The Answer Engineer

Classic search crawled, indexed and ranked pages; a person compared ten links and clicked one. AI search fans a full question out into several searches, keeps the passages that answer it, adds what the model learned in training, and writes one answer that names a few businesses. The decision moved from your website (after the click) into the answer (before it). SEO still matters, but ranking is no longer the finish line.

An assistant answers from two places: what the model learned in training (frozen at a cutoff date) and what it retrieves live. With retrieval, it interprets the prompt, fans it out into searches, pulls pages, splits them into passages, keeps the passages that answer the question, writes a grounded answer and sometimes cites its sources. Passages, not whole pages, are the unit AI works with, and being mentioned is not the same as being cited.

Much of SEO carries over, with a new meaning. Keywords become prompts. Backlinks become mentions and agreement across many sources. Ranking position becomes share of answer: how often you are named for the prompts that matter. Click-through rate becomes AI referral visits. Crawlability, quality content, clear site structure and authority still form the foundation.

Before AI can mention you, its crawlers must load your pages. robots.txt decides which crawlers may fetch which paths. Many AI crawlers read the raw HTML and do not run JavaScript, so content that appears only after scripts run may be invisible. Login walls, errors, slow responses and blocked bots stop them too. Some controls are separate: for example, OpenAI lists GPTBot (training) and OAI-SearchBot (search) separately, and Google-Extended governs use for Gemini, not Google Search.

Make your facts easy for machines to read. Write key facts in plain text on the page. Add schema.org structured data (usually JSON-LD) for your organisation, locations, products, prices, hours and FAQs, and keep it in sync with the visible page. Publish discovery files such as llms.txt and brand.json that point AI to your most important pages and facts. They are emerging conventions and support varies, so they add to plain, consistent facts and do not replace them.

Write so that any passage can be lifted out and still make sense. Use real questions as headings, put the direct answer in the first one or two sentences, then add detail. Name the subject in each section, include specifics (prices, places, numbers, steps), and use comparison tables, lists and FAQs. Show who wrote it and when it was updated.

AI learns about you from everywhere your brand appears: your website, business profiles, directories, review sites, social profiles, marketplaces, Wikidata and more. Keep the core facts identical everywhere: name, address, phone, hours, prices, categories and descriptions. Connect your official profiles to your site with sameAs links in your structured data.

AI trusts agreement. What others say about you, in reviews, social media posts, local news, best-of lists, comparison articles and videos, often matters as much as your own site. Find which third-party sources assistants cite for your prompts, earn honest mentions there, respond to reviews, and correct errors with the publisher. Never buy or fake reviews.

Stale pages fade, and stale facts get repeated. Watch for answers that give old hours, prices or features, or that are wrong outright (hallucinations). Run a correction workflow: find where the wrong fact comes from, fix it at the source and on your site, update structured data and profiles, then re-check the answers. Review key fact pages on a schedule and whenever something changes.

Buyers ask long, specific questions. Build a prompt set from real customer language: sales calls, support tickets, reviews, forums and on-site search. Group prompts by intent (discover, compare, decide), persona and location. Track the same prompts across several assistants, because each can answer differently.

Measure the whole chain: are you mentioned, are you cited, is the answer accurate, do AI crawlers read your pages (server logs), and do people arrive from AI answers (referrals in analytics) and convert? Rank prompts by what their pages actually earned. A prompt can win without being cited, and a cited prompt can bring no one. Visits beat mentions.

Put it all together as a loop: audit (prompts, answers, loadability, footprint), fix (the highest-impact problems first), publish (answer-shaped pages and corrected facts), measure (mentions, citations, crawler reads, visits) and double down on what wins. Run it on a steady cadence. The capstone project runs one full loop.