Key takeaways
- Answer engines extract passages, not pages. Optimise each section to stand alone.
- Lead every page and every H2 with a 40 to 60 word direct answer.
- FAQPage, Article, Product and BreadcrumbList schema make extraction reliable.
- Named authors, reviewers, dates and outbound citations are ranking evidence, not decoration.
- Track citations in ChatGPT, Perplexity and AI Overviews separately from blue-link rankings.
What AI SEO actually changes
Classic SEO optimised a page so a person would click it. Answer engine optimisation, often shortened to AEO or GEO, optimises a passage so a model will quote it. The unit of competition moved from the URL to the paragraph, and that single shift explains most of what follows.
Google AI Overviews, ChatGPT search, Perplexity and Copilot all work the same way underneath: retrieve a set of candidate passages, rank them for relevance and trustworthiness, then synthesise an answer with citations. If your paragraph cannot be lifted out and understood on its own, it will not be cited even when the page ranks well.
The practical consequence is that clarity now beats keyword density. A page written as a sequence of self-contained answers outperforms a longer page that buries the answer in narrative.
The 60-word answer rule
Open every page with a quick answer of roughly 40 to 60 words that resolves the query completely. Put it above the fold, before any history, definitions or storytelling. Models weight early passages heavily, and readers who bounce after ten seconds still leave satisfied.
Repeat the pattern at section level. Each H2 should be phrased as the question a user would type, and the first sentence beneath it should answer that question outright. Supporting detail, nuance and examples come after the answer, never before it.
Avoid hedging language in the answer sentence. Statements such as 'it depends on your use case' are unciteable. Give the default recommendation first, then list the conditions that change it.
Structured data that answer engines read
Four schema types cover most sites. Article with headline, datePublished, dateModified, author and reviewedBy signals freshness and accountability. FAQPage exposes question and answer pairs directly to extraction. BreadcrumbList explains where the page sits in the hierarchy. Product or SoftwareApplication with aggregateRating suits tool and review pages.
Keep the schema truthful and in sync with the visible page. Marking up an FAQ that does not appear on the page is a manual action risk and, more practically, produces answers that contradict the page a user lands on.
Validate with the Rich Results Test after every template change. Schema silently breaks when a field becomes undefined for a subset of routes, and a single malformed block can invalidate the whole graph.
Technical foundations that still decide everything
Server-render the content. If the answer only appears after client-side hydration, some crawlers will read an empty shell. Check by disabling JavaScript and reloading the page.
Ship one H1 per page, descriptive alt text on every meaningful image, a self-referencing canonical, a title under 60 characters and a meta description under 160. These are unglamorous, and they are still the difference between indexed and ignored.
Keep the sitemap generated from the same data source as the routes so new pages appear the day they publish, and keep robots.txt permissive except for genuinely private paths.
Internal linking as a retrieval map
Build hub pages for each cluster and link every child page back to its hub with descriptive anchor text. Retrieval systems use link context to understand what a page is about almost as much as the page text itself.
Link laterally between comparison pages, tool pages and tutorials that a reader would plausibly visit next. Three to eight contextual internal links per long page is a healthy range.
Avoid generic anchors. 'Read more' teaches a model nothing, while 'Claude vs ChatGPT pricing comparison' teaches it exactly what sits on the other end.
Measuring citations, not just clicks
Search Console still tells you about impressions and position, but it does not tell you whether ChatGPT quoted you. Build a small monthly routine: ask each major answer engine the 20 queries that matter most to your business and record which domains get cited.
Watch referral traffic from chatgpt.com, perplexity.ai and copilot.microsoft.com in your analytics. The volume is smaller than organic search but the intent is unusually high.
Treat a lost citation the same way you treat a lost ranking: identify the page that replaced you, compare answer clarity, freshness and evidence, then rewrite the weakest of the three.
A 30-day implementation plan
Week one: audit titles, descriptions, H1s and canonicals across every route, and fix duplicates. Week two: add or repair Article, FAQPage and BreadcrumbList schema on templates rather than individual pages.
Week three: rewrite the opening 60 words of your top 20 pages into direct answers and convert vague headings into questions. Week four: add author and reviewer blocks, refresh dates honestly, and build the citation-tracking spreadsheet.
Re-run the audit quarterly. Answer engines re-crawl aggressively, so improvements that used to take months to show now often register within two to three weeks.
References
- [1] Google Search Central documentation on structured data — developers.google.com
- [2] Schema.org Article and FAQPage specifications — schema.org
- [3] Google guidance on AI features in Search — support.google.com
Frequently asked questions
Is AEO different from SEO?
It is an extension of it. The technical foundations are identical, but AEO adds passage-level clarity, question-shaped headings, schema and visible evidence so that answer engines can quote you safely.
Does keyword density still matter?
Very little. Cover the topic and its natural vocabulary, then spend the remaining effort on structure and evidence, which is what retrieval systems actually score.
How long should an SEO page be in 2026?
As long as the question needs. A 700-word page that answers one question completely beats a 3,000-word page that answers it partially in paragraph nine.
How fast do AI answers update?
Most answer engines re-crawl popular pages within days and reflect content changes within two to four weeks, which is considerably faster than the old ranking cycle.