A practical GEO checklist to get your website cited by AI assistants

Getting cited by AI assistants comes down to six jobs: let AI crawlers reach your pages, publish an llms.txt file, restructure key pages so they answer questions directly, add FAQ and organisation schema, keep your business details consistent across the web, and measure what the assistants actually say about you. None of it requires a rebuild, and most of it is checkable in a single sitting. This is the do-this-now companion to the conceptual GEO pieces on this blog: a checklist you can hand to a developer and a content lead and start working through today.

Step one: can AI crawlers actually reach your site?

Start with the unglamorous plumbing, because it is where sites fail silently. Open your robots.txt file and check for rules that block AI user agents, either deliberately or through an old blanket disallow. Then check the layers above it: firewalls, CDN bot-protection settings and security plugins often block AI crawlers by default, and no robots.txt edit will fix a block that happens at the network edge.

The practical test is simple. Fetch a key page with each crawler's user-agent string and confirm you get a normal response rather than an error or a challenge page. Do the same for your sitemap. While you are there, make sure your important content is present in the raw HTML: many AI crawlers do not execute JavaScript the way Googlebot does, so a page that renders entirely client-side can look empty to them.

Names and behaviours are current at the time of writing, and the labels above are a simplification: Google-Extended, for one, is a robots.txt control rather than a crawler in its own right. Agent lists change, so revisit this table's logic, not just its rows, each quarter. The decision is a business one: blocking training crawlers protects content from reuse, but it also removes you from the answers your buyers are reading.

Step two: publish an llms.txt file

llms.txt is a plain-markdown file at your site root that gives AI systems a curated index: who you are, what you do, and links to your most useful pages with one-line descriptions. It is an emerging convention rather than an enforced standard, and adoption by the major platforms is uneven, so treat it as low-cost insurance rather than a silver bullet.

It earns its place on this checklist because it is quick to write and it forces a valuable editorial decision: which ten to twenty pages would you want an AI assistant to read first? Write the file by hand rather than generating it from your sitemap. Lead with a short description of the organisation, group links under clear headings, and describe each link in terms of the question it answers. Keep it current whenever you publish or retire a cornerstone page.

Step three: make every important page answer-shaped

AI assistants quote passages, not pages. When one assembles an answer, it retrieves a handful of chunks from across the web and cites the ones it leans on. A page that buries its point beneath a scene-setting introduction gives the retrieval step nothing quotable, no matter how good the material further down is.

AI assistants do not rank ten blue links. They quote the clearest passage they can find, from whoever wrote it.

The fix is structural, and you can apply it page by page. Open with a direct answer to the question in the title, in two or three sentences a machine could lift verbatim. Use question-shaped headings that match how people actually ask. Keep each section self-contained, so it still makes sense when read without the rest of the page. Prefer lists and tables for anything comparative or sequential. Define terms plainly and attribute your claims, so that a passage can be checked rather than simply believed. This is the same discipline behind our SEO and content work: writing that serves the reader first tends to serve the machines as a by-product.

Step four: add FAQ and organisation schema

Structured data is how you state, in a format machines parse without guessing, what a page is and who published it. Two types do most of the GEO work. FAQPage schema wraps genuine question-and-answer pairs so that assistants can match them to user queries cleanly. Organization schema, referenced from every page, establishes your name, logo, location and official profiles.

Three rules keep schema honest and useful. First, mark up only what is visible on the page; schema that describes content the reader cannot see is a trust problem waiting to be discovered. Second, validate everything with a schema testing tool before it ships, because one syntax error can invalidate the whole block. Third, connect your entities: use sameAs links to your official profiles, and give your organisation a stable identifier so every page points at the same entity rather than a slightly different copy of it.

Step five: is your entity consistent everywhere?

AI systems build a picture of your organisation from every mention they can find: your site, directories, social profiles, review platforms, press coverage. When those sources disagree about your name, address, category or description, the picture goes blurry, and an organisation a model cannot pin down is an easy one to describe wrongly.

Run a consistency pass. Pick one canonical form of your business name and use it everywhere, ruthlessly. Align your address, phone number and description across your site footer, contact page, Google Business Profile, LinkedIn and every directory you control. Write a plain one-paragraph description of what you do and reuse it verbatim wherever a profile asks for one. Then check the places you do not control: if an old directory listing or a stale wiki entry describes you wrongly, request a correction, because assistants may be reading it.

Step six: close the loop with measurement

You cannot manage what you never look at, and plenty of organisations have never once asked an AI assistant the questions their buyers ask. Build a small, repeatable loop. Write down the ten to twenty questions that matter commercially: the "best X in Y" queries, the "how do I" queries, the comparisons. Each month, put them to ChatGPT, Claude, Perplexity and Google's AI results, and record three things: were you mentioned, were you cited as a source, and was what the assistant said about you accurate.

Pair that with your analytics. Referral traffic from AI platforms is visible in most analytics tools and worth segmenting, so you can see for yourself how those visitors behave rather than assuming. Watch your server logs for AI crawler activity to confirm the access work from step one is holding. Feed what you learn back into the checklist: an unanswered question becomes a new answer-shaped page, a wrong claim becomes an entity correction, a competitor's citation becomes a brief to beat. This is the same measure-and-iterate habit that drives our analytics and CRO practice, applied to a new surface.

How OlDevs can help

OlDevs is a full-stack technology studio in Vancouver, working since 2014 across development, AI and performance marketing, which is exactly the mix this checklist demands: crawler access and schema are engineering work, answer-shaped pages are editorial work, and the measurement loop sits between them. One accountable team runs the audit, ships the fixes and reports on what the assistants say, with a working demo every week and every deliverable owned by you. Because we also build with these systems through our AI development practice, we optimise for how they actually retrieve and cite, not for folklore.

If you want to know where your site stands today, we will tell you plainly, including the parts that are already fine. Request a quote and we will reply within one business day.