SEO copywriting — four rules Google contradicts in its own words

Guides to SEO copywriting consist largely of numbers nobody can trace to a source. A title should be sixty characters. A description a hundred and sixty. The phrase should make up one to three per cent of the text. You should write "for position zero".

All four are contradicted in Google's documentation or do not appear in it at all. Below we quote where, and with what.

How this differs from writing in general

By one thing, not a list of techniques. The text has to answer a question you did not choose — a stranger asked it, in three words, typed into a search box in circumstances you know nothing about.

Everything that separates this craft from writing follows from that. It is why phrase research is part of writing rather than a marketing task. It is why the same service needs different texts depending on how people talk about it. And it is where the most common failure comes from: a good text that answers the question its author asked rather than the one the reader did.

A second consequence is less obvious: the same phrase is typed with different intent, and a text serves only one of them. "CRM system" is typed by somebody who does not know what one is; by somebody comparing three; and by somebody looking for the price of one in particular. Three different texts, three different endings. Writing one that tries to serve all three produces a text that serves none — and that is the most common reason a good text brings no enquiries.

Recognising intent needs no tool. Type the phrase into Google and look at what the engine has already decided counts as a good answer: all definitions mean a definitional question; all comparisons mean a comparative one; all offer pages mean a buying one, and an article will not get in. Five minutes that settle whether four thousand words are worth writing.

How to find the phrase in the first place and where topics come from: content marketing. Here we start from the point where you already know what you are writing about.

The title: what Google does with it

The first thing to accept: your title is a proposal, not a setting.

Google's documentation lists nine sources the engine may take the displayed title from — the <title> element is only one. The others include headings, among them <h1>, the content of og:title, any other text on the page, anchor text on the page, and the text of links pointing to the page from elsewhere. Google substitutes a title when, in its words, it has "detected an issue on the page".

The second thing demolishes the rule everyone repeats:

> There's no limit on how long a `<title>` element can be, the title link is truncated in Google Search results as needed, typically to fit the device width.

There is no length limit. Truncation happens "typically to fit the device width". The reader's device decides where the cut falls, not a character count. The sixty-character rule is industry folklore, convenient because it is easy to measure.

The practical conclusion is different and more useful. Since the end is cut and the end is movable, the words that distinguish your page belong at the front. A company name glued to the front of every title eats up that position in every result. Second: since <h1> is one of the substitution sources, it is worth having it say the same thing as the title — otherwise you are supplying Google with the alternative yourself.

The meta description: "sometimes"

The description is treated as a mandatory field Google fills in the results. It is not. It is a candidate.

> Snippets are primarily created from the page content itself. However, Google sometimes uses the meta description HTML element if it might give users a more accurate description of the page than content taken directly from the page.

The fragment under the title is built primarily from the page content. The meta description gets used when Google judges it a better fit. The same documentation page adds a sentence twinned with the one about titles: there is no length limit on a description either, and truncation again follows device width.

So why write one? For two reasons. On the queries where it is used, it is the only sentence in the search result whose content you control entirely. And Google recommends something at this point worth taking literally: descriptive, specific language rather than a list of keywords, and a separate description for every page.

There is a third, less obvious conclusion. Since on most queries the fragment will be cut from your own text, the text itself has to be answer-shaped. A paragraph opening with "in today's world, more and more companies" is unusable as a search fragment and unusable for anything else.

Text structure: order, not a template

One practical thing follows from the two sections above, more important than any template: the text has to begin with the answer.

Not with an introduction to the field, not with how important the subject is these days, not with the history of the question. There are three reasons and all are checkable. The reader arrived with a question and checks in two seconds whether they have landed right. Google cuts the results fragment from the page content, so the first paragraph is candidate number one. And an AI summary, if one appears, reaches for what can be summarised — and a paragraph beginning "more and more companies recognise" contains nothing quotable.

The order that follows:

1. The answer — one or two sentences. If the question is "what does it cost", the first paragraph holds a range or the reason there is none.
2. The qualification — what that answer depends on. This is where the text stops being a generality.
3. The detail — only here the cases, exceptions and numbers.
4. What to do — a specific next step, doable without you.

That inverts the schoolroom introduction–body–conclusion and feels unnatural to many writers. A test that helps: read the first paragraph alone and check whether somebody who stops there has taken away something true. If not, it is not a first paragraph, it is a run-up.

The second rule concerns what sits between sections: one section, one thought. A section that cannot be summarised in a sentence is two sections. That is not a search requirement, it is the condition for a text to be readable in jumps — and most people read in jumps.

