Content SEO
Content SEO: writing content that ranks and gets cited by AI.
The same qualities decide whether a page ranks, whether an engine lifts an answer from it, and whether an assistant is willing to repeat it: it answers the real question, it shows where its facts came from, and it is still correct a year later.
How do I write content that ranks in Google and gets cited by AI?
Content that ranks and gets cited is built around what the searcher is actually trying to accomplish, states its answer at the top of each section, shows where its facts came from, and is kept accurate as things change.
The structural rules are unglamorous: real questions as headings, a complete answer in the first two or three sentences underneath, every number attached to a named publisher and a year, and structured data that matches the visible text exactly.
Volume is not a lever. One page that fully answers a question outperforms five that partially answer the same one, and near-duplicate pages mostly compete with each other.
What Google means by helpful content
Google publishes a list of self-assessment questions for content quality, and they are more useful than most SEO checklists because they are written from the reader's point of view rather than the optimiser's.
The questions are blunt. Does the content provide original information, reporting, research or analysis? Does it provide a substantial, complete or comprehensive description of the topic? Would someone leave feeling they have learned enough about a topic to help achieve their goal? Does it leave a reader feeling they need to search again to get better information from another source?
That last one is the standard worth internalising. A page fails when the reader finishes it and searches again. Everything else on this page is a technique for making sure they do not need to.
There is no separate quality bar for AI. Google's guidance for its AI features points back at the same fundamentals, and an assistant asked to summarise a shallow page produces a shallow summary that names someone else. Depth, accuracy and clarity are the requirement for all three outcomes: the ranking, the lifted answer, and the recommendation.
From a question to a published page
This is the sequence we use, and the first two steps are the ones most often skipped and most often responsible for a page that never performs.
Decide what the reader is trying to accomplish
Before any outline, write one sentence describing the job the reader came to do. Someone searching "root canal cost" wants a number and the reason it varies. Someone searching "root canal or extraction" wants a decision framework. Someone searching "emergency dentist near me" wants a phone number and opening hours. Three very different pages, and building the wrong one is a total loss no matter how well it is written.
- Look at what currently ranks: the format Google is showing is evidence about intent
- Separate research questions from booking questions and build different pages for them
- Write the job down at the top of the brief so the draft can be judged against it
Check you are not already competing with yourself
Before creating a page, search your own site for the topic. Most sites that struggle with content do not have too little of it, they have four overlapping pages splitting the same signals. Consolidating three thin pages into one strong page and redirecting the others is often the highest-return content task available.
Outline as questions, not as topics
Each heading should be a question a real person would ask, phrased the way they would ask it. This forces the page to answer rather than describe, and it lines the page up with how snippets, People Also Ask panels and assistants work. Search Console's query report is the cheapest source of real phrasings, because it shows what you already appear for.
See the remaining steps: From a question to a published pageHide the remaining steps: From a question to a published page
Answer first, then explain
Under each heading, resolve the question in the first two or three sentences, then give the supporting detail. If the honest answer is that it depends, say what it depends on in the same breath. This costs nothing in reading quality and it is the difference between a section that can be lifted and one that cannot.
- No warm up paragraph and no restating of the question
- Make each answer true when read alone, with no pronouns pointing at earlier text
- Keep the depth underneath: the short answer is not the whole page
Attach the evidence in the visible sentence
Name the publisher and the year where the fact came from, and link the original document rather than a blog that summarised it. This is good practice for readers, it is what makes a passage safe for a machine to reuse, and it is the single fastest way to look more credible than the pages you are competing with, most of which cite nobody.
- One named source beats three vague appeals to what studies show
- State the year whenever a figure is more than about eighteen months old
- If you cannot find a source, make the point qualitatively instead of inventing a number
Add the structured data last, and keep it honest
Article, FAQPage where there is a genuine FAQ, Breadcrumb, and the business types that apply. Markup goes on after the content is final, so it describes what is actually there. Do not expect FAQ markup to produce a rich result: Google restricts those to a narrow set of sites now, so the real benefit is an unambiguous machine reading.
