SEO Copywriting in 2026: Writing for Search, Readers and AI Answers

SEO copywriting is writing web content that ranks in search and persuades the reader to act. It combines search intent analysis, keyword research and page structure with persuasive writing, and in 2026 it also means writing passages clean enough for AI-generated answers to quote and attribute.

You publish content and it disappears. The pages go live, the word counts are respectable, and six months later the Search Console numbers show impressions climbing while clicks sit in single figures. Somewhere a competitor with worse design and a smaller budget is taking the traffic.

The gap is rarely effort. It is that the writing was briefed against a keyword rather than a reader, produced without anything only your business could say, and structured so neither a skimming buyer nor a retrieval system can find the answer in it. Meanwhile the cost of producing generic text has fallen to nothing, so generic text now competes with infinite supply.

This guide covers what changed and what to do about it: how to establish intent before writing, how to structure a page so it survives extraction, how to get real experience into the copy, where AI tools help and where they damage you, and the production process that makes the quality repeatable across a team.

What SEO Copywriting Means in 2026

SEO copywriting is writing that earns a ranking and then earns the action. It serves the person searching and the systems deciding what to show them, and when those two goals conflict the reader wins, because a page nobody acts on is worth nothing however well it ranks.

The discipline has changed more in the last two years than in the previous ten. Two forces did it. The first is generative search. A meaningful share of queries now return an AI-assembled answer above the results, which changes what a click is worth and rewards writing that can be extracted and cited rather than merely indexed.

The second is the collapse in the cost of producing text. Anyone can generate five thousand words on any subject in a minute, which means volume has stopped being a competitive advantage and the things a model cannot fake have become the whole game. What survives is writing that carries evidence. Original numbers, specific experience, named examples, and a point of view that could only come from doing the work.

The Three Jobs a Page Has to Do

Every page that ranks and converts is doing three jobs, and most underperforming pages are doing one of them well and the other two badly. It has to match the intent behind the query, so the reader arrives and immediately recognises the page as the right one.

It has to be readable by machines, which means clear structure, direct answers near the top, and content that holds together when a passage is lifted out of context. It has to move the reader towards a decision, which means answering the objections that actually stop people buying rather than the ones that are comfortable to answer.

Writing that does all three is rarer than it should be, mostly because the three jobs are usually assigned to three different people who never speak.

What It Is Not

It is not writing a target number of keyword mentions into a document. Keyword density as a concept has been useless for a decade and actively harmful for about half that. It is not producing long content for the sake of length. Long pages tend to rank because subjects that need covering properly take words, and adding words to a subject that does not need them produces a page that bores people.

It is not a task you can hand to a generalist writer with a keyword and a word count. Our SEO copywriting work starts with the search data and the commercial goal, because those determine what the piece has to be before anyone writes a sentence.

Where the Reader Actually Is

Writing improves when you are honest about the state the reader arrives in, which is rarely the state the brief assumes. They are on a phone more often than not. They have four other tabs open with your competitors in them. They have been burned before by someone who promised this exact thing.

They are not reading. They are scanning for the part that tells them whether to keep going, and they will give the page somewhere between five and fifteen seconds to provide it.

They are also, frequently, not the decision maker. They are the person tasked with shortlisting, who will have to justify the choice to someone else. That means the page needs to give them the argument, not just the pitch, because they will be repeating it to a colleague without you in the room.

Writing for that reader changes concrete things. The answer goes at the top. The proof goes in a form that can be copied into an email. The objections get handled explicitly, because they will be raised by the person who was not on the call.

Content that equips an internal champion outperforms content that impresses a browser, and the two are written differently.

Why the Old SEO Copywriting Playbook Stopped Working

The tactics that worked in 2018 are not merely less effective now. Several of them actively damage a site, and they persist because they are easy to teach and easy to invoice for.

Tactics That Have Stopped Paying

Writing to a keyword density target produces stilted copy that readers abandon and that Google has been able to detect for years. The phrase should appear where it belongs and nowhere else. Publishing a separate page for every minor keyword variation splits your own signal. Fourteen near-identical pages about the same intent compete with each other and none of them wins.

Stuffing a 300-word answer inside a 3,000-word page to hit a length target teaches readers that your content wastes their time, and the behavioural signals follow. Spinning existing content into slightly different words at scale is the practice Google’s spam policies name as scaled content abuse, and the enforcement has become considerably less forgiving.

Writing exclusively for informational queries and expecting commercial return. Those are the queries most likely to be answered by an AI Overview without a click.

What Replaced Them

Intent matching replaced keyword matching. The question is not whether the phrase appears but whether the page does the job the searcher wanted doing. Topical depth replaced page count. One thorough page plus a cluster of genuinely distinct supporting pieces beats forty thin ones, and it is cheaper to maintain.

Demonstrated experience replaced claimed authority. “We are experts” is worthless. “Here is what happened when we did this eleven times, including the two that went badly” is not. Extractability replaced optimisation. Writing so a machine can lift a clean, correct, self-contained passage is now a core skill rather than a nice extra.

The Uncomfortable Part

Some content that used to earn traffic will not earn it again, and no amount of rewriting will bring it back. Definitional articles answering a simple factual question are the clearest case. If the whole value of the page was telling someone what a term means, an AI Overview now does that above your result and the click does not happen.

The honest response is to stop producing that sort of page and redirect the effort towards queries where the searcher still has to reach a supplier, and towards content that earns citations even when it does not earn clicks. This is a real reduction in the addressable traffic for some businesses. Pretending otherwise leads to reporting that blames the writer for a structural change.

What to Do With the Content You Already Have

Most businesses reading this have a back catalogue rather than a blank page, and the instinct to leave it alone and publish something new is usually wrong. Start by sorting what exists into four groups using Search Console data over sixteen months. Pages earning clicks are working. Leave them alone except for an annual refresh of any figures or dates that have gone stale.

