Key takeaways
- SEO-optimised content answers a specific search intent in a format Google and answer engines can extract.
- GEO depends on SEO, though not at the ranking level: at the level of indexability and topical coverage.
- Keyword density, the 1,000-word threshold and systematic keyword placement no longer work.
- Open every section with a self-contained 30 to 80 word answer, because that is the format answer engines lift.
Writing SEO-optimised content is no longer about placing a keyword in the right spot. You write for two readers: the human searcher, and the machine that decides whether to cite you.
What Counts as SEO-Optimised Content in 2026?
SEO-optimised content answers an identified search intent, in a format engines can understand, extract and cite. Three conditions, in that order.
The first condition has not moved in ten years. Before writing a single line, you need to know the precise question your reader is asking and where they sit in their decision process. Keyword research gives you that information. Without it, you produce text nobody is looking for.
The second condition has changed in nature. Google still assesses how relevant a page is to a query, but it no longer reads your article as a single block. It analyses it passage by passage and selects the ones that answer best. Answer engines work the same way, only more aggressively: they almost never cite a whole article, they lift extracts of a few dozen words.
The third condition is new. Now that producing text is free and unlimited, volume no longer separates anyone from anyone. What sets you apart is what only you can write: a figure you measured, a case you handled, a method you have applied for years.
I sum the logic up in one sentence with my clients: you do not optimise twice, you optimise for SEO and then make the result extractable.
Read also: Outsourcing SEO: should you delegate in 2026
Why Does GEO Depend on SEO?
An answer engine can only cite what it has crawled, indexed and judged relevant. Every filter upstream of the citation is an SEO filter. The only link in the chain that genuinely belongs to GEO is the extraction format.
The chain breaks down into five stages: your page is crawled, indexed, selected to answer a query, split into passages, then one passage gets cited. The first three belong to classic organic search. A site blocked at crawl level, poorly linked or orphaned never clears the first hurdle, however good its writing.
The SEO → GEO dependency chain
Crawl → Indexing → Selection → Passage splitting → Citation
The first four stages are SEO. The last one depends on how you write.
What the 2026 Data Actually Shows
A qualification belongs here, because an oversimplified claim would be wrong. The relationship between organic ranking and AI citation has loosened considerably. The Ahrefs analysis published in March 2026, covering 863,000 queries and 4 million cited URLs, found that 38% of pages cited in AI Overviews also appear in the top 10 for the same query. That figure stood at 76% in July 2025. The remainder splits almost evenly between positions 11 to 100 and pages absent from the top 100 entirely.
Put plainly: ranking first no longer guarantees a citation, and not ranking first no longer rules one out. This shift does not break the link between SEO and GEO, it moves where that link applies.
The explanation lies in query fan-out, a technique Google has confirmed: the initial query is split into several sub-queries, and the pages that surface most often across those derived SERPs are the ones feeding the answer. The perimeter widens rather than disappearing. You no longer rank for a query, you rank for the constellation of questions surrounding it.
That is SEO. Just harder.
My Take on the GEO Services Being Sold Today
In my view, a good share of the GEO services on the market today are repackaged SEO audits. The work performed is often sound; the problem sits in the invoicing. What gets presented as a new discipline remains, roughly 80% of the time, organic search done properly. An honest provider will tell you that if your site is not indexed correctly, no amount of GEO optimisation will produce results.
The 3 SEO Writing Rules I Removed From My Method
I used to recommend these three rules. They were widespread, I taught them, and they still appear in most guides published today. They no longer produce results.
The three abandoned rules are keyword density, the 1,000-word minimum threshold, and systematic keyword placement in so-called hot spots. I replaced them with semantic field coverage, length matched to the topic, and care given to the title tag and H1 only.
| Abandoned rule | What I used to recommend | Why I stopped | What I do instead |
|---|---|---|---|
| Keyword density | Repeat the keyword 10 times per 1,000 words | Google interprets meaning, not frequency. Mechanical repetition degrades readability without improving rankings. | I cover the topic’s semantic field: entities, synonyms, related questions. |
| The 1,000-word threshold | Write at least 1,000 words per page | The correlation between length and ranking came from the effort invested, not the word count. The Ahrefs study of 75,000 brands found almost no relationship between a site’s content volume and its AI visibility. | I write the length the topic requires, then cut. |
| Hot spot placement | Place the keyword in the title tag, URL, H1, first paragraph, two headings, alt attribute | Everyone has applied this checklist for a decade. It differentiates nothing. | I focus on the title tag and H1, and let the rest follow the subject naturally. |
What survived the cull: search intent, content structure and internal linking. All three still serve the reader, Google and answer engines alike.