Link it into the site properly
A new page with no internal links pointing at it is hard to find and easy to ignore. Link down from the relevant hub page, across from two or three genuinely related pages, and out to the service page if there is a commercial next step. Use descriptive link text that says where it goes, never a bare read more.
The whole sequence works on existing pages too, and usually pays back faster there. A page that already has impressions is a page with a proven audience and a fixable execution problem.
Why one deep page beats five thin ones
The most common structural mistake is publishing a page per keyword variant instead of a page per question.
Five pages targeting near-identical phrases split every signal you have: internal links, external links, engagement and crawl attention. They also confuse a search engine trying to decide which one to show, which frequently results in the weakest of the five being chosen or none of them performing.
One comprehensive page on the subject, supported by separate pages for the questions that genuinely deserve their own answer, concentrates those signals instead. The test for whether a subtopic deserves its own page is simple: could you write eight hundred honest words about it that would not fit naturally inside the main page?
This structure also happens to be what an assistant handles well. A clear hub with unambiguous children is easier to interpret than a flat sprawl of overlapping pages, and the internal links tell a machine which page you consider authoritative on what.
What happens to the click when a summary appears
Pew Research Center studied the browsing data of 900 US adults across 68,879 Google queries in March 2025, of which 12,593 produced an AI summary. This is independent research rather than a vendor study, which makes it the most reliable measurement available of this behaviour.
SourcePew Research Center, browsing-panel study of Google AI summaries, 2025
That last figure is the one that changes how content should be written. If almost nobody clicks the citation, then the value of being in the answer is the answer itself carrying your name and your framing. Content written purely as click bait for a summary that will never be clicked is wasted effort.
The pre-publish checklist
Ten minutes with this list catches most of what would otherwise be found six months later.
- Every heading is a question someone would actually ask, in their words
- Each section answers its heading in the first two or three sentences
- Every number has a named publisher and, where it matters, a year
- Every external link points at the primary document rather than a summary of it
- Nothing on the page promises an outcome that cannot be guaranteed
See the full checklist: The pre-publish checklistHide the full checklist: The pre-publish checklist
- The page does not duplicate an existing page on the same site
- Internal links point in from a hub and out to the logical next step, with descriptive link text
- Images have meaningful alt text and the page reads correctly with styles disabled
- The title and meta description are unique across the whole site and describe this page specifically
- There is a date in the calendar for reviewing whatever on this page will age
The last two are the ones that get skipped. Duplicate meta descriptions across a site are extremely common, and a page with no review date becomes a liability the moment a price or a regulation changes.
What makes a passage safe to quote
Both a human editor and a machine are making the same judgement: can this be repeated without it becoming wrong?
A quotable passage names its subject rather than relying on the heading, contains its own qualifiers rather than putting them in a footnote, and attributes any claim that is not self-evident. It does not say "research shows", it says who, when and how many.
It also does not overreach. A sentence that claims a result nobody can guarantee is unsafe to repeat, which is a practical reason to write honestly rather than merely an ethical one. Hedged language such as "typically" and "designed to" is not weakness here, it is precision.
Finally it stays true. The fastest way to lose a citation you have earned is to let the page go stale while the world moves on. Put the pages that carry prices, thresholds, product names or regulations on a review cycle and check them on a schedule rather than when someone complains.
Habits that quietly sink a content program
Publishing to a schedule rather than to a question. A monthly post quota produces pages nobody searched for, and those pages still consume crawl budget, internal link equity and editorial attention. Consistency matters, but consistency of standard beats consistency of frequency.
Generating bulk text and shipping it unedited. Google's spam policies target content produced at scale primarily to manipulate rankings rather than automation itself, so the risk is not the tool, it is publishing material nobody checked, nobody sourced and nobody would defend. If no human is accountable for a page being correct, it should not be live.
