Most of the effort in a blog post goes into the draft. Most of the ranking gets decided before the draft and after it. That mismatch is why so many carefully written posts sit on page three while rougher ones built on the right question quietly pull traffic for years.
This is a walk through the life of one post, from the moment before you type to three months after it goes live, with the effort put where it counts.
What "Ranking" Means Now
Position one is no longer the prize it was. For a large share of informational searches, an AI Overview now sits above the results and answers the question directly, with a handful of source links tucked inside it. Google's AI Mode goes further and handles whole conversations without the user ever leaving the results page.
The click numbers reflect that. Studies through 2025 and 2026 put the drop in click-through on queries with an AI Overview somewhere between a third and more than half, depending on device and market. At the same time, pages that get cited inside those overviews earn noticeably more clicks per impression than pages that don't. The traffic hasn't vanished. It has moved to fewer, better-chosen sources.
So ranking in 2026 means three things at once: appearing in the classic results, being one of the sources an AI answer draws on, and being good enough that the person who reads the summary still wants the full version. The rest of this piece is built around earning all three, and the rough time split looks like this:
| Stage | Share of total effort | What most people actually spend |
| Choosing the question and reading the results | 30% | 5% |
| Working out what you know that the results don't | 20% | 0% |
| Drafting and structuring | 30% | 80% |
| Proof, publishing checks, follow-up | 20% | 15% |
Before You Type: Pick a Question Someone Actually Asks
Keyword volume is the wrong starting point. A term with 20,000 monthly searches and an AI Overview that answers it in three sentences will send you almost nothing. A question with 400 searches that an AI summary handles badly can send you every one of those people.
Start from a question, written the way someone would type it or say it. "Best running shoes" is a keyword. "Which running shoes hold up for someone who's 90 kilos and runs on pavement" is a question, and it's the kind of long, specific query that AI search has made more common, not less. Google's own guidance on its AI features says users are asking longer, more specific questions and following up with more of them.
Three sources for questions worth answering:
- The questions you get asked repeatedly by customers, readers or clients, especially the ones you're slightly tired of answering. Repetition is demand.
- Search Console's query report for your existing pages, filtered to queries where you get impressions but few clicks. Those are questions people are asking near your content that your content isn't quite answering.
- Forum threads and comment sections in your niche where someone asks something and the replies are all "it depends." That's a gap an AI summary can't fill and you can.
Before moving on, write the question at the top of your document in full. Everything below it has to answer that question and nothing else.
Read the Results Page Like a Competitor
Spend twenty minutes on the current results for your question before you write a word. Not skimming. Reading. You're looking for four things.
First, the format that wins. If every top result is a comparison table, a listicle won't compete. If they're all long tutorials, a 600-word answer won't either. The format is a signal about what searchers wanted last time, and you either match it or beat it deliberately.
Second, what the AI Overview says, if there is one. Read it twice. Note which pages it cites and what it lifted from each. Then ask the harder question: what did it get vague about, skip, or flatten? Those gaps are where a click still happens.
Third, what every top result says. If nine out of ten pages make the same five points, those five points are the price of entry. You have to cover them, briefly, and they won't earn you anything on their own.
Fourth, what none of them say. This is the only thing on the page worth more than twenty minutes, and it feeds directly into the next stage.
Decide What You Know That the Top Results Don't
Here is the uncomfortable truth about ranking in 2026. If your post could have been written by someone who had only read the top ten results, it is the top ten results, rearranged. Search systems have become very good at recognising that, and AI summaries make it worse: they compress commodity content into a paragraph, and nobody clicks through to read the same paragraph stretched to 1,500 words.
Google's advice on this hasn't changed in years, and it now matters more than it used to: unique, non-commodity content, made for the people reading it. The word doing the work there is "unique." What can you put on the page that a summariser can't generate and a competitor can't copy without doing the work?
● Something you did. You tested the method, ran the tool, cooked the recipe four times, made the mistake. First-hand experience shows up in details nobody invents: the step that took twice as long as expected, the setting that turned out not to matter.
● Something you counted. A survey of your own customers, twelve months of your own numbers, a dataset you built. Original figures get cited, by other writers and by AI systems, because they can't be found anywhere else.
● Something you believe. A clear view on which option is best for whom, stated plainly, with the reasoning shown. "It depends" is what the summary already says.
● Something you noticed. A pattern across many cases, a common mistake, a reason the standard advice fails for a specific kind of reader.
Pick one. It becomes the spine of the post. If you can't find one, the honest move is to write a different post, because this one will be invisible.
Draft One: Write the Answer First
Open with the answer. Not the history of the topic, not why the question matters, not three paragraphs about your journey. The direct answer to the question at the top of your document, in two to four sentences, before the first heading.
This isn't only about readers, though readers reward it. It's how AI systems and featured results identify the passage that answers the query. A page whose answer is buried under 300 words of preamble is a page whose answer gets missed, or gets extracted with the wrong context.
Compare two openings for a post on fixing a slow WordPress site:
Weak: "In today's fast-paced digital world, website speed is more important than ever. A slow site can hurt your rankings, frustrate visitors and cost you conversions. In this guide, we'll explore everything you need to know about speeding up WordPress."
Strong: "Most slow WordPress sites have one of three problems: a bloated theme, too many plugins loading on every page, or cheap shared hosting. Fixing them in that order usually takes an afternoon. Here's how to tell which one you've got."
The second one answers, promises something specific, and gives the reader a reason to keep going. Then the body earns the promise: each section takes one part of the answer and gives it the depth the summary didn't.