In my view, these rules never worked for the reasons attributed to them. They correlated with well-researched content without causing it. A writer who reached 1,500 words had dug into the topic, and that was what Google rewarded.
How Do You Structure Content So It Gets Read, Ranked and Cited?
Structure is the only piece of work that serves all three of your readers at once. A clear plan helps the human find their answer, helps Google grasp your topical coverage, and helps answer engines isolate a citable passage.
The Inverted Pyramid: Answer First
Put the answer at the top and the explanation underneath. The technique comes from journalism and works for three reasons: the hurried reader gets what they came for, Google identifies the section’s subject immediately, and the answer engine has a self-sufficient extract to lift.
Apply it at two levels. Across the article: most-searched questions at the top, technical detail below. Within each section: one direct statement, then the explanation.
Atomic Content: Every Section Must Stand Alone
Answer engines lift short passages, roughly 30 to 80 words, and fold them into a composite answer. A passage that opens with “as we saw above” or “in this context” becomes unusable the moment it leaves the page.
The markers of a genuinely self-contained section:
- It opens with a statement that answers the question posed by its heading.
- It contains no implicit reference to an earlier section.
- It names its subject explicitly rather than using a pronoun.
- It stays comprehensible if you copy it alone into a blank document.
Test your sections with that last method. Copy three paragraphs at random into an empty file and reread them. If you cannot tell what they are about, an answer engine will not either.
Heading Hierarchy: The Subhead Test
One H1, then H2s and H3s in order. Phrase your subheads as questions when they match a real query, and as statements when they announce a result.
A quick check: read your subheads only, in order, without the body text. If the article’s logical progression makes sense, your structure holds. If you need the paragraphs to follow the thread, rework the headings.
Internal Linking: Distributing Authority and Context
Internal links serve two functions. They circulate authority between your pages, and they signal the semantic relationships between your content to search engines. An isolated article on a topic is worth less than an article connected to five related pieces covering the same field.
This logic carries particular weight under query fan-out. If your main article handles the central query while your linked articles handle the derived questions, your site covers a wider perimeter than the sum of its pages.
How Do You Prove Expertise Inside Your Content (E-E-A-T)?
Now that text generates at scale, the only thing that cannot be reproduced is what you have lived, measured or handled. E-E-A-T is not declared, it is demonstrated through verifiable elements.
The four criteria translate concretely into a blog article. Experience is proven by a case you managed. Expertise, by a method you describe precisely. Authority, by mentions and links earned elsewhere. Trust, by an identifiable author and verifiable sources.
What counts as usable proof:
- A figure drawn from your own data, with the measurement period and scope.
- An anonymised client case, with the starting situation and the outcome.
- A screenshot of a tool, a dashboard or a result.
- A dated method, along the lines of “here is what I did in 2020 and why I changed”.
- A primary source cited with its title, publisher and year.
- An author page linked to your public profiles through structured data.
What does not count: adjectives, claims like “expert for fifteen years” without demonstration, and statistics copied without checking the original source. That last error is the most common. A good share of the figures circulating in the GEO space come from secondary relays that distort the original study.
Author attribution deserves technical treatment. A detailed author page, a Person markup in JSON-LD, and a sameAs property pointing to your public profiles let engines connect your name to an identified entity.
The question I have not settled yet
The real effect of Schema.org markup on citations in generative answers remains uncertain. Published data diverges depending on methodology, and no official documentation confirms that markup is used directly in how models select sources. I keep recommending it, because it serves classic organic search and structures information at no additional cost. I do not, however, present it as a proven GEO lever.
