SEO Copywriting in 2026: Write for Search Intent, Not Keywords

Somewhere around 2015, SEO copywriting stopped being about keywords and started being about intent. Somewhere around 2024, AI-generated summaries started answering questions before anyone clicked. Both shifts scared a lot of writers. Neither one changed the fundamental job: be the most genuinely useful answer to a question someone actually asked.

This is the working method I use for client content in 2026. It assumes you are a business owner or writer, not a technical SEO, and that you would like more of the right people to arrive on your site without pretending to be a robot.

What changed, and what did not

Three things genuinely changed in the last few years, and it is worth being precise about them because the panic has been less precise than the reality.

Zero-click results grew. A large share of informational queries are now answered directly in the results page or by an AI summary. That has hollowed out traffic for thin definitional content — “what is a CRM” pages are not the asset they were in 2019.

Evidence of experience started mattering more. Search quality guidelines lean hard on demonstrable first-hand experience. Generic synthesis is abundant and cheap; a page that shows someone actually did the thing is not.

Structure became machine-legible. Whether your content gets quoted in an AI summary depends heavily on whether a machine can extract a clean, self-contained answer from it. Which, pleasingly, is also what makes it easy for humans to skim.

What did not change: pages that satisfy the person who clicked outrank pages that do not, over any horizon longer than a few months. Everything below is in service of that.

The four intents, and how to tell them apart

Every query has an underlying job. Getting this wrong is the single most common reason a well-written page never ranks: it answers a question nobody was asking in that search box.

The reliable way to identify intent is not a tool. It is to search the query yourself, in an incognito window, in the country you are targeting, and look at what is already ranking. If the first page is all listicles, the search engine has decided that query wants a listicle. You may write a magnificent essay instead, and it will lose.

How to research intent in 40 minutes

Here is the routine, timed, that I run before writing any article.

Minutes 0–10: harvest questions. Collect real phrasings from your customer interview transcripts, your inbox, sales-call notes, the “people also ask” box, Reddit and forum threads, and your site’s own search log if you have one. You want the awkward, specific, badly-typed versions. “why isnt my website getting enquiries” is worth more than “website conversion optimisation”.

Minutes 10–25: read the top five results properly. Not skim — read. For each, note the format, the depth, the sub-questions it answers, and crucially what it fails to cover. Your angle lives in that gap. If all five are theoretical, your differentiator is a worked example with real numbers. If all five are shallow listicles, yours is the one with actual method.

Minutes 25–35: write the outline as questions. Every H2 becomes a question a reader would genuinely ask, in their words. This one habit does more for both readability and machine-extractability than any technical tactic.

Minutes 35–40: decide the one thing only you can say. Your proprietary data, your client story, your contrarian position, your process. If you cannot name it, you are about to write a page that adds nothing, and the algorithm will notice even if your team does not.

Structure: the shape search engines can read

The structural pattern that performs consistently in 2026 looks like this:

1. H1 that matches the query’s intent, not necessarily its exact wording.
2. A direct answer in the first 60–80 words. Do not warm up. If someone asked “how long should a homepage be?”, the answer appears before the anecdote about your client in Dunedin.
3. Question-shaped H2s, each followed by a self-contained answer of 40–120 words before you elaborate.
4. One table, list or checklist per major section — extractable, quotable, scannable.
5. Evidence of first-hand experience: numbers you measured, screenshots you took, mistakes you made.
6. A clear next step that is not a pop-up.

The self-contained answer rule is the one people skip. It means each section should make sense if lifted out of the page entirely, because increasingly that is exactly what happens to it.

Writing the title tag and meta description

Title tags are still the highest-leverage 60 characters on any page, and most CMSs will generate a bad one if you let them.

• Keep it under about 60 characters so it does not truncate. Front-load the words that match intent.
• Make a promise the page keeps. Inflated titles win the click and lose the ranking, because people bounce straight back.
• Include a differentiator — a number, a year, a qualifier: “18-point”, “in 2026”, “for small clinics”.

Meta descriptions do not directly affect ranking. They affect click-through rate, which affects everything downstream. Write about 150 characters, in second person, that name the reader’s problem and state what they will be able to do after reading. Then check it renders sensibly — search engines rewrite roughly two thirds of them, and yours should be good enough to survive.

Internal links are a ranking decision

Internal linking is the most under-used lever available to a writer, largely because it feels like admin rather than craft. It is craft.

Three rules:

• Link with descriptive anchor text. “Read our guide to brand voice” tells both a human and a crawler what is at the other end. “Click here” tells them nothing.
• Point articles at money pages deliberately. Every informational article should have at least one contextual link to the service page it naturally leads to — placed where the reader has just realised they need help, not bolted to the bottom.
• Build topic clusters. A pillar page linking to six deeper articles, each linking back, tells search engines you have depth on a subject rather than one lucky post.

Writing for AI summaries without writing for robots

Getting quoted in an AI answer follows a short list of properties, all of which happen to be good writing anyway:

• Answer first, elaborate second, at both page and section level.
• Attribute your facts. Say where a number came from and when it was measured. Unsourced claims are the first thing a summariser discards.
• Define your terms in one clean sentence somewhere on the page.
• Use consistent naming. If it is a “copy audit”, call it that every time — not “copy review” on paragraph four and “messaging assessment” on paragraph nine.
• Date your content and update it visibly. Recency is a signal, and an honest “updated August 2026” beats a hidden timestamp.

What does not work: stuffing questions you never answer, writing 3,000 words when 900 would do, or generating content wholesale and hoping. Synthesised text with no first-hand experience in it is the single most abundant commodity on the internet right now, and abundance is the opposite of a ranking advantage.

Refreshing old pages beats writing new ones

Most sites over two years old are sitting on more opportunity in their archive than in their content calendar. The refresh routine, done quarterly:

1. Pull every page ranking in positions 5–20 for a term you care about. These are the ones close enough to move.
2. For each, re-run the top-five read. What do the current winners cover that you do not?
3. Add the missing sections, update every number and date, remove anything now untrue.
4. Improve the title tag against current intent.
5. Add two internal links in, one internal link out.
6. Update the visible “last updated” date — honestly.

In my client work, a good refresh cycle routinely outperforms writing an equivalent number of new posts, at roughly a third of the effort. It is unglamorous, which is precisely why the opportunity is still there.

The on-page checklist

1. Have you searched the query yourself and matched the dominant format?
2. Is the direct answer within the first 80 words?
3. Is every H2 a question a real person would type or ask?
4. Does each section stand alone if extracted?
5. Is there at least one thing on this page that only you could have written?
6. Are all numbers sourced and dated?
7. Title tag under 60 characters, promise kept by the page?
8. Meta description around 150 characters, written in second person?
9. At least three descriptive internal links, one to a service page?
10. One table, list or checklist per major section?
11. Images — if any — carrying real alt text, not filenames?
12. Does the page load fast on a mid-range phone on mobile data?

None of this is a trick. It is the same bargain search has always offered: be the most useful answer, make it easy to tell that you are, and the traffic follows. The tools change every eighteen months. The bargain does not.