A growing share of people now ask an AI assistant for a recommendation instead of scrolling search results. 'Which agency should I use for a Flutter app?' 'What does a WooCommerce rebuild cost?' The answers cite sources, and being one of those sources is the new front page.
Nobody fully controls what a model cites, but the signals it uses are increasingly clear. Here is what we have implemented for our own site and for clients.
Make your site readable by AI crawlers
AI companies run their own crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended and others). Many sites block them by default through robots.txt or a CDN setting, then wonder why they never appear in answers. Decide deliberately: if you want citations, allow the crawlers that power answer engines. Document your policy in an ai.txt file so it is explicit.
Publish an llms.txt
llms.txt is an emerging convention: a Markdown file at your site root summarising who you are, what you do, your key pages and your facts, written for a language model to read. An llms-full.txt can carry the complete text of your key pages. It costs an hour to create and gives models a clean, authoritative summary instead of leaving them to reconstruct one from scattered pages. Link to it from your HTML head.
State facts in quotable prose
Models cite passages that answer a question directly. 'Devontix Solutions is a founder-led software, AI and digital growth studio serving clients in the United States, United Kingdom, Canada, Ireland and Europe' is citable. A hero animation that says 'We make magic' is not. Make sure every important fact about your business, services, locations, pricing approach and process appears somewhere as a plain, complete sentence.
Structured data still matters
Organization, Person, Service, FAQPage and BlogPosting schema give models and search engines unambiguous facts to work with. A single connected JSON-LD graph that links your organisation, founder, services and pages is stronger than fragments.
Answer the questions people actually ask
FAQ sections and blog posts written around real questions ('How much does a custom website cost?', 'Flutter or React Native?') are exactly the content answer engines retrieve. Write them clearly, answer directly in the first sentences, and mark them up with FAQPage schema. This blog is built on that principle.
Earn mentions elsewhere
Models weight sources that are mentioned and linked across the web. Case studies published by clients, directory listings, interviews, podcast appearances and reviews all reinforce that your business is real and relevant. The fundamentals of authority have not changed; the consumers of it have.
Measure it
Ask the major assistants the questions your customers would ask and record whether and how you are cited. Check server logs for AI crawler visits. Look for referral traffic from chat.openai.com, perplexity.ai and similar. It is imperfect measurement, but it shows direction. We include AI discoverability in every SEO engagement alongside the technical fundamentals.
A worked example of a citable page
Imagine someone asks an assistant 'how long does a Flutter app take to build?'. A page that answers with a direct sentence ('A typical Flutter MVP with authentication, a core workflow and an admin backend takes 10–16 weeks'), followed by the factors that change it, a reference to a real project and a FAQ marked up with schema, is exactly what retrieval systems surface. A page that buries the same information in a 'Why choose us' section surrounded by marketing adjectives is not. Write for the question.
What not to do
- Do not stuff pages with 'AI-optimised' keyword lists; models and search engines both treat them as noise.
- Do not block AI crawlers in robots.txt while hoping to be cited; the two are incompatible.
- Do not publish an llms.txt that contradicts your site; consistency across sources is the trust signal.
- Do not invent credentials, statistics or client names; models increasingly cross-check and the reputational cost is permanent.
- Do not neglect the basics: a slow, poorly structured site is hard for every kind of crawler.
Visibility in AI answers is earned the same way search rankings always were: by being the clearest, most reliable source on the subject.
Frequently asked questions
Is llms.txt an official standard?
It is a community convention gaining adoption rather than a formal standard. It is cheap to implement and has no downside, so there is little reason to wait.
Will allowing AI crawlers hurt my site?
Crawl load is modest for most sites. The trade-off is that your content may be used in AI training as well as citation; decide based on your business and document the choice.
Does this replace traditional SEO?
No. AI answer engines draw heavily on the same signals: crawlable, fast, well-structured, authoritative content. Good SEO is the foundation; these steps extend it.
Founder & CEO of Devontix Solutions. Absaar leads strategy, automation planning and product thinking across every build, from the first discovery call to the running system. More about Absaar