--- title: "Drasyo insights" canonical: "https://drasyo.com/insights" description: "Evidence-backed writing on SEO, answer-engine optimisation, generative-engine visibility and automation from the Drasyo team." count: 11 --- # Drasyo insights Evidence-backed writing on search, answer-engine and generative-engine visibility, automation, and how sites are actually built. Everything here comes out of client work. ## Accessibility is a build gate, not a retrofit The failures an audit finds late — unreachable controls, invisible focus, unlabelled buttons, text that fails contrast — are all consequences of decisions made early. Treated as a gate the build has to pass, they cost almost nothing; treated as a phase, they cost a rebuild. URL: https://drasyo.com/insights/accessibility-as-a-build-gate Category: websites Author: Drasyo Published: 2026-09-02 Updated: 2026-09-05 ### Questions **Is an automated checker enough?** It catches roughly the mechanical third: missing alt text, contrast failures, unlabelled fields. It cannot tell you whether the focus order makes sense or whether a custom control announces what it does. Run it, then use a keyboard. **What does WCAG AA require in practice?** For most sites: sufficient contrast on text and controls, everything operable by keyboard with visible focus, meaningful names on every interactive element, content that survives 200% zoom, and no information carried by colour alone. --- ## Which Core Web Vital actually moves when you fix things Cumulative layout shift is usually fixed in an afternoon by declaring image dimensions and reserving space for late content. Largest contentful paint responds to how the hero is delivered. Interaction latency is the one that needs architecture, not tuning. URL: https://drasyo.com/insights/which-core-web-vital-moves Category: websites Author: Drasyo Published: 2026-08-26 Updated: 2026-09-03 ### Questions **Which should we fix first?** Layout shift, almost always. It has the clearest causes, the cheapest fixes and the most direct effect on how a page feels to use — a page that reflows under someone's thumb reads as broken regardless of how fast it loaded. **Do lab scores predict what real users get?** Not closely. A lab run uses one device profile on a throttled connection; field data covers everything your actual visitors use. Use lab runs to find causes and field data to decide whether you have a problem at all. --- ## How to tell whether AI assistants cite your site There is no analytics panel for assistant citations. What works is a fixed list of the questions your customers actually ask, run against the assistants they actually use, on a schedule, with who gets named written down each time. URL: https://drasyo.com/insights/are-assistants-citing-your-site Category: geo Author: Drasyo Published: 2026-08-19 Updated: 2026-09-02 ### Questions **Can we see assistant traffic in analytics?** Partially and unreliably. Some assistant referrals arrive with a recognisable referrer and most arrive as direct traffic, so analytics will understate it by an unknown amount. Treat any figure you derive that way as a floor, not a measurement. **How often should we re-run the check?** Monthly is enough to see movement without mistaking noise for a trend. Answers vary between runs even with no changes on either side, so a single check tells you very little and three months of the same list tells you a lot. --- ## Writing an FAQ an answer engine will actually quote An answer is quotable when it survives being removed from the page. That means it names its own subject, answers in the first sentence, and contains no pronoun or phrase that depends on what came before it. URL: https://drasyo.com/insights/faqs-an-answer-engine-will-quote Category: aeo Author: Drasyo Published: 2026-08-05 Updated: 2026-08-28 ### Questions **How long should an FAQ answer be?** Long enough to be complete and short enough to be lifted whole — in practice two to four sentences. Answers past roughly a hundred words tend to get truncated mid-thought, and a truncated answer is worse than a brief one because it misrepresents you. **Does FAQ structured data guarantee we get quoted?** No. Structured data makes an answer easier to find and parse; it does not make it worth quoting. Marking up a vague answer just helps a model discover that the answer is vague. --- ## What a qualifying agent should and should not do The safe scope for a qualifying agent is narrow and worth defining before it is built: gather facts, classify the enquiry, route it, and set an expectation. Anything that commits the business — a price, a date, a technical assurance — belongs to a person. URL: https://drasyo.com/insights/what-a-qualifying-agent-should-not-do Category: automation Author: Drasyo Published: 2026-07-29 Updated: 2026-08-27 ### Questions **Can the agent give a price if we publish our prices anyway?** It can quote a published figure verbatim and link to where it is published. What it must not do is calculate, estimate or apply a published range to a specific job, because that reads as a commitment and it is the one thing customers remember exactly. **What happens when the agent does not understand the enquiry?** It should hand over and say so. An agent that guesses produces a wrong classification, wrong routing and a wrong first reply. Escaping to a person with the raw message intact costs one step; a misroute costs the enquiry. --- ## What is generative engine optimisation? Generative engine optimisation is making your pages easy for an AI assistant to read, trust and cite. It shares foundations with SEO but rewards different things: parseable HTML, unambiguous entity signals, and specific quotable claims rather than keyword coverage. URL: https://drasyo.com/insights/what-is-geo Category: geo Author: Drasyo Published: 2026-07-22 Updated: 2026-08-06 ### Questions **Is GEO just SEO with a new name?