The first time a buyer asks ChatGPT "who is the best company for X" and your name isn't in the answer, that's uncomfortable. If it keeps happening, you have a strategy problem.
AI assistants do not pick three names at random. They pick based on signals, and only some of those overlap with traditional SEO. The signals are not secret, and the work to improve on them is largely the same work that makes a site better in other ways. Most sites just haven't done it yet.
This piece covers what moves the needle, what is hype, and how to start measuring whether anything is working.
How AI assistants decide who to cite
The mechanics vary by assistant, but the broad shape is consistent.
When a user asks a question, the assistant either pulls from its training-data-era memory (what it learned about your space during its last training run), retrieves fresh content via a real-time search or browsing tool, or both. The "who to cite" decision draws on the trustworthiness of the source, the structural clarity of the page, the specificity of the answer to the question, and corroboration from other sources.
In practice, a few things now carry more weight than they used to:
· Pages that lead with the answer instead of burying it under 300 words of preamble.
· Structured data that tells the assistant what the page is about without it having to guess.
· Authoritative-looking sites with a clear identity, real Organization schema, consistent name and contact info, and visible expertise.
· Content shaped around the questions buyers ask rather than keyword phrases.
The five things that move the needle
1. Real Schema.org JSON-LD on every important page. That means Organization on the homepage, Service on each service page, Article on each blog post, FAQPage where you have real Q&A, and BreadcrumbList everywhere. Most sites still have none of this, or have it implemented sloppily through a plugin that fights with the theme. Done correctly, structured data is the strongest citation signal we have measured. We go into it in more detail in the schema markup worth adding in 2026.
2. Answer-first content structure. When you write a piece that answers a question, put the answer in the first sentence or two, then explain. This is uncomfortable for writers who learned to "set up" an answer with context. AI engines do not read that way. They extract the most direct version of the answer and quote it, so lead with the part you want quoted.
3. Clear, single-topic pages. One question per page, answered well, beats a comprehensive guide that tries to cover ten things. The "comprehensive guide" still has its place for human readers, but for citation, the specific page usually does better. A page titled "What is managed WordPress hosting?" that answers that exact question in the first 200 words gets cited more than a 4,000-word ultimate guide that mentions managed hosting in passing.
4. A real Organization identity. Use one name, one physical or operational base and one website, and keep them consistent everywhere you appear: your site, your structured data, your social profiles and directories. AI engines cross-check these to judge authority. A site that looks like one coherent business gets treated as one, and a marketing site that could belong to anyone gets treated as anonymous.
5. llms.txt at the root. A small plain-text file that tells modern AI crawlers where your highest-signal content lives. Adoption is still uneven, but the cost of adding one is trivial. It is the AI equivalent of a sitemap that says "if you're going to learn about us, start here."
What does not move the needle
A few things get more attention than they deserve.
Keyword density. Stuffing your service name into every paragraph does not get you cited. It signals "low-trust content" to engines that have spent five years learning to ignore it.
AI-generated content at scale. Publishing 80 AI-written articles a month may briefly improve impressions. It almost always hurts citation rates over time, both because the content reads as low-trust and because the search engines themselves have gotten quite good at detecting it.
"AI SEO" tools that promise citation in a week. The things that get a site cited build up over months. Tools that promise faster results are usually either selling tracking dashboards, automated content generation, or both.
Backlinks alone. Links still count for traditional ranking, but for AI citation they carry less weight than the structural and authority signals above. A page with no backlinks and clean structure can get cited; a page with great backlinks and bad structure often does not.
How to measure it
Real measurement is mostly manual right now. The honest, repeatable workflow:
Pick the 10-20 questions a buyer would ask about your space. Be specific. "Best managed WordPress host for agencies" is a real question; "WordPress hosting" is not.
Ask each question across at least three assistants: ChatGPT, Claude, and Perplexity is a reasonable starting set. Note which sources get cited, and repeat it once a month so you can watch the trend.
Some tools automate this. They are useful as a directional signal, especially for tracking a larger set of queries, but they don't replace the manual workflow. Checking by hand makes you read what the assistant actually said about your space, and that is often more useful than the citation counts.
What this looks like in a real engagement
When we take on AI visibility work for a site, the first month is mostly fundamentals. We audit the structured data, review the content structure on the highest-value pages, add real Organization schema to the homepage and llms.txt where it's missing, and draw up a fix-list for the technical SEO basics that AI engines also rely on.
The second month is content shape: rewriting the top 10 pages to lead with the answer, adding FAQPage schema where there is real Q&A, and tightening the internal linking so the engines can build a coherent picture of what you do.
By the third month, manual citation checks usually start showing movement. Not always; sites with thin authority or in extremely competitive spaces take longer. But for most business sites in defensible niches, the work compounds visibly within a quarter.
This is the kind of work our SEO & AI Visibility service is built around, with honest reporting and without vanity dashboards or AI-content-at-scale shortcuts. If you want a read on where you currently stand and what the realistic path forward looks like, get in touch and we'll give you a straight answer.
