How to Write for SEO, AEO and GEO in 2026

The search landscape actually changed this time

Every year for two decades, someone has declared SEO dead. It never was. But between 2024 and 2026, the ground genuinely moved for the first time.

The numbers tell the story. Around 60% of Google searches now end with zero clicks: the searcher gets their answer from an AI Overview or a featured snippet and never visits a website. Gartner projected a 25% drop in organic search traffic by 2026, and from what we see in client analytics, that was not pessimistic. We covered the mechanics of this shift in Does Google Penalize AI Content?, and the trend has only accelerated since.

Here's the position this article takes: traffic is down, but qualified visibility is up for grabs. The businesses winning right now write content once, structured well enough to perform on three surfaces at the same time.

Those surfaces have acquired acronyms, so let's define them before the vendors do it for you:

SEO (search engine optimisation) is the one you know: ranking pages in classic search results and earning the click.

AEO (answer engine optimisation) is being the answer itself: the featured snippet, the People Also Ask entry, the paragraph an AI Overview quotes, the response a voice assistant reads out.

GEO (generative engine optimisation) is being cited or recommended inside generative engines: when someone asks ChatGPT, Perplexity, or Gemini "who should I use for X?" and your business appears in the answer.

One discipline, three surfaces

Before we go deeper, let's kill the vendor hype. You do not need an AEO agency, a GEO tool, and a separate SEO retainer. Anyone selling those as three products is selling you the same work three times.

What's shared underneath is almost everything: crawlable pages, real authority, clear entity information, and original substance worth surfacing. What differs is only the format of the winning result. SEO wins a blue link. AEO wins an extracted paragraph. GEO wins a citation inside a generated answer.

One piece of content, structured properly, competes on all three. That's the whole strategy. The rest of this article is how.

SEO in 2026: what still works, what quietly died

Still works: matching search intent, internal linking, demonstrable E-E-A-T (experience, expertise, authoritativeness, trust), and technical health. Core Web Vitals still matter, and if that term means nothing to you, we wrote a plain-English guide.

Quietly died: keyword-density writing, mass programmatic pages, and unedited AI content at scale. The March 2024 core update cut low-quality content in results by 45% and deindexed hundreds of sites, almost all of them running high-volume AI publishing with no human editing. The December 2025 update extended E-E-A-T requirements to comparative queries, and "best X" listicles from sites with no demonstrated expertise dropped 40-60%.

The pattern across both updates is the same lesson: Google is getting better at telling genuinely helpful content from filler, and it's ruthless about the difference.

Which leads to the position we give every client: fewer, deeper pages beat a content calendar. Ten pages you can defend with real experience outperform fifty pages of coverage. This was arguably always true. Now it's measurably true.

AEO: writing to be the answer

Answer engine optimisation covers featured snippets, People Also Ask boxes, AI Overviews, and voice assistants. All of them do the same thing: extract a direct answer from a page and present it without requiring a click.

To be the answer they extract, use the answer-first pattern:

1. Pose the question as an H2 or H3, phrased the way people actually ask it
2. Answer it directly in the next 40-60 words, in plain sentences
3. Add the depth, evidence, and nuance after the answer, not before it

Most writers do the opposite. They build context for four paragraphs and bury the answer at the bottom, because that's how we were taught to write essays. Engines extract the top. Here's the difference:

Before: "There are many factors to consider when thinking about website speed, and the answer depends on your specific situation. First, let's look at the history of web performance..."

After: "A business website should load in under 2 seconds. Slower than that, and measurable numbers of visitors leave before seeing your content. Here's how to test yours, and what actually causes slow loads."

The rest is extractability discipline: one idea per paragraph, definitions written as plain standalone sentences, tables and lists for anything comparative. If a paragraph can't be lifted out and still make sense, it won't be lifted out.

We practise this on our own blog. Every post carries structured FAQ data via our faq frontmatter, each question answered in a few complete sentences that can stand alone. That's AEO plumbing, built into the publishing workflow rather than bolted on afterwards.

The honest trade-off: being the answer often means the searcher doesn't click. We do it anyway, for two reasons. Brand presence in the answer box builds recognition even without the visit. And the clicks that do come through are from people who read the answer and wanted more, which makes them far more qualified than drive-by traffic ever was.