Headings: a table of contents, not decoration

Headings have a role here that their appearance does not reveal: read on their own, without the text between them, they should tell the whole story. That is both a test of structural quality and the way most people read — scrolling and stopping at whatever looks like their question.

Hence the practical rulings:

• A heading carries content, not a category. "Introduction", "Summary" and "Our approach" say nothing. "What Google does with your title" says everything.
• One `<h1>` per page, saying the same as the title. The reason is concrete and follows from the section above: <h1> is one of the nine substitution sources. A mismatch hands the engine an alternative to your own proposal.
• Hierarchy without skips — h2 for sections, h3 for sub-points within them. Skipping from h2 to h4 to get a smaller heading is using markup as styling. Size changes in the presentation layer, not in the structure.
• A question as a heading works if it is a question somebody actually asks. Forcing every heading into a question produces a contents list that reads like an interrogation.

The same content at two addresses

The fear of duplicate content is inflated in this industry beyond what the documentation supports. Two things get confused constantly.

Duplicated content on your own site — the same text at two addresses, the www and non-www versions, a product in two categories — is not a penalty. It is a selection problem: Google has to decide which address to show. The documentation states that if you do not specify a canonical, Google "will identify which version of the URL is objectively the best version to show to users in Search". Specifying it yourself — by redirect, by rel="canonical" or through the sitemap — lets you consolidate the signals from the individual URLs into a single preferred one rather than scattering them.

Copying other people's content is an entirely different matter, and there the prohibition is explicit, in the spam policies, alongside scraping and rewriting other people's texts with synonyms.

Practically, for a writer: the problem is not repeating a phrase between your texts, it is repeating a subject. Two of your articles answering the same question compete for the same slot, and Google will pick one anyway. That is a situation for a merge, not for optimisation — and we had it six times over in one section.

Keyword density does not exist

Not in the sense that it is not worth watching. In the sense that Google nowhere states a percentage — no threshold, no range, no counting method.

What does exist has a different shape. The spam policies carry a category called keyword stuffing:

> Keyword stuffing refers to the practice of filling a web page with keywords or numbers in an attempt to manipulate rankings in Google Search results.

It is worth reading the examples Google supplies, because they settle the doubt better than the definition: lists of phone numbers without substantial added value, blocks of text listing cities and regions a page is trying to rank for. Those describe pages built out of keywords, not how often a phrase appears in a paragraph.

The rule that replaces counting: name the thing the way a customer names it, synonyms included, because those are what match the query. If you are counting occurrences you are solving the wrong problem — and if you had to count to reassure yourself you had not overdone it, you probably had.

Position zero: "You can't"

A whole branch of advice teaches how to format text to land in the featured snippet — the answer box above the results. To the question of how to mark a page as a featured snippet, Google's documentation answers in two words:

> You can't.

The explanation follows: Google's systems determine whether a page would make a good featured snippet for a given request, and if so, elevate it. There is no markup, no schema, no format that forces it.

What remains true from that advice? That writing in the shape of an answer is good for the reader — a definition in the first two sentences of a section, a list where the question is about a list, a table where the point is a comparison. If Google elevates it as well, good. But it is not a technique with a success rate, it is a way of writing that stands up without a search engine.

The only real lever here works the other way: you can opt out. More on that two sections down.

When Google answers instead of linking

This is the change that revalues everything above, and the thing the guides are silent about, because they are older than it.

Pew Research Center measured it on browsing data rather than survey self-reports. The result:

• 8% of visits to a results page with an AI summary end in a click on a result,
• 15% of visits without a summary — nearly twice as often,
• 1% of visits end in a click on a link inside the summary,
• the session ends on the results page in 26% of visits with a summary against 16% without.

Sample: 900 US adults, 68,879 searches, March 2025. We publish the authors' caveat with the number: the summary for any given query changes over time, and the result describes how results appeared when the data was collected. This is American data from one month — not European and not multi-year.

The 1% is the most important figure and the least quoted. An AI summary carries links to sources, but they are clicked in one visit in a hundred. Appearing in a summary is therefore not a substitute for a position — it is something of a different kind.

From Google's side the mechanism is query fan-out: one user question is broken into multiple related searches across subtopics and sources. So you are no longer answering the query, but potentially one of a dozen side questions nobody typed.

The conclusion for a writer is hard. Traffic from questions a summary settles in two sentences is structurally gone — "how long does SEO take", "what is a CMS", "what image formats are there". No title fix recovers it, because the problem is not position, it is that the question was closed before the list of results.

What survives is what a summary cannot do: your price, your case, your availability, your judgement on a specific situation and your view on when not to do something. A text built purely from general knowledge of a field was always weak; it has now stopped being visible as well.