Read the full breakdown: Habits that quietly sink a content programHide the full breakdown: Habits that quietly sink a content program
Writing for a word count. Length is a consequence of covering a question properly, not a target. Padding a four hundred word answer to two thousand words makes the answer harder to find, which is the opposite of what both readers and extraction need.
Chasing keywords the business cannot serve. Ranking for a treatment you do not offer, or a city you do not cover, produces traffic that cannot convert and inquiries your front desk has to turn away. Relevance to the business is part of content quality, not separate from it.
Never updating. Content decays quietly. A page that was excellent in 2024 can be actively misleading now, and a misleading page damages the trust that everything else here depends on.
What to read next
The neighbouring pieces of the same job.
Answer engine optimization (AEO)
The formatting discipline in detail: how to write a section an engine can lift, and what stops a good page being quoted.
Answer engine optimization (AEO)Do I need a blog for my business?
Whether publishing is worth the time for your business, and what to do instead if the honest answer is no.
Do I need a blog for my business?What are keywords and how do I choose them?
How to pick the questions worth writing about before committing a month of effort to the wrong ones.
What are keywords and how do I choose them?What is a meta description?
Why Google rewrites them, what they are actually for, and how to write ones that earn the click.
What is a meta description?Want an honest read on the content you already have?
We will tell you which pages are worth improving, which should be merged, and which are quietly working against the rest of the site.
Straight answers.
How long should a page be?
Long enough to answer the question completely and no longer. There is no word count that ranks, and padding a short answer makes it harder for both readers and engines to find the part that matters.
A useful test: could a knowledgeable person read this and still need to search again? If not, it is long enough. If yes, the missing part is what to add, not another five hundred words of context.
How often should I update content?
Set the cycle by what ages, not by the calendar. Pages carrying prices, thresholds, regulations, product names or statistics need a scheduled review, typically quarterly. Explainer pages about things that do not change can go much longer.
Changing a date without changing the content does nothing. Update because something is now wrong or incomplete, and say what changed.
Does AI-written content get penalised by Google?
Google's spam policies target content created at scale primarily to manipulate search rankings, rather than the use of automation in itself. The judgement is about the output: is it original, accurate, useful and accountable to someone.
In practice, unedited generated text tends to fail on exactly the qualities that matter here. It rarely cites primary sources, it often repeats claims that cannot be verified, and nobody on the team can defend it. Treat any draft as a draft, whoever or whatever produced it.
Should one page cover several related searches?
One question, and whatever phrasings of that question naturally belong to it. Trying to cover several distinct questions on one page usually means none of them is answered in a way that can be lifted.
If two phrasings mean genuinely different things to the person searching, they need different pages. If they mean the same thing, they belong on one.
Should I write for people or for search engines?
For people, in a structure machines can parse. Those are not in conflict. Answer-first writing, honest headings and clear sourcing make a page easier to read and easier to extract from at the same time.
Where they do conflict, the tell is usually keyword repetition. If a sentence reads awkwardly because a phrase has been forced into it, it is hurting both audiences.
How do I get my content cited by AI?
Write answers that are complete, self-contained and sourced, make the business behind them unambiguous, and keep the facts current. Those are the qualities that make a passage safe to repeat.
There is no guarantee. No vendor publishes how its assistant selects sources, and the same prompt can return different names on different days, so treat citations as an outcome you improve the odds of rather than one you can buy.
Where this comes from.
Primary documentation and published research behind the guidance on this page.
- Google Search Central: Creating helpful, reliable, people-first content (opens in a new tab)Google's published self-assessment questions for content quality and experience.
- Google Search Central: Control your snippets in search results (opens in a new tab)Google's own statement that it may rewrite a meta description.
- Google Search Central: Introduction to structured data markup (opens in a new tab)
- Google Search Central: Spam policies for Google web search (opens in a new tab)Includes the link spam policy that covers bought and exchanged links.
Talk to the team
A short call, a look at how the business currently shows up, and a straight answer on what we would do first.