Pages earning impressions and almost no clicks are the priority. Google considers them relevant and readers reject them, which usually means a weak title, a buried answer or content thinner than the pages above it. These are the cheapest wins on the site. Pages earning neither are either targeting terms nobody searches or have never been indexed properly. Check indexation first, then decide whether the subject is worth owning at all.

Pages competing with each other for the same term should be merged into one stronger piece with a redirect from the loser. Updating a page that already has impressions beats publishing a new one, because the hard part, convincing Google the page is relevant, is already done. Our ongoing SEO checklist covers how often to run this review.

Search Intent: The Decision That Shapes Everything

Intent is the first decision and it determines format, length, structure and call to action. Get it wrong and the quality of the writing cannot rescue the page.

The Four Intents and What Each Needs

Informational intent means the searcher wants to understand something. They need a clear direct answer near the top, then depth, then related questions. The conversion offer should be low commitment, because they are not buying today. Commercial investigation means they are comparing options before choosing. They need honest comparison, criteria for deciding, evidence, and a frank account of where each option fits. Pretending there is one right answer loses them.

Transactional intent means they are ready to act. They need specifics: what you do, how it works, what it costs or how pricing is structured, and an obvious way to start. Long preamble here is an obstacle. Navigational intent means they want a particular page or brand. There is little to write and much to make findable.

How to Establish Intent Properly

Search the term and read what already ranks, because the results are Google’s own answer to what the query means. If the first page is all guides, the intent is informational and publishing a service page will not rank however good it is. If it is all product and service pages, a guide will not rank either.

Look at the search features too. An AI Overview or a People Also Ask block signals a question to be answered. A local pack signals geographic intent. Shopping results signal transactional. When the results are mixed, the query has mixed intent and you have a choice to make. Usually the right one is to serve the dominant intent well rather than to serve both badly on one page.

Where Intent Analysis Goes Wrong

The most common error is assuming intent from the words rather than checking the results. “Best CRM for small business” looks commercial and often returns listicles rather than product pages, which tells you what format can actually win.

The second is writing for the intent you wish people had. A business selling a considered service wants transactional queries, and most of its search demand is informational. Writing service pages for informational queries produces pages that rank for nothing.

The third is ignoring how intent shifts. Terms that were informational two years ago become commercial as a market matures, and content planned against stale intent data slowly stops working. Recheck the results for your priority terms at least twice a year.

A Worked SERP Analysis

Take the term “lead generation agency” and work through what the results tell you before writing a word. The first page returns a mix: two or three agency service pages, several listicles ranking agencies, and usually a guide on how to choose one. That mix is the signal.

It tells you the intent is commercial investigation rather than transactional. People searching it are comparing rather than buying, which means a service page alone will struggle and a piece that helps them compare has a route in.

It also tells you what format can win. If three of the top five are listicles, publishing a fourth listicle means competing on the incumbents’ terms with less authority. Publishing the honest buyer’s guide none of them wrote is a better bet. Check the search features. A People Also Ask block listing questions about cost, contract length and how to judge results tells you exactly which sections the page needs.

Twenty minutes reading the results saves rewriting the piece later, and it is the step most often skipped because it produces no deliverable.

Keyword Research for Writers

Most keyword research is done by someone who will not write the piece and handed over as a spreadsheet. The writer then has a list of phrases and no idea what the page is supposed to be. Research done properly produces a brief, not a list.

What the Writer Actually Needs

One primary keyword, which is the phrase the page is genuinely about. Not the highest-volume phrase in the cluster, but the one that matches what the page will do for the reader. Eight to twelve secondary terms that represent genuine subtopics, each earning a section rather than a mention. If a secondary term cannot carry a section, it is a variant rather than a subtopic and it does not need planning for.

The questions people ask around the subject, which come from People Also Ask, from forums and from your own sales team’s inbox. Those become headings and FAQ entries. The pages currently ranking, with a note on what each covers and what none of them covers. That gap is usually where the piece earns its place.

Choosing the Primary Keyword

Volume is the least important input and the one most people lead with. Relevance comes first. A term with 200 searches a month that describes exactly what you sell is worth more than 20,000 searches of people who will never buy.

Difficulty comes second, assessed honestly against your site rather than in the abstract. A site with modest authority will not win a term dominated by national publishers, and the effort is better spent where it can succeed. Commercial intent comes third. Terms containing “agency”, “services”, “cost”, “best” or “near me” convert at a different order of magnitude from terms containing “what is”.

Pick the phrase a real person would type, not the phrase that looks tidiest in a spreadsheet. Awkward compound keywords give away that a page was built for a tool rather than a reader.

Building the Cluster

Once the primary term is fixed, group the remaining terms by whether they belong on this page or on a different one. Terms a reader would expect answered on this page become sections. Terms that represent a genuinely separate question become their own piece, linked from here.

The test is the same one that governs architecture generally. If two terms would be satisfied by the same content, they are one page. If a reader searching each would want something different, they are two, and merging them creates keyword cannibalisation.

A Worked Brief

Here is what a usable brief looks like, as opposed to a keyword list. Primary keyword: SEO copywriting. Intent: informational with commercial investigation underneath, because searchers are usually deciding whether to hire someone or do it themselves. What the piece must prove: that we understand how the discipline changed in the last two years, and that we have a repeatable process rather than a talented individual.

Sections required, from the People Also Ask block and the sales inbox: what it is, how long pages should be, whether AI can be used, how long ranking takes, and how it differs from content writing. The gap: every ranking page treats AI use as either forbidden or unlimited. None sets out a workable division of labour. That is our angle.

Evidence needed: the division of labour we actually use, one example of a generated claim we caught in the accuracy pass, and the interview questions we ask delivery teams. Links planned: the SEO copywriting service page, the keyword cannibalisation post, the two AI search guides, and Google’s own guidance on generative AI content.

That brief takes thirty minutes to write and removes almost every reason a draft comes back wrong.

