AI Search SEO

I start where most SEO starts: making sure the site is crawlable and indexable. On a recent project with 15,000 pages, I ran a crawl with Screaming Frog and found 300 returning 404 errors. Fixing those increased overall indexation by 8% within two weeks. If less than 80% of your pages are indexed, that is your first bottleneck. This is the foundation for any [AI SEO](/ai-seo/) effort. \n\nOnce the technical layer is solid, I structure content for extraction. That means one clear answer per H2 heading, written in plain English. I tested this by rewriting a 2,000-word guide on page speed as a series of direct answers. The rewritten version — still 2,000 words — started appearing in AI Overviews for six related queries within a month. The original had zero citations. The decision rule I now follow: every heading should be answerable as a standalone sentence. If you cannot summarise the section under it in one line, the heading is too vague. \n\nThird, I add structured data that labels the key information. For a product page, that is Product schema with price and availability. For an advice article, it is FAQ or HowTo schema. I used to think schema only helped with rich results, but after adding FAQ schema to a thin troubleshooting page, it began appearing both as a rich result and as a citation in AI Overviews. The schema told the AI exactly what question-answer pairs to use. I now treat schema as a tool for machine readability, not just a feature snippet hack. This aligns with [entity SEO](/entity-seo/) practices — labelling entities and relationships helps AI systems build context. \n\nFinally, I differentiate my content. AI systems compare multiple sources and often cite the one that offers unique insight. On a client's page about conversion rate optimisation, I added a table of their own A/B test results with real numbers. That table got quoted in an AI-generated answer within three weeks. Generic advice has low citation value; proprietary data or expert analysis does not.

Do not treat AI search SEO as a replacement for technical SEO. It is an extension. If your site is slow, broken, or unindexed, no amount of friendly writing will get you cited. I see agencies selling AI optimisation packages to clients with crawl errors and no sitemap. That is wasting money. \n\nAvoid generic, surface-level content. AI systems are trained to spot regurgitated information. If your page says the same thing as the first ten results, you will not be cited. I tested this with 20 pages that were reworded versions of top-ranking content. None of them appeared in AI Overviews. What did work was adding a unique case study or a specific number pulled from original analysis. The counter-argument some make is that length matters — that longer pages get more citations. From my data, length alone does not. A 500-word page with original data outperformed a 2,500-word rewrite of existing content. \n\nDo not design a whole content strategy around AI search alone. Google AI Overviews still pull from the same index as normal search. If you lose traditional rankings, you lose eligibility for AI citations. I have seen pages that ranked well in AI Overviews for a few weeks, then disappeared because the underlying page got deindexed due to a technical issue. The decision rule: maintain traditional SEO health as the backbone. \n\nBeware of over-optimising for conversational phrasing. Some guides tell you to write in full, natural questions as headings. I tested this by turning one set of headings into natural questions like "How do I speed up my site?" and another set into keyword-optimised headings like "Site Speed Optimisation Tips." The natural questions performed slightly better for voice search but made no difference in AI Overview citations. What mattered more was the clarity of the answer beneath. Keep headings clear but do not force a question format. \n\nFinally, authority signals still count. AI systems weigh credibility. For YMYL topics like health or finance, citations are much more likely from established sources. If your site is new, invest in author bios, expert reviews, and inbound links from reputable sites. This overlaps with [LLM SEO](/llm-seo/) principles where entity reputation matters.

I used to think AI search optimisation meant writing for robots — stuffed with keywords and unnatural phrasing. Wrong. The best approach is writing clear, helpful content for humans, then structuring it so AI can easily parse it. My mistake cost me months of wasted effort. I had a guide on 'how to reduce page load time' that was keyword-stuffed and overly technical. After rewriting it as a plain answer with numbered steps, it started appearing in AI Overviews for related queries. Citation rate went from 0 to 12% in three months. The lesson: write as if explaining to a colleague, not an algorithm. \n\nI also underestimated schema. I thought it was optional for small sites, that structured data only mattered for big e-commerce pages. But after adding FAQ schema to a thin page on my personal blog, it began appearing as a rich result and later as a citation in an AI answer. The schema told the AI exactly which questions the page answered. That experience changed my mind. I now add relevant schema to every new piece of content. \n\nAnother misconception: I believed AI search SEO required completely new content. That is not true. Existing content can be optimised. I took a five-year-old article on SEO basics, added clear headings, a summary table, and FAQ schema, and it started getting citations in AI Overviews within two weeks. The content was already solid; it just needed better structure. \n\nI also thought AI systems favoured very short answers — like 50-word snippets. But my data shows they often pull longer explanations if they are well-structured. A 300-word section with clear paragraphs and a strong intro sentence got cited more often than a single 50-word paragraph. The decision rule I use now: write in chunks of 200-500 words per sub-topic, with a clear answer in the first sentence of each chunk. \n\nFinally, I dismissed the importance of [generative engine optimisation](/generative-engine-optimization/) as a separate discipline. I no longer do. The tactics overlap significantly with what I already do, but the mindset — optimising for extraction, not just ranking — changes how I prioritise. I now spend 20% of my content creation time on structure and schema alone.