job. Be ruthless.
• Sharpen: replace vague claims with specific ones and weak verbs with strong ones.
• Commit: turn hedges into recommendations; say what you actually think.
• Front-load: move the answer to the top of each section.
• Read aloud: anything that trips your tongue trips the reader too.

Hedging is the silent killer. "It might be a good idea to consider possibly testing" is four qualifiers stacked into saying nothing. "Test this" is a sentence with a spine. Readers came for a recommendation, not a survey of what could theoretically be true. When you have done the work to have a view, state it plainly, and let the supporting detail earn the confidence.

Most people treat publishing as the finish line. It is the starting line. Content decays. Facts go stale, the results shift, competitors catch up, and the searcher's expectations move. A page that ranked last year can quietly slide while you write new ones, and a disciplined refresh program often beats publishing more.

I would rather refresh a page that already has traction than publish a new one from zero. The existing page has history, links, and rankings to build on. A real update, new information, removed dead weight, fresh examples, often recovers lost positions faster and cheaper than any new piece, because you are improving an asset instead of starting one.

What a real refresh involves

• Update facts, prices, screenshots, and anything time-sensitive.
• Add information gain that has appeared since you published.
• Remove advice that is now wrong or no longer applies.
• Re-check the search results: has the intent or the bar shifted?
• Improve the structure and extraction-friendliness with what you have learned.

Build a schedule. I review high-value pages on a cadence, not when I happen to remember. Pages tied to fast-moving topics get checked more often than evergreen ones. The point is that maintenance is planned work, not a reaction to a ranking drop you noticed three months too late, by which time the recovery is harder.

If you cannot tell whether a page worked, you cannot get better at writing them. Most people measure the wrong thing, usually rankings alone, and then wonder why traffic does not turn into anything. Performance is whether the page did the job: brought the right people, answered them, and moved them to the next step. Measure that.

Rankings are an input, not an outcome. A page can rank first and convert nothing because it pulled the wrong audience or answered the question so completely the reader had no reason to go further. Start from the job the page was supposed to do, then pick the metric that proves it happened. The metric follows the goal, not the other way around.

• Impressions and position: is the page being shown for the right queries?
• Click-through rate: is the title and snippet earning the click?
• Engagement: are people reading, or bouncing back to the results?
• Conversions or the intended action: did the page move them forward?
• Citations in AI answers: is the page being quoted by generative engines?

Read the metrics together

No single number tells the truth. High impressions with a low click-through rate means your snippet is weak or the intent is mismatched. Good clicks with fast bounces means the page over-promised and under-delivered. Strong rankings with no conversions means you ranked for the wrong job. The pattern across metrics tells you what to fix; one metric in isolation usually misleads.

Define what success looks like before you publish. If you decide afterward, you will pick whatever metric happens to look good.Shmul