Structuring a Page So It Can Be Read and Extracted

Structure used to be about readability. It is now also about whether a machine can lift a clean passage from your page and attribute it to you.

The Structural Rules That Matter

One H1 stating what the page is. H2s for the major questions the page answers. H3s for the components of each. No skipped levels, ever. Phrase headings as the questions people actually ask, because those headings are what a retrieval system matches against and what a scanning reader uses to decide whether to stop.

Keep paragraphs short and self-contained. A paragraph that depends on the previous three to make sense cannot be extracted, and a reader skimming on a phone will not reconstruct the argument for you. Put the answer before the explanation in every section. The reader who wants the short version gets it, and the reader who wants the reasoning reads on.

Writing Self-Contained Passages

This is the skill that has changed most and the one writers find hardest, because good prose usually builds. A self-contained passage restates enough context to stand alone. Instead of “this means you should avoid it”, write “thin service pages rank poorly, so avoid publishing them”.

Pronouns are the usual culprit. “It”, “this” and “they” referring back across paragraphs make a passage useless the moment it is lifted. Numbers and named thresholds travel well and vague quantities do not. “Under 2.5 seconds” survives extraction. “Reasonably fast” does not.

Formatting That Earns Its Place

Tables are the most underused format in SEO copywriting. Any comparison of three or more things across two or more attributes belongs in a table, and tables are extracted readily. Numbered lists suit sequences where order matters. Bulleted lists suit sets where it does not. Using the wrong one quietly misleads the reader about whether sequence matters.

Bold should mark the single most important claim in a paragraph, which makes a skim-read coherent. Bolding half a paragraph marks nothing. Avoid decorative structure. A page broken into thirty tiny sections to look scannable reads as fragmented and gives every heading less weight.

A Worked Restructure

Here is a section that fails extraction, and the same section fixed. Original: “We covered the three metrics above. It is worth noting that they are measured differently from what most people assume. This is because the data comes from real users rather than from a test, and that distinction matters a great deal when you are trying to work out why your scores look the way they do.”

Nothing in that paragraph can be lifted. “They”, “this” and “it” all refer backwards, and the actual fact never arrives. Fixed: “Core Web Vitals are measured using field data from real Chrome users, not from a simulated lab test. That distinction explains why a page can score well in Lighthouse and still fail Core Web Vitals: the lab test runs on your machine, while the assessment uses the slowest quarter of your actual visitors.”

The second version states the subject, carries the fact, includes the qualifying detail and stands alone. It is also clearer for a human reading in sequence, which is the point worth noticing. Writing for extraction and writing clearly turn out to be nearly the same instruction.

Writing the Opening That Wins the Snippet

The first fifty words under the H1 do more work than any other part of the page. They decide whether an AI Overview cites you, whether a featured snippet picks you up, and whether the reader stays.

What the Opening Has to Contain

The direct answer to the query, stated as plainly as you can manage. Who it is for, so the wrong reader can leave and the right one knows they have arrived. Why it matters, in a clause rather than a paragraph. That is roughly fifty words and every one of them has to earn its place.

What to Leave Out

No preamble. “In today’s competitive market” tells the reader nothing and burns the extraction window. No announcement of what the article will do. “In this guide we will explore” delays the answer the reader came for. No label. Writing “Quick answer:” or “TL;DR” above the paragraph announces the page as machine furniture. The paragraph should simply be the first paragraph, because that is what it is.

No hedging. A sentence qualified three ways cannot be extracted as an answer and reads as though you are not sure.

A Before and After

Weak: “Conversion rate optimisation is something many businesses are talking about in today’s digital world. In this comprehensive guide, we will explore everything you need to know about CRO and how it can help your business grow.” That is forty words and it contains no information.

Stronger: “Conversion rate optimisation is the practice of increasing the share of visitors who complete a desired action, using research and testing rather than guesswork. It suits businesses with existing traffic and disappointing enquiry volume, and it usually pays faster than buying more traffic.” That is forty-two words, answers the query, qualifies the audience and states the benefit. It can be lifted whole.

Openings That Fail, and Why

Four patterns account for most weak openings, and all four are easy to spot once named. The scene-setter opens on the state of the industry. “The digital marketing world moves quickly and businesses must keep pace.” The reader already knows, and you have spent the extraction window agreeing with them.

The throat-clear announces the article. “In this comprehensive guide, we will walk you through everything you need to know.” That is a promise rather than information, and it delays the thing they came for. The definition-dodge answers a slightly different, easier question than the one asked. Someone searching how to fix a problem gets a paragraph explaining what the problem is. It reads as evasion even when it is not.

The over-qualified answer hedges the claim into meaninglessness. “There are many factors and results may vary depending on circumstances.” Nothing in that can be cited, and nothing in it helps. The fix in every case is the same. Write the sentence you would say if a client asked you the question directly across a table, then start there.

Headlines, Title Tags and Meta Descriptions

These three are usually written last in five minutes, and they determine whether any of the work gets seen.

The H1 and the Title Tag Are Different Jobs

The title tag appears in search results and has to win a click against nine competitors. It is 50 to 60 characters, keyword near the start, written for someone choosing. The H1 appears on the page and has to confirm the reader made the right choice. It can be longer, warmer and more specific, because the click has already happened.

Making them identical wastes one of them. Vary the phrasing so the H1 adds something the title could not fit.

Writing a Meta Description That Works

Meta descriptions do not influence rankings and they strongly influence click-through, which is worth the same in the end. Write 150 to 160 characters. Lead with the primary keyword naturally, state the specific benefit, and end with an implicit reason to click rather than a command.

The common failure is a description that would suit any page on the site. If you could paste it onto three other pages without noticing, it is not doing its job. Count the characters rather than estimating. A description truncated mid-word looks careless in exactly the moment you need to look competent.