** No, though they overlap. SEO competes for a position in a ranked list. GEO competes to be the source a model cites when it composes an answer. Ranking helps but does not decide it — we regularly see sites outranked in search that are cited more often, because their pages are easier to parse and quote. **How do I know if AI assistants are citing my site?** Ask them. Query the assistants your customers use for the questions your business answers, and record who gets named. It is manual and it is the only reliable signal available today, because none of the major assistants report referral attribution the way a search engine reports impressions. --- ## Why enquiries go cold in the first hour An enquiry that waits is an enquiry that keeps shopping. The fix is rarely faster humans: it is one intake surface, an immediate acknowledgement that sets an expectation, and enough qualification captured up front that the first real reply can be useful. URL: https://drasyo.com/insights/why-enquiries-go-cold Category: automation Author: Drasyo Published: 2026-07-08 Updated: 2026-08-20 ### Questions **Is an automatic reply not just an obvious robot response?** It is, and that is fine when it does a job. A short acknowledgement that confirms what was received and says when a person will reply is useful information. What reads badly is an automatic message pretending to be a person, or one that says nothing beyond thank you. **Do we need a chatbot on the website?** Not necessarily. A chat widget adds another inbox to watch, and an unwatched one is worse than none. Start by making sure the channels you already have — form, email, phone, social messages — all land in one place with a record. --- ## How to shape content for answer engines Answer engines extract a passage rather than a page. To be extracted, put a 40 to 60 word complete answer directly beneath the question it answers, before any elaboration, and let the depth follow underneath. URL: https://drasyo.com/insights/answer-engine-content-shape Category: aeo Author: Drasyo Published: 2026-06-30 Updated: 2026-07-28 ### Questions **How long should an answer block be?** Roughly 40 to 60 words. Shorter tends to lack enough context to stand alone once it is separated from your page; longer gets truncated at an arbitrary point, which often cuts the qualifier that made the claim accurate. Write it as though it will be read with nothing else around it, because it will be. **Does FAQ schema still help?** It helps engines locate question and answer pairs unambiguously, which is worth the small effort. It does not make a weak answer strong. Schema describes structure; the extraction still depends on whether the answer text is genuinely self-contained. --- ## Keywords that end in a booking, not a visit A keyword is worth pursuing when the person typing it has already decided to act. Sorting a keyword list by volume optimises for traffic; sorting it by how close the phrase sits to a decision optimises for revenue, and the two lists rarely overlap. URL: https://drasyo.com/insights/keywords-that-end-in-a-booking Category: seo Author: Drasyo Published: 2026-06-24 Updated: 2026-08-14 ### Questions **Should I ignore high-volume informational keywords entirely?** No, but scope them honestly. Informational pages build topical coverage and earn citations from assistants. Just do not expect them to convert, and do not judge them on enquiries — judge them on whether they get quoted and linked. **How do I judge intent without a tool?** Search the phrase and read what the engine already returns. If the results are guides and definitions, the intent is informational. If they are service pages, listings and prices, the intent is transactional. The engine has already told you what it thinks the query means. --- ## Local search when your whole market is one island In a market the size of Malta, proximity barely separates anyone — a competitor twenty minutes away is still local. What decides local results here is category precision, review recency and consistent business data, not radius. URL: https://drasyo.com/insights/local-search-one-island Category: seo Author: Drasyo Published: 2026-06-10 Updated: 2026-08-14 ### Questions **Does adding town names to my pages help in Malta?** Rarely, and it usually hurts. Building a page for every locality on an island this size produces near-duplicate content competing with itself. One strong service page, with your service area stated plainly once, outperforms fifteen thin ones. **How much do reviews matter for local ranking?** Recency matters more than total count. A business with forty reviews spread across the last year generally reads as more active than one with two hundred that stopped three years ago. Steady arrival is the signal, not the number. --- ## Static-first is the cheapest performance win left Static generation removes the largest sources of slow pages: server work at request time, client-side content assembly, and render-blocking third-party requests. Most sites that fail Core Web Vitals fail on those three, not on anything exotic. URL: https://drasyo.com/insights/static-first-core-web-vitals Category: websites Author: Drasyo Published: 2026-05-19 Updated: 2026-07-02