A growing share of searches now end in an answer instead of a click, and the brands quoted in those answers are the ones publishing specific, structured, checkable content. Roughly two thirds of searches already resolve without a visit to any website, and the figure is higher where an AI overview appears. The page still matters. It just has a different reader.
We build sites and write content, and we have spent this year rebuilding our own around exactly this question. Here is what moves the needle, in the order we would do it.
What answer engines quote
Facts with edges: a number, a date, a price, a definition, a rule, a step. Models cite passages they can lift without having to interpret them, which is why a page that says “video production in Dubai starts around AED 12,000 for a single-location corporate film” gets quoted and a page that says “we deliver world-class video solutions” does not.
The test we use on our own drafts: read a paragraph and ask whether a stranger could repeat one sentence from it as a fact. If not, it is decoration.
Structure the page like an answer
- One question per heading. Headings phrased the way people ask, answered immediately underneath
- The answer in the first two sentences. Then the detail, then the nuance
- Tables for anything comparative. Formats, prices, specifications and turnarounds are quoted far more often as rows than as prose
- An FAQ with real questions. Four to six, written in the words customers use, not the words your industry uses
- Dates and places. Entities with specifics attach to a brand in a way that adjectives never do
The build side
Crawlers that feed answer engines are less tolerant than a browser. Content that only exists after JavaScript runs is a risk, so the words that matter belong in the served HTML. Beyond that the list is short and mostly unglamorous: valid structured data for the organisation, the article and the FAQ; a clean heading hierarchy with one H1; a canonical that matches the URL; a sitemap that lists every page; and pages that load fast enough not to be abandoned mid-crawl.
An llms.txt file at the root is worth adding too. It is a plain-text index of the site written for language models, and it costs an hour. We publish one at steermedia.com/llms.txt, listing every service with its starting price in AED, because a model that can read a price can quote it.
What we publish, and why
Three kinds of page earn citations in our experience. Pricing, because almost nobody publishes it and every buyer searches it. Process, because “how does X work” is the shape of most questions put to an assistant. And local rules, because regulation changes, dates are specific and the answer is checkable: filming permits, packaging regulations, advertising permits.
Our guide to video production cost in Dubai exists for the first reason, and it is the single page that brings the most qualified enquiries. The UAE ad spend piece covers the shift to zero-click in more depth.
What does not work
Adjectives, gated PDFs, pages that answer five questions badly, and claims with no number behind them. An answer engine cannot quote a superlative, and a reader who arrives from one has already had the summary. What brings them to the site is wanting the detail underneath it, which means the detail has to be there.
We build this into the sites we design and the content we run, because the two halves only work together. A perfectly marked-up page with nothing specific on it gets crawled and ignored.