Draft one is allowed to be rough. What it's not allowed to be is padded. If a paragraph exists to reach a word count, cut it. Length that comes from substance ranks; length that comes from filler gets summarised and skipped.
The Structure Machines and Humans Both Read
Structure is where a good draft becomes a page that can be used. Three rules do most of the work.
Headings are questions, or close to it. "Pricing" is a label. "How much does it cost to run this each month?" is a heading that matches how people search, and it tells a summariser exactly what the section answers.
The first sentence under each heading answers it. Everything after that is explanation, evidence and detail. If a reader lifted only the first sentence of every section, they'd have a complete short version of the post. That's also what an extraction system does.
Comparisons go in tables, sequences go in numbered steps, everything else stays in short paragraphs. Tables aren't decoration; they're the format a comparison is actually understood in, and they're the format most likely to be pulled into a results page whole.
Here's the skeleton of a post built this way. Copy it into your editor and replace the labels.
Title: the question, or the answer, in plain words Direct answer (2 to 4 sentences, no heading above it) Section 1: [question the searcher has first] First sentence answers it Evidence, example, or first-hand detail Section 2: [the next question that follows naturally] First sentence answers it Table, if this is a comparison Section 3: [the part where your unique knowledge lives] This is the longest section. It's the reason to click. Section 4: [common mistake or edge case] Close: what to do next, one paragraph Author line, updated date, sources |
Notice that the unique section isn't first. The commodity questions come first because searchers need them answered to trust you, and they're short because the answers are known. The depth goes where the value is.
The Proof Layer
Trust used to be something a domain earned over time and passed down to its pages. It still is, but individual pages now carry more of the load, and the signals are concrete and visible.
A real author, with a name, a short line on why they're qualified to write this, and a page or profile that exists elsewhere. Anonymous posts on informational topics are at a clear disadvantage, and "by Admin" is the same as anonymous.
Sources that are linked, not just mentioned. When you cite a figure, the reader should be able to reach it in one click. This also makes the page easier for AI systems to verify against, which affects whether it gets cited.
Evidence of experience, in the body, not the bio. Your own screenshots, your own photos, your own numbers, the detail only someone who did it would know. Stock images signal nothing; a photo of your actual setup signals everything.
A visible date, and an honest one. An "updated" date that's changed without the content changing is noticed, and it costs more trust than it buys.
None of these are tricks. They're the things a careful reader already looks for, made explicit so that neither the reader nor the system has to guess.
The Hour Before You Publish
The post is written. This is the only checklist in the piece, and it takes about an hour.
| Check | What good looks like |
| Title | Contains the question or its answer in plain words, under about 60 characters, no clickbait promises the post doesn't keep |
| URL | Short, readable, three to five words, no dates or numbers you'll regret when you update it |
| Meta description | The direct answer or the specific promise, in one or two sentences, written for a person deciding whether to click |
| Internal links in | At least two existing posts on your site now link to this one with descriptive anchor text |
| Internal links out | The new post links to your related pages where a reader would naturally want more |
| Images | Compressed, with alt text that describes what's in them, and at least one that's yours |
| Structured data | Article schema with author and dates; FAQ or HowTo only where the page genuinely has that content |
| Phone read | You've read the whole thing on your phone, and the answer is visible without scrolling |
The phone read is the one people skip and the one that catches the most. Paragraphs that looked short on a monitor turn into walls, tables overflow, and the direct answer slides below the fold.
The First Thirty Days
Publishing is the middle of the process, not the end. In the first month, three things matter and one thing doesn't.
Links from your own site matter. Go back through your older posts on related topics and add a link to the new one wherever a reader would benefit. This is the fastest way to tell search systems the page exists and where it fits.
Putting it in front of the people who already asked the question matters. The forum thread from earlier, the newsletter, the customers who asked. Early readers who stay and read to the end are the strongest early signal a page can have.
Watching the right report matters. Search Console has started separating impressions from AI Overviews and AI Mode into their own segments, and a dedicated generative AI performance report is rolling out to sites gradually. Look for queries where the page is appearing that you didn't target. Those tell you what the page is actually about in the system's eyes, which is sometimes different from what you intended.
What doesn't matter yet is the ranking position. Pages move around for weeks after publishing. Rewriting a post in week two because it's sitting at position eleven is like repotting a plant every day to see if the roots are growing.
Three Months Later: Update, Merge or Let Go
At ninety days the post has settled enough to judge. There are three outcomes and each has a right response.
It's getting impressions and clicks for the question you targeted, or close to it. Update it. Add what you've learned since, tighten the sections that readers skip, refresh any figures, and change the date only if the content actually changed. A post that ranks is worth more than three new ones that might, and it compounds with every honest update.
It's getting impressions for a different question than the one you wrote it for. The page has found its own audience. Either reshape it toward that question, or split that section into its own post and let the original stay focused. Don't leave it half-serving two intents.
It's getting almost nothing, and a look at the results shows why: a stronger page, a format you didn't match, or a question that turned out to be fully answered in the overview. Merge whatever's useful into a related post that does perform, redirect the URL, and let it go. Thin pages that nobody reads are dead weight on the rest of the site.
The Post That Still Ranks Next Year
Everything here is aimed at one property: durability. Posts built on commodity information have a short life now, because the summary replaces them. Posts built on something only you could have written last, because there's nothing to replace them with.
That's the whole shift in a sentence. The systems have become good at answering questions that many people can answer. What's left, and what gets cited, clicked and returned to, is the question only you can. Write that one.