Headings Inside the Page

Every H2 is a small headline and most are written as labels. “Benefits” tells a reader nothing; “What Changes in the First Ninety Days” tells them whether to read on. Phrase headings as questions where the section answers a question, because that matches how people search and how retrieval systems index.

Keep them honest. A heading that promises more than the section delivers is the fastest way to lose a reader who was otherwise ready to trust you.

Writing Body Copy That Holds Attention

Ranking gets the reader to the page. The body copy decides whether they are still there when the argument for choosing you arrives.

Rhythm

Uniform sentences send people to sleep regardless of the quality of the thinking. Vary the length deliberately. A short sentence lands. A longer one, with a subordinate clause and room to develop the idea, gives the reader somewhere to settle before the next point arrives. Then a short one again.

The same applies to paragraphs. Two lines, then four, then one. Uniform blocks of identical length read as generated, because that is what generated text tends to look like. Read a section aloud. Anywhere you run out of breath or trail off bored is a place to rewrite. Our older piece on narrative in content writing covers rhythm in more depth.

Specificity

The single biggest quality difference between competent and forgettable writing is how specific it is. “Improve your site speed” is advice nobody can act on. “Convert the hero image to WebP and remove the lazy-load attribute” is. “Many businesses struggle with this” is filler. “We reviewed a service page with 624,515 impressions and 23 clicks” is evidence.

Every time you write a general claim, ask whether you can replace it with the specific instance it came from. Usually you can, and the page gets better every time you do.

Validate Before You Prescribe

Advice that arrives without acknowledging the difficulty reads as lecturing, and readers who feel lectured leave. Say what makes the thing hard first. “Most businesses do not have the traffic for valid A/B testing” earns the right to then explain what to do instead.

This matters more in commercial writing than people expect, because the reader is usually someone who has already tried and failed at the thing you are describing. Writing that implies it was easy tells them you have not done it either.

Cut the Padding

Transitional phrases that add nothing. Sentences restating the previous sentence. Paragraphs summarising what the reader just read. Introductions to sections that announce what the section will cover before covering it. The heading already did that. Hedging stacked three deep. “It could potentially be argued that this may sometimes help” means nothing. Decide what you think and write it.

Anything you wrote to reach a word count. Readers can tell, and so increasingly can the systems ranking the page.

Three Rewrites Worth Studying

Vague to specific. Before: “Make sure your images are optimised for the web.” After: “Serve images as WebP, size them to the largest viewport that displays them, and lazy-load everything except the hero.” The first sentence describes a category of action. The second can be handed to a developer.

Claim to evidence. Before: “Thin service pages don’t perform well.” After: “We reviewed a 300-word service page carrying 624,515 impressions and 23 clicks across sixteen months, at an average position of 43.” The first asks the reader to take your word. The second shows them the thing.

Lecture to peer. Before: “You need to understand that testing requires significant traffic to be valid.” After: “Most UK business sites do not have the traffic for valid A/B testing, which is awkward because the advice to test is everywhere. Below about a thousand conversions a month, use sequential testing instead.”

The first tells the reader what they have failed to grasp. The second acknowledges the bind they are actually in and offers a way through it.

Voice, and Why Most Business Writing Has None

Most business content is written in a voice that belongs to nobody: careful, balanced, faintly corporate, and indistinguishable from every competitor. That voice exists because it is safe. Nobody gets in trouble for it. It also gives a reader no reason to prefer you, and it is precisely the register a language model produces by default, which means it is now the sound of content produced at zero cost.

A voice worth having takes positions. It says this approach is better than that one and explains why. It admits what it does not know. It uses the words the trade actually uses rather than the polished equivalents. The practical test is whether someone could remove your logo and attribute the piece to three competitors without anyone noticing. If they could, the writing has no voice.

Getting there is mostly permission rather than skill. Writers hedge because they are not sure they are allowed to commit, and the fix is agreeing at brief stage what the business is willing to say plainly. Our SEO copywriting briefs include the position the piece is taking, precisely so the draft does not arrive neutral.

Proving Experience in the Writing

Google’s guidance on helpful content keeps returning to experience, expertise, authoritativeness and trustworthiness. Of the four, experience is the one that is hardest to fake and easiest to demonstrate, and most business writing skips it entirely.

What Demonstrated Experience Looks Like

Numbers from your own work rather than from a study you found. “Across our client base, the most common LCP failure is a lazy-loaded hero image” is something only someone doing the work can write. Specific instances with enough detail to be checkable. Named clients where you have permission, anonymised but concrete where you do not.

The things that went wrong. Writing that only describes successes reads as marketing. Writing that says “we tried this on three accounts and it worked on one, and here is what distinguished it” reads as expertise. Opinions with reasoning attached. Taking a position that some readers will disagree with signals that a person wrote it and thought about it.

How to Get It Into the Writing

The material usually exists and lives in people who are not writing the page. The fix is process rather than talent. Before drafting, spend twenty minutes with whoever actually delivers the service. Ask what clients get wrong most often, what surprised them recently, and what they wish prospects understood before the first call.

Ask for a number. Any number. How many of these have you done, how long does it usually take, what proportion fail the first time. Ask what they would tell a friend not to bother with. The honest negative is the most credible thing you can publish and almost nobody does it.

The Trust Signals Around the Writing

Attribute the piece to a named author with a real biography and relevant credentials. An unattributed page claiming expertise is asking the reader to trust nobody in particular. Date the content and update it honestly. A page marked as updated that has not changed is worse than an old date, because readers check.

Cite primary sources for load-bearing claims, and link to them. Google’s own helpful content guidance is explicit that this is what separates useful pages from assembled ones. Say when you do not know. A page that flags an untraceable statistic rather than repeating it confidently earns more trust than one that never admits uncertainty.

The Interview That Produces the Evidence

Twenty minutes with the person delivering the service will give you more usable material than a day of desk research. These are the questions that work. What do clients get wrong most often before they come to us? This produces the misconception the piece should correct, and it is usually not the one the marketing team assumed.

