Video: "How-to SEO for Google AI Overview (Tutorial by RankYa)" by RankYa on YouTube.
Why this matters now
For the last twenty years, ranking on Google meant fighting for one of ten blue links. That's still useful — but a growing slice of queries now return an AI Overview at the top of the page, with two or three websites quoted as the source. If your business is one of those quoted sources, you get the click and the implied endorsement of being "the answer." If you're not, you may as well be on page two.
The same pattern repeats inside ChatGPT, Claude, Gemini and Perplexity. They generate one answer, cite a handful of pages, and the rest of the web is invisible to that conversation. RankYa's tutorial focuses on Google specifically, but almost every tactic transfers.
1. Pick a real question, not a topic
The single biggest mistake is writing a page about "kitchen renovations." Google's AI doesn't pull from topics — it pulls from answers. Reframe the page around a specific question a real person would type, like "How long does a small kitchen renovation take?" or "What are the best plants for a sun-filled garden?"
The question becomes the page's H1. The first paragraph is the answer in 40–60 words. Everything after is the supporting detail. That structure mirrors exactly how Google's AI extracts and quotes content.
2. Lead with the answer (the "answer-first" rule)
Don't open with throat-clearing — "In today's competitive landscape…" is a tell that the AI will skip you. Open with the direct answer, then back it up. Think of the lead like a featured snippet: short, complete, quotable on its own.
If a busy reader (or an LLM) only reads the first paragraph, they should still leave with the right answer. Everything else is depth for the people who want more.
3. Structure your headings as sub-questions
Each H2 and H3 should preview a useful chunk on its own. RankYa's point here is subtle but matters: AI models scan headings to understand a page's outline and to choose which section to quote. Vague headings like "Key Considerations" tell the model nothing. "How long does the planning stage typically take?" tells it everything.
4. Describe steps in prose, not bullet points
This one feels counter-intuitive. We're all trained to use bullet lists for steps. RankYa's argument — and it lines up with what we see in the wild — is that AI Overviews prefer self-contained paragraphs for each step, because they can lift one paragraph and quote it as a complete answer. A bullet point on its own is too thin.
The technical name for this is "semantic distance." A paragraph contains enough surrounding context for the AI to be confident it understands the meaning. A six-word bullet doesn't.
5. Hand-write nested Schema markup
Plugins generate flat, generic schema. To actually help Google's AI understand your entity, you want nested JSON-LD that ties everything together: Organization → WebSite → Article → author → sameAs. Add FAQPage for any Q&A blocks and HowTo for any process you describe in prose.
It looks intimidating the first time. Once you've built one, you're copy-pasting structure for every page after that. (Every NordSys page is built this way — view source on this article and you'll see a full @graph with five linked entities.)
6. Add complementary media on the same topic
RankYa stresses that Google rewards "topical depth" — and AI models read it as a signal that you're a serious source on this subject. The fix is small: alongside your written answer, include an image with a descriptive filename and alt text, embed a short video (your own, or a credited one like the embed above), and offer a downloadable PDF version. The page now has four formats covering one question. The AI sees a richer entity.
7. Build a topic cluster, not a one-off article
A single article on "kitchen renovation timelines" is fine. Surround it with five supporting articles — "How to brief a kitchen designer", "What's the difference between flat-pack and bespoke kitchens?", "Average UK kitchen renovation costs in 2026" — all internally linked, and you've built topical authority. AI models lean heavily on this. They'd rather quote a site that clearly knows the whole space than one that has a single article on it.
8. Write for people; let humans earn the links for you
The closing point in the video is the most important one and the easiest to forget. AI models still weigh authority signals — links, mentions, citations on third-party sites. The fastest way to earn those is to write something a real person actually wants to share. Optimised content nobody reads earns nothing.
Practical version: read your draft aloud. If it sounds like a human you'd trust, ship it. If it sounds like SEO copy, rewrite it.
The one-page checklist
Run this against any page you want to rank in AI answers. If you can tick six of the eight, you're already ahead of most of your competition.
- One specific question as the page's H1.
- 40–60 word direct answer in the opening paragraph.
- Headings phrased as sub-questions, each previewing its section.
- Steps written as paragraphs, not bullets, for semantic distance.
- Hand-written nested JSON-LD linking Organization, WebSite, Article (and FAQPage / HowTo if relevant).
- Complementary media: image with descriptive alt, embedded video, optional PDF.
- Topic cluster: at least three internally-linked supporting pages on related sub-questions.
- Reads naturally when spoken aloud — no SEO throat-clearing, no adjective soup.
What the video gets right (and what to layer on top)
RankYa's tutorial is solid and unfussy. It treats AI Overviews as an extension of classic SEO craft — better content, cleaner structure, real schema — rather than a separate dark art. That's the right mental model.
What we'd add for 2026: pay equal attention to off-Google AI surfaces. ChatGPT and Perplexity weight third-party mentions (Reddit, listicle round-ups, podcast transcripts) far more heavily than Google does. A monthly piece of earned coverage on a respected industry site moves your visibility inside ChatGPT faster than a month of on-page tweaks.
And don't forget llms.txt at your site root — the emerging convention for telling LLM crawlers which pages are your canonical, citable content. It's a five-minute job that'll only get more useful.
Where this connects to NordSys
This is the day-job. Our SEO & AI Ranking service runs every one of these tactics on your site every month — content with citations, hand-written nested schema, topical clusters built around real customer questions, and the off-site authority work that gets you quoted in AI answers. You get a one-page report each month showing what moved.
If you'd rather just do it yourself with the checklist above, brilliant — that's what tutorials are for. If you'd rather hand it off, we're a phone call away.