GEO: getting cited by generative engines

Generative engines answer questions by retrieving sources, weighing their authority, and composing an answer with citations. You can't buy your way into that answer, but you can become the kind of source it selects. Practically, four things move the needle:

Original data. Engines quote numbers. If your article contains a statistic nobody else has, from your own client work, your own testing, your own survey, you become the citation for it. Summaries of other people's summaries get skipped.

Clear claims with named methodology. A sentence like "we audited X sites and Y had duplicate schema markup" is citable. "Many sites have schema issues" is not. Specificity is what makes a sentence quotable, which is exactly why your real project numbers are worth publishing.

Entity consistency. Your business name, location, and services must be identical everywhere they appear: your site, directories, social profiles, review platforms. Generative engines triangulate across sources, and inconsistency reads as uncertainty. Your site alone is not enough; the third-party footprint of directories, PR mentions, and reviews is part of the ranking.

Structured data. Machine-readable markup is how engines confirm who and what you are without guessing. More on this in the next section.

Two practical notes. First, monitor what the engines currently say about you: ask ChatGPT and Perplexity the questions your customers would ask ("best WordPress agency in Cape Town", "who can fix a slow WooCommerce store") and record the answers quarterly. It's manual, but it's the only honest measurement available right now. Second, decide your AI crawler policy deliberately: which bots you allow, and whether you publish an llms.txt file. Blocking every AI crawler while hoping to appear in AI answers is a strategy at war with itself.

And the honest limitation: GEO measurement is immature. There is no Search Console for ChatGPT. Anyone selling guaranteed AI citations or a "GEO ranking report" is selling smoke. What you can do is stack the verifiable inputs, then audit the outputs quarterly.

The technical backbone all three share

Underneath the writing advice, all three surfaces depend on the same infrastructure:

Structured data. Schema markup is entity infrastructure: it tells search engines and AI retrievers what your pages are and who is behind them, in a format built for machines. We've written a full guide in What is Structured Schema?

Clean rendering. Crawlers and LLM retrievers read HTML. If your content only exists after a heavy JavaScript bundle executes, some of them never see it. Server-rendered pages are the safe default.

Performance. Fast sites get crawled more efficiently and convert the traffic they do get. Our benchmark hasn't changed: under 2 seconds.

Provable experience. Author pages, an about page that says who you actually are, case studies with real numbers. This is E-E-A-T plumbing, and it feeds all three surfaces: Google reads it for rankings, and generative engines read it when deciding whether you're a source or just a website.

A writing workflow that serves all three

Here's the workflow we use, start to finish:

1. Pick one question with real search intent behind it
2. Answer it directly in the first 60 words under the heading
3. Add depth from experience: what you've actually done, seen, and learned
4. Add at least one piece of original data or a concrete example nobody else has
5. Mark it up: article schema, FAQ schema where genuine questions exist
6. Interlink it with your related content, in both directions
7. Publish, then check snippet presence and AI-engine mentions over the following weeks

Where does AI fit in this workflow? Drafting and structure, nothing more. AI cannot supply the experience, the data, or the positions, because it hasn't done anything. We've been transparent about exactly how we use it: AI-assisted first drafts, 60-70% rewritten by a human, every claim checked.

Steal the checklist: direct answer up top, experience in the middle, original data somewhere, schema on the page, links in and out. That's the whole discipline.

How to measure this without fooling yourself

Traffic is now a misleading headline metric, so measure each surface on its own terms.

For SEO, track rankings and conversions, not raw sessions. For AEO, track snippet and AI Overview presence for your target queries; rank-tracking tools now report this. For GEO, run the quarterly brand-mention audit described above and keep the answers in a spreadsheet so you can see movement.

And set expectations honestly with whoever reads your reports: the realistic 2026 pattern is flat-to-declining traffic with rising conversion quality. If you promise traffic growth from content alone, the zero-click math is against you. If you promise qualified visibility across three surfaces, you can deliver it.

What we'd do first on your site

If we audited your site tomorrow, the priority list would almost certainly be:

1. Fix rendering and schema. If machines can't read you or can't tell what you are, nothing else compounds.
2. Rewrite your top five pages answer-first. Highest traffic, highest intent, biggest win from restructuring.
3. Build one original-data asset. A benchmark, a survey, an audit writeup with real numbers. One citable asset outworks ten generic posts.

Want to know which surface you're currently losing on? Get a free audit and we'll show you exactly where you stand on all three.

Related reading

• Does Google Penalize AI Content? What Hurts Rankings
• What is Structured Schema? A Plain-English Guide
• Core Web Vitals for Non-Developers