What surprised you recently? This produces the detail nobody else has published, because it has not made it into anyone’s content yet. How many of these have you done, and how long does it take? Any number is better than none, and delivery people usually have them to hand.

What goes wrong, and how often? The honest failure rate is the most credible thing you can publish. What would you tell a friend not to bother with? This produces the negative recommendation that distinguishes a real opinion from marketing copy. What question do prospects ask on every single call? That is your FAQ, written by the market rather than guessed.

Record it, transcribe it, and pull the specifics into the brief before drafting. The material almost always exists inside the business and simply never reaches the page.

Using AI Tools Without Producing Scaled Content Abuse

Most content teams now use AI somewhere in the process. The question is not whether to, but where it helps and where it produces material that will damage the site.

What Google’s Position Actually Is

Google’s stated position is that it focuses on the quality of content rather than how it was produced. Using automation to generate content whose primary purpose is manipulating rankings violates the spam policies, and not all automated production is spam. The named risk is scaled content abuse: generating many pages without adding value. That is the behaviour the enforcement targets, rather than the use of a tool.

The practical reading is that a generated page indistinguishable from a hundred others is a liability, and a piece where AI did the drafting while a human supplied the evidence, judgement and editing is not. Google’s guidance on generative AI content is worth reading in full rather than through secondhand summaries, because the secondhand versions tend to be either alarmist or permissive beyond what it says.

Where AI Genuinely Helps

Research synthesis. Reading twenty competing pages and summarising what they all cover and what none of them does is tedious and a model is good at it. Structural drafting. Turning an agreed outline into a first draft that a human then rewrites saves real time, provided everyone understands the draft is raw material.

Variation at the edges. Meta descriptions, alt text, and headline options benefit from being generated in quantity and chosen by a person. Editing support. Asking a model to find the passages that hedge, repeat or fail to stand alone is a useful second pass.

Where It Reliably Fails

Anything requiring first-hand experience, which is exactly the thing that now differentiates a page. Specific factual claims. Models produce confident, plausible, wrong numbers, and a published wrong number is worse than no number. Voice. Generated prose has recognisable habits: uniform rhythm, contrastive framing, tidy triplets, and a reluctance to commit to a position. Readers notice even when they cannot name what they are noticing.

Judgement about what to leave out. A model asked for a comprehensive piece produces a comprehensive piece, and the editorial skill is deciding what does not belong.

A Workable Division of Labour

Human decides the intent, the angle and what the piece must prove. Human gathers the evidence and the experience. AI drafts against that brief if it helps. Human rewrites, adds the specifics, removes the padding and takes a position. Human checks every fact against a primary source. Nothing a model asserted gets published as fact without verification, however plausible it sounded.

A Failure Worth Learning From

A piece drafted with AI assistance came through our accuracy pass carrying a confident claim that Google had reduced the Largest Contentful Paint threshold to 2.0 seconds in March 2026. It was plausible. It was phrased with the certainty of documentation. Several SEO blogs say it, which is presumably where the pattern came from.

It is also false. Google’s published threshold remains 2.5 seconds, and checking took four minutes against the primary source. Had it shipped, the piece would have told UK businesses to build to a standard that does not exist, and anyone who checked would have concluded we did not know the subject.

The lesson is not that AI is unusable. It is that a model reproduces what is commonly written, and what is commonly written is frequently wrong in exactly the places where being wrong is most damaging. Every load-bearing number gets checked against a primary source. Not the first blog that states it confidently, and not a summary of the source, but the source.

Writing for AI Overviews and Assistants

Being cited inside a generated answer is becoming its own objective, and the writing choices that earn it are mostly the ones that make a page good anyway.

What Gets Cited

Clear, direct, self-contained statements that answer the question the assistant is trying to answer. Specific, verifiable detail. Named thresholds, dated announcements, concrete steps and real figures are easier to attribute and safer to cite than general advice. Content structured so that the relevant passage can be located without reading the whole page. Question-phrased headings and short paragraphs do this.

Sources the system already treats as reliable on the subject, which is a function of the site’s broader authority as much as the individual page.

Practical Changes to How You Write

Front-load every section. The first sentence under a heading should be the answer, not the setup. Define terms on first use, briefly, even when the audience probably knows them. A passage that assumes context does not survive being lifted. Repeat the subject rather than relying on pronouns across paragraph boundaries.

Include the qualifying detail inside the sentence that carries the claim. “LCP should be under 2.5 seconds at the 75th percentile” is citable; “it should be under 2.5 seconds” is not.

Accept the Trade-Off Honestly

Writing for extraction costs a little elegance. Restating the subject where a pronoun would flow better makes prose slightly heavier. That is a real cost and worth paying on informational content that stands to be cited. It is not worth paying on a persuasive service page, where the reader is a person and the flow matters more.

Decide which mode each page is in before writing rather than applying one standard everywhere. Our guides on Google AI Overviews and appearing in ChatGPT cover the mechanics in more detail.

Internal Linking as an Editorial Job

Internal links are usually added afterwards by someone checking a box. Treated as an editorial decision during writing, they do more for rankings than most technical work.

Why the Writer Should Own It

The writer is the only person who knows where a reader’s attention naturally turns towards a related question. A link placed there gets clicked. A link inserted afterwards to satisfy a quota does not. Links also tell Google what the target page is about, through the anchor text. The writer is best placed to choose anchor text that is both descriptive and natural in the sentence.

Doing it during drafting rather than afterwards also surfaces gaps. If you keep wanting to link to a page that does not exist, you have found your next piece.

The Rules Worth Following

Use descriptive anchor text that would tell someone what the target page covers if they read it out of context. Never “click here” or “read more”. Link where the reader would want to go, not where it is convenient to place a link. Mid-paragraph, at the moment the related question arises, beats a block of links at the end.