Which of your texts are exposed

You can estimate this without any tool, in fifteen minutes. Go through your own list of texts and answer one question about each: can this be answered correctly in two sentences, knowing nothing about the person asking?

If yes — the text is exposed. "What is a CMS", "what image formats are there", "how does http differ from https". A summary will answer and will be right.

If no — it is safe, for a specific reason: the answer depends on something the model does not have. What a website costs in my industry. Whether it is worth it in my case. What went wrong for somebody who did this before me. Here the answer needs somebody else's experience, not the sum of what has already been written online.

This is not an argument for abandoning definitional texts — they have their place in a site's structure and in a sales conversation. It is an argument for no longer measuring their success by search traffic, because that traffic will keep shrinking however well they are written.

What you genuinely do not have to do for AI

An offer has grown up around the above — AI optimisation, GEO, files for models, special markup. Google's documentation settles it in two sentences:

> You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add.

>

> There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.

If somebody offers you optimisation for AI summaries as a separate service with a separate price, those are the two sentences to put on the table. It does not mean there is nothing to do — it means the thing to do is what it always was, and that there is no shortcut available for purchase.

Four controls that genuinely exist

Everything above is out of your hands. Here is what is in them — and note that all four work by switching off, not on.

• `nosnippet` — no text fragment from this page in results at all. It also excludes the page from featured snippets.
• `data-nosnippet` — excludes one part of the page, not the whole. An attribute on an HTML element. The right tool when you do not want one specific thing readable from the results — a price table, say, or a quotation that reads differently out of context.
• `max-snippet:[n]` — caps the fragment at n characters. A low value usually rules out a featured snippet but, as the documentation notes, does not guarantee it.
• `noindex` — out of the index entirely. Right for pages that must exist and must not appear: thank-you pages after a form, internal search results, drafts.

The cost is obvious and should be counted before use: less fragment in the results is also less reason to click. Restricting the fragment holds up where the content is the product, or where out of context it misleads. As a reflex defence against AI it does not hold up, because it cuts you out of ordinary results too.

Anchor text is an editorial decision

Anchor text — the words the reader clicks — does three things at once, and authors overlook two of them.

It tells the reader what is on the other side. It tells Google what the destination page is about. And, as we saw in the section on titles, it is one of the nine sources Google may take the destination page's title from — including anchor text of links pointing at it from elsewhere. A page linked everywhere as "offer" may appear in results as "offer".

Hence two rules, both free. Descriptive anchors rather than "click here" — the name of the subject, not an instruction. And the same anchor for the same destination across the site, because scattered naming scatters the signal.

Link architecture — what to connect to what, how not to leave pages with no link at all — is a separate subject, and we cover it with our own measurement under website positioning.

Text length: there is no number

There is no number. No Google document states a minimum or an optimum, and every circulating threshold — a thousand words, fifteen hundred, two thousand — comes from correlations in somebody's dataset, not from guidance.

Length is set by two things: how much the question actually requires and what the answers already out there contain. A text about setting up a redirect should be three hundred words, because that is how long the answer is. A text about choosing a system for a business should run to several thousand, because without them it is a list of names.

We have our own counter-example and it is not flattering. The Polish-language edition of the section you are reading held six documents, over twenty-one thousand words and about twelve hundred searches a month across every phrase they declared. Length was never the problem there.

A test that works better than a counter: cut a paragraph and check whether the reader lost anything. If not, it was there for the volume.

Our texts rank for something other than intended

You do not get to choose the query. We will show it on ourselves, because that is the most honest form of the argument.

We checked how far the query actually carrying each text sits from the phrase it was written for. Four cases:

• a text on content marketing — we wanted "content marketing", we get "content publishing strategy";
• a text on SSL certificates — we wanted "ssl certificate", we get "ssl vs https";
• a text on backups — we wanted "website backup", we get "backup solution";
• the Google Business Profile section — we wanted "google business listing", we get "company in google".

The caveat in the same breath: this is measurable on four documents because only twenty-three of our hundred and sixty-five articles have any impressions at all. That is not a representative sample — it is everything we have, and that fact is itself part of the lesson.

What to do about it: the gap between the intended phrase and the real one is a free rewrite brief. If Google shows you for a phrase you did not choose, and that phrase has demand, the honest response is to write the text that is genuinely about it — rather than pushing for years on the one chosen at a desk.

What this text deliberately leaves out

• How to find a phrase and how to read the report — website positioning.
• Where topics come from, the calendar, who should write — content marketing.
• Alt text and image file names — image copyright.
• Whether a business needs a blog — business blog.
• Product and category descriptions in a shop — those follow separate rules and we cover them in the ecommerce section.