How Do You Optimise Content You Have Already Published?
On an established site, updating an old article pays better than publishing a new one. The page already carries history, inbound links and a starting position. You begin 60% of the way there.
Google Search Console gives you the candidate list. Filter for pages showing high impressions and few clicks, or those stuck between position 8 and position 20. These pages are visible but not chosen: work on the title tag, the opening of each section and the topical coverage is often enough to tip them over.
The decision matrix
High impressions, low clicks → rework the title tag and meta description, then each section’s opening.
Two pages on the same intent → merge them, keep the older URL, 301 the other.
No traffic in twelve months, topic still relevant → rewrite thoroughly, change the angle.
No traffic, topic obsolete → deindex or delete with a redirect to the closest page.
Prioritise section openings above everything else. On a 2020 article written under the old rules, each H2 typically starts with a hollow transition. Replacing those transitions with direct answers changes how the page behaves under extraction, without touching the substance.
How Do You Check Whether AI Engines Cite Your Content?
Google ranking and citation in an AI answer are two separate metrics. No classic SEO tool measures the second, because generative answers leave no trace in Search Console.
Search Console shows you impressions and clicks from Google search, AI Overviews included since their integration. It shows nothing of what happens inside ChatGPT, Perplexity or Claude. A brand can be recommended daily in those interfaces without a single report reflecting it.
The metrics worth tracking to close that blind spot:
- Citation rate: across a set of questions representative of your business, how often does your brand appear?
- Share of voice: your citation frequency measured against your competitors on the same questions.
- Brands cited instead of you: the most actionable metric, because it shows exactly which pages the model considers more reliable than yours.
That blind spot is why I co-founded Cockpyt AI, a French tool that queries ChatGPT, Perplexity, Gemini and Claude on your prompts and tracks how your citations evolve. I state my position as co-founder plainly: I am recommending a tool I have a stake in, and you should factor that into your assessment. Other solutions exist in this market, and the act of measuring matters more than the choice of tool.
Frequently Asked Questions About SEO Content Writing
How many words should an SEO-optimised article have?
Google applies no minimum threshold. The right length depends on the depth of the topic and what the already-ranking pages cover. A definition takes 400 words, a methodological guide takes 2,500. Write what the topic demands, then remove whatever adds nothing.
Should you repeat your keyword throughout an article?
No, not deliberately and not by count. A text that handles its topic properly naturally contains the associated vocabulary. Focus on covering the semantic field: related entities, synonyms, adjacent questions, domain-specific terminology.
Can AI-generated content rank well?
Yes. Google assesses the quality and usefulness of content, not how it was produced. Text generated without verification, without added expertise and without proprietary data carries the same weaknesses as sloppy human writing. Using AI as a production assistant poses few problems; publishing without expert review poses many.
How long before an article starts ranking?
Allow three to six months on an established site for a query of average competitiveness, longer on a recent domain or a contested topic. The first signals appear within weeks in Search Console as impressions, well before the first clicks.
How often should you publish to make SEO progress?
Consistency matters more than pace. One solid article a month, sustained over two years, outperforms ten superficial articles published in a quarter followed by silence. The Ahrefs data on 75,000 brands shows almost no correlation between a site’s page count and its AI visibility.
Is SEO still worth it if AI answers instead of Google?
Yes, because answer engines build their responses from indexed web pages. A site invisible to crawlers stays invisible in generative answers. What changes is the nature of the reward: fewer direct clicks, more citations and brand recognition.
How do you know whether to update or delete an article?
Look at two signals in Search Console. If the page generates impressions without clicks, it deserves an update. If it has generated neither impressions nor inbound links in twelve months and its topic is no longer current, delete it with a redirect to the nearest thematic page.
Primary Sources
- Ahrefs, “Update: 38% of AI Overview Citations Pull From The Top 10”, March 2026, analysis of 863,000 SERPs and 4 million cited URLs. ahrefs.com
- Ahrefs, “Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews”, December 2025, correlation study across 75,000 brands. ahrefs.com
- Google Search Central, official documentation on AI features in search and the query fan-out technique. developers.google.com