Do not link the same URL more than twice in a piece. Beyond that you are repeating yourself and diluting the signal. The same principle governs site architecture generally, which our SEO work addresses first. Link up to the commercial page the piece supports at least once, early enough that a reader who leaves halfway has seen it.

Never guess a URL. A broken internal link is worse than no link, and slugs are easy to get wrong from memory.

Choosing Targets Well

Prefer editorial pages over announcements. Company news and awards posts are legitimate pages and they are not evidence of expertise on a subject, so they make poor link targets from a guide. Prefer pages you want to rank over pages that already do. Authority flows along the link, and pointing it at a page that needs it is the whole point.

Check the target is still the best page on the subject. Sites accumulate several pages covering similar ground, and linking to the weaker one wastes the signal.

A Worked Linking Pass

Here is what the linking pass looks like on a finished draft, done properly rather than as a box-check. Read the piece and mark every point where a reader might reasonably think “I would like to know more about that”. Those are your link positions, and there are usually six to ten in a long guide.

For each, decide whether a page exists that genuinely answers it. If one does, link it with anchor text describing the target. If none does, note it: you have just found a brief for the next piece. Then check the commercial link. The piece should point at the service page it supports at least once, placed early enough that a reader who abandons halfway has still seen it.

Then check the targets. Are any of them announcements rather than editorial pages? Are any of them weaker than another page on the same subject? Is any URL guessed rather than copied? Finally count. On a 3,000-word guide, six to ten contextual links spread through the piece reads as helpful. Two links clustered at the end reads as compliance.

The whole pass takes fifteen minutes and it is the difference between links that get clicked and links that sit there.

Writing Service Pages and Blog Posts That Support Commerce

These are different jobs with different rules, and most sites write both the same way.

Service Pages

A service page is a sales document that has to rank. The commercial job comes first and the search job is a constraint on how you do it. Cover what the service is in plain terms, who it suits and who it does not, how it works step by step, how pricing is structured, what results look like with real examples, who delivers it, and what happens when someone gets in touch.

Answer the objections that actually stop people. Cost, lock-in, how long it takes, what happens if it does not work, and why they should not do it themselves. A page that avoids these leaves the reader to resolve them alone, which they do by leaving.

Length follows from covering that properly, which for a competitive UK commercial term is usually 1,500 to 2,500 words. Thin service pages are the most common reason a page collects impressions and no clicks.

Blog Posts

A blog post is an editorial piece that has to earn attention and pass the reader towards something commercial. Choose subjects from search data and from what your sales team gets asked, not from what is interesting to write. Every post should carry at least one contextual link to the service page it supports, placed where the reader would naturally want it.

Resist the temptation to sell throughout. A post that pitches in every section reads as an advert and gets abandoned. Earn the reader’s attention for the length of the piece, then make the offer once, clearly.

How the Two Work Together

The service page makes the commercial case and the posts provide the evidence that the business knows the subject. A service page with no supporting content is asserting expertise. A service page behind fifteen genuinely useful pieces is demonstrating it, and Google can see the difference in the link structure as well as the content.

Our lead generation pages work this way, and the posts that support them are planned alongside them rather than afterwards.

Handling the Objections Directly

Most service pages avoid the four questions that actually decide the sale. Addressing them in the copy converts better than any amount of polish. On cost, say how pricing is structured even when you cannot publish a number. “Retained monthly, scoped to the number of priority pages” tells a buyer whether to keep reading. Silence makes them assume the worst.

On duration, give the honest range and the reason. “Two to three quarters on competitive commercial terms, because that is how long it takes for a new page to reach its position” is more persuasive than an optimistic figure, because a buyer who has done this before knows the optimistic figure is untrue.

On risk, describe what happens if it does not work. A named review point at ninety days with an agreed measure is worth more than a guarantee nobody believes. On doing it themselves, say when they should. “If your architecture is sound and your content is thin, hire a writer rather than an agency” costs you the wrong client and earns the trust of the right one.

A page that handles objections honestly is doing the job a salesperson would do on a call, which is what a service page is for.

Blog Formats That Reliably Work

Not all editorial formats earn their keep. These four do, and they map to different points in a buyer’s thinking. The honest buyer’s guide helps someone choose between options including ones you do not sell. It ranks for commercial investigation terms, it builds more trust than any amount of self-description, and most competitors will not write it because it means naming circumstances where they are the wrong choice.

The process walkthrough shows how something is actually done, step by step, with the decisions and the things that go wrong. It demonstrates competence more convincingly than a case study and it ranks for how-to intent. The diagnostic piece helps a reader work out what is wrong. It performs unusually well because the reader arrives with a symptom rather than a solution, which is how most people actually search.

The benchmark or original data piece publishes something only you know, from your own client work. It earns links and citations, which is the hardest thing to buy. What reliably fails is the news-reaction post published three days after everyone else, and the definitional explainer on a term an AI Overview now answers above your result.

Writing for the Person Who Has Been Burned

A large share of buyers for any marketing service have bought it before and been disappointed. Writing that ignores this reads as naive. They are looking for the tell that you are the same as the last one. Vague promises, undated case studies, results quoted without a baseline, and a refusal to name a timescale all read as warning signs to someone with a bad experience behind them.

Writing that lands with this reader does the opposite of reassurance. It names the failure modes. “Most programmes get cancelled in month four because the reporting shows cost before it shows return” tells them you have watched it happen. It gives numbers with their conditions attached. “Position 43 to 19 in twelve weeks on five pages, with clicks following the quarter after” is checkable and modest. “300% increase” is neither.

It also states plainly where you are not the right answer. A page that only describes fit reads as a pitch. A page that describes misfit reads as advice, and advice is what a burned buyer is actually shopping for. Credibility now comes from precision and admitted limits rather than from confidence. That is a change in what persuasive writing looks like, and most service copy has not caught up with it.

Editing: The Pass That Makes the Difference

Most writing improves more in editing than in drafting, and most content operations skip it because the draft looks finished.

