Someone looking for a development partner increasingly does not open Google. They ask an assistant, get three recommendations with reasoning, and act on them. If you are not in that answer, you were never in the running.
Why this is a different problem
Search ranking is competitive — you are placed against others on a page and the user chooses. An AI answer is synthesised: the model reads several sources and produces one response. You are either part of it or invisible, and there is no second page to be on.
That changes what helps. Ranking rewards authority and links. Citation rewards being clearly, factually, and unambiguously useful on the specific question being asked.
Write in citable units
Models extract and attribute passages. Content that is easy to extract gets cited more than content that is merely good.
- Use headings that are the actual question someone asks.
- Answer directly in the first sentence under the heading, then elaborate.
- State specifics — timeframes, technologies, constraints — rather than adjectives.
- Avoid burying the answer under three paragraphs of preamble.
Ship llms.txt
An emerging convention: a plain-text file at your domain root that describes what your organisation does and links to your key pages in a form built for machine consumption rather than layout.
Adoption is still uneven and it is not yet a guaranteed win. It costs very little, and generating it from the same data that drives your navigation means it cannot drift out of date.
Structured data matters more than it did
Schema.org markup — Organization, Service, FAQPage, BreadcrumbList — gives assistants unambiguous facts instead of inferences drawn from prose. If you want to be described accurately, state the facts in a format designed for being read.
Server-rendered HTML is a hard requirement
Many AI crawlers do not execute JavaScript. A client-rendered single-page application may be entirely invisible to them, no matter how good the content is. This is the single most consequential technical factor, and it is one many otherwise-modern sites fail.
What does not change
Being genuinely worth citing. Models synthesise from sources that contain real information — specific, verifiable, useful. Thin content built to satisfy a keyword performed poorly in search and performs worse here, because there is nothing in it to extract.