The Passes Worth Running Separately

Trying to fix everything in one read means fixing nothing properly. Run these as distinct passes. The structure pass asks whether the sections are in the right order, whether any is missing, and whether any exists only because it was in the outline. Move and cut before polishing.

The accuracy pass checks every number, date, name and claim against a primary source. This is the pass that protects you, and it is the one most often skipped. The specificity pass hunts general statements and replaces them with the concrete instance behind them. The rhythm pass is read aloud. It catches uniform sentences, repeated openers and paragraphs that have gone on too long.

The cut pass removes everything that does not earn its place. If the piece is not shorter after this pass, you did not do it.

What to Look For Specifically

Three consecutive sentences starting with the same word, which reads as mechanical even when the content is good. Paragraphs over about eighty words, which fail on a phone regardless of quality. Hedged claims. Decide what you think. Sections that restate an earlier section in different words, which is the commonest form of padding and the hardest to see in your own work.

Any sentence you could delete without losing meaning. Delete it.

A Useful Test

Ask whether a senior colleague would be comfortable sending this to a prospective client as evidence of what the business knows. If the answer is no, find the specific passage causing the hesitation. It is usually either a claim that cannot be supported or a section that says nothing.

That test is more reliable than any checklist, because it forces a judgement about whether the piece is actually good rather than whether it complies.

A Worked Edit

Here is a paragraph going through the passes, which shows what each one actually removes. Draft: “It is worth noting that there are a number of different factors which can potentially influence how quickly your content will begin to rank in search engines, and these can vary quite considerably depending on a range of different circumstances that are specific to your particular situation and your website.”

Fifty-two words carrying no information. The structure pass asks whether the paragraph belongs at all, and it does, because readers genuinely want to know. The specificity pass replaces the abstraction with the actual factors: “Three things decide how quickly a page ranks: your site’s existing authority, how strong the incumbent pages are, and whether internal links from established pages point at the new piece.”

The accuracy pass checks that those three are defensible and that nothing has been overstated. The rhythm pass reads it aloud and finds it acceptable, though the sentence is doing a lot of work, so it splits: “Three things decide how quickly a page ranks. Your site’s existing authority, how strong the incumbent pages are, and whether internal links from established pages point at the new piece.”

The cut pass finds nothing further to remove. Fifty-two words became thirty, and the thirty say something.

Measuring Whether the Writing Worked

Content reporting tends to measure activity rather than outcome, which is why content budgets are fragile.

The Metrics That Tell You Something

Impressions for the target query tell you whether Google considers the page relevant at all. This moves first, usually within weeks. Average position tells you how far there is to go. Movement from 40 to 22 produces almost no extra clicks and is genuine progress, and reporting it prevents the work being written off at the halfway point.

Click-through rate at a given position tells you whether the title and description are doing their job. This is the fastest thing to fix when it is wrong. Scroll depth and time on page tell you whether the body copy holds people, which is the only real measure of whether the writing is any good.

Assisted conversions tell you whether the piece contributes to enquiries, which most informational content does indirectly rather than directly.

The Metrics That Mislead

Total pageviews across the blog. A rising number driven by one viral post that attracts nobody who buys is not progress. Bounce rate on informational pages. A reader who arrives, gets the answer and leaves has been served well. Word count published per month, which measures effort rather than result and quietly incentivises padding.

Rankings for terms nobody commercially relevant searches. The measurement discipline here mirrors the one in our lead generation reporting.

Setting Expectations Before You Start

A new piece on a competitive term takes two to three quarters to reach its position, and saying so in writing at the outset prevents the conversation in month two. Report leading indicators monthly and lagging indicators quarterly, and show the baseline every time so the comparison is visible.

Be explicit about what has not moved. Reports that only carry good news stop being believed at exactly the moment you need them to be.

A Worked Report

Here is how a content report should read at the ninety-day mark. Baseline at publication: the target term returned 4,100 impressions a month to the site, all of them to an unrelated page, with no page ranking in the top 50 for the exact query.

Week twelve: the new piece holds position 18 for the primary term, has picked up 340 secondary query impressions a month, and has earned 61 clicks in the month. Scroll depth shows a median of 64%, which for a 3,000-word guide means readers are getting past the halfway point in reasonable numbers.

Two assisted conversions are attributed to sessions that began on the page. The honest framing: the piece is progressing normally and has not yet reached its position. Expect the commercial return in the following quarter, not this one. Reported against the baseline, that reads as progress. Reported as “two enquiries”, it reads as failure, which is how good content programmes get cancelled at the point they are about to start working.

When a Piece Underperforms

A piece that has not moved after twelve weeks needs diagnosis rather than replacement, and the diagnosis is usually quick.

If impressions are near zero, Google has not understood what the page is about or has not indexed it. Check indexation in the URL Inspection tool, then check whether the primary term appears in the title, the H1 and the opening paragraph. A page that never states its subject plainly will not be matched to it.

If impressions are healthy and position is beyond 30, the page is relevant but outgunned. Compare it honestly against the three pages ranking above it. Usually they cover something you left out, or they sit on a site with more authority on the subject, or they have internal links you have not built.

If position is inside 15 and clicks are low, the ranking is working and the result is not being chosen. Rewrite the title and the meta description first, because that is a twenty-minute job with a measurable outcome.

If clicks arrive and scroll depth is shallow, the opening is not delivering on the title’s promise. That is a mismatch between what you advertised and what the page does, and it is fixed at the top of the page rather than throughout.

Resist rewriting the whole piece before you know which of these it is. Wholesale rewrites lose the signal the page has accumulated, and three of the four problems above are fixed in under an hour.

A Repeatable Production Process

Quality that depends on one talented writer having a good week is not a system. This is the sequence we use.

Before Writing

Agree the intent, the primary keyword and what the piece has to prove. One paragraph, written down, signed off. Read the pages currently ranking and note what they all cover and what none of them does. Interview whoever delivers the service for twenty minutes and come away with at least three specifics and one number.

Build the outline with every H2 and H3, a note on what each section proves, and the internal and external links planned in.

While Writing

Write the body sections in full before the introduction and conclusion, because you cannot introduce a piece you have not written. Write each section completely before moving on. Skipping ahead and returning produces sections that do not connect. Keep the brief visible and check each section against what the piece is supposed to prove.

After Writing

Run the editing passes separately, in order. Verify every factual claim against a primary source and record where each came from. Check the mechanical requirements: heading hierarchy, paragraph length, link count and placement, title and description character counts. Our SEO copywriting checklist covers the mechanical pass.

Have someone who did not write it read it. The author cannot see what is missing.

Why the Process Matters More Than the Writer

A good writer without this process produces inconsistent work, because the quality depends on how much they happened to know about the subject. An average writer with it produces reliably useful pages, because the specifics come from the interview, the structure comes from the brief and the errors get caught in the accuracy pass.

The process is what makes quality repeatable across a team, and repeatability is what a content programme actually needs.

Who Does What

The process only works when the roles are explicit, and on small teams one person wears several hats without noticing which one they are wearing. The strategist owns the brief: intent, primary keyword, what the piece must prove, and which commercial page it supports. This is a thirty-minute job and skipping it costs days later.

The subject expert supplies the evidence in a twenty-minute interview. They do not write and they are not asked to, because delivery people are usually poor writers and excellent sources. The writer owns the outline, the draft and the internal links. They are accountable for whether the piece is any good, which means they need the authority to change the brief when the research contradicts it.

The editor owns accuracy and the cut. Critically, this is not the writer. An author cannot see what is missing from their own piece, and the accuracy pass is exactly where self-review fails. On a team of two, the writer and editor roles can rotate between pieces. They cannot be collapsed into one person on the same piece.

Cadence

Two well-made pieces a month beats eight thin ones, and it is easier to sustain. Build in a review point at ninety days for each piece: check the position, decide whether it needs strengthening, and feed what you learn into the next brief.

Revisit the top-performing pieces annually. Updating a page that already ranks is usually cheaper and more productive than publishing a new one, and it is the most commonly neglected work in any content programme.

How GLO Approaches SEO Copywriting

We start with the Growth Strategy rather than the keyword list, because writing aimed at nobody in particular reads that way. That work establishes the brand archetypes, the tone of voice, the commercial objectives and the audiences most likely to buy. For each audience we build vertical profiles covering the job titles that sign off, the pain points driving the decision, the objections that slow it, and the lead magnets that earn an enquiry.

Those objections become the sections. That is the practical link between strategy and copy, and it is why our pages tend to address the awkward questions rather than avoid them. The keyword research then sits inside that picture rather than replacing it, and the brief for every piece states what it has to prove for which audience before anyone writes.

You can see how this connects to the wider digital marketing programme, and the SEO work that determines whether the writing has foundations to stand on.

Bringing It Together

SEO copywriting in 2026 rewards the things a model cannot produce: first-hand experience, specific evidence, and a willingness to take a position. The points worth carrying away:

• Intent determines format, length and structure, and no amount of good writing rescues the wrong format
• The first fifty words decide whether you are cited and whether the reader stays, so answer the question there
• Self-contained passages survive extraction, and passages built on pronouns do not
• Experience is the differentiator, and it comes from interviewing the people who do the work
• AI helps with research, drafting and variation, and never with facts, judgement or voice
• Editing in separate passes improves a piece more than redrafting it

If your service pages collect impressions and no clicks, the writing is usually the reason. Book a Growth Strategy session and we will look at what your highest-impression pages are failing to say.

Frequently Asked Questions

What Is SEO Copywriting?

SEO copywriting is writing web content that ranks in search and persuades the reader to act. It combines search intent analysis, keyword research and page structure with ordinary persuasive writing skill. The distinguishing feature is that the search requirements shape the brief before drafting begins, rather than being applied to finished copy as an afterthought.

In 2026 it also covers writing that can be extracted and cited by AI-generated answers, which rewards direct opening statements and self-contained paragraphs.

How Long Should an SEO Page Be?

Long enough to cover the subject properly and no longer, which depends entirely on intent rather than on any target. A competitive commercial service page usually needs 1,500 to 2,500 words to answer what a buyer asks. A comprehensive guide competing with established pages often needs considerably more. A page answering one narrow factual question may need 600.

The reliable method is to read the pages currently ranking, establish what the subject genuinely requires, and then cover it better. Writing to a word count produces padding, and padding is visible to readers and to ranking systems.

Can I Use AI to Write SEO Content?

Yes, within limits that matter. Google’s position is that it assesses content quality rather than production method, and that using automation to generate pages whose purpose is manipulating rankings breaches the spam policies. The practical division is that AI is useful for research synthesis, drafting against an agreed outline, generating variations and supporting the edit. It is unreliable for facts, incapable of first-hand experience, and recognisable in voice.

Every factual claim needs verifying against a primary source before publication, whoever or whatever drafted it.

How Long Does SEO Copywriting Take to Rank?

A new page on a low-competition long-tail term can rank within four to eight weeks. A competitive commercial term usually takes two to three quarters. The variables are your site’s existing authority, how strong the incumbent pages are, and whether the new piece is supported by internal links from pages that already carry weight.

Impressions and average position move first and clicks follow, so judging a piece on clicks at week six will tell you it failed when it is progressing normally.

What Is the Difference Between SEO Copywriting and Content Writing?

Content writing produces material for an audience without search performance as a primary requirement. It covers newsletters, social posts, brochures and thought leadership where distribution comes from somewhere other than search. SEO copywriting starts from what people are searching, what those searchers want, and what the page must do commercially, then writes to satisfy all three.

The skills overlap heavily. The difference is the brief, and specifically whether search demand and commercial intent shaped the piece before the first sentence was written.