LLM Optimization: How to Get Cited by AI Without Buying the Hype

What LLM optimization means

LLM optimization (sometimes LLMO, sometimes GEO for generative engine optimization, sometimes AEO for answer engine optimization) is the practice of making your organization and your expertise easy for large language model products to find, understand and cite. The products in question are ChatGPT, Gemini, Claude, Perplexity, Microsoft Copilot and Google’s own AI features in search.

The labels differ mainly in who is selling them. The underlying job is the same: when someone asks an assistant a question your business can answer, or asks it to recommend a firm like yours, you want your pages to be among the sources it reads and your name to be in what it writes.

If you want the mechanics of how an assistant assembles a recommendation, we cover that in how AI search engines find and recommend businesses. This article is about the practice and the market around it: what to actually do, and how to tell a sound offer from a speculative one.

What is genuinely new

Four things deserve attention that a traditional SEO program might not have covered.

Letting the right crawlers in

Most assistants that answer with current information search the web and read pages as they go, using their own crawlers or a search partner’s. Several providers run separate crawlers for model training and for search, so you can decline one and allow the other. Many websites block AI crawlers without anyone deciding to: a security plugin, a hosting default or a content delivery network setting does it quietly. Check your robots file and your bot-protection settings, and make a deliberate choice for each provider.

Passages that stand on their own

Assistants quote and paraphrase passages, not whole pages. A paragraph that says “It depends on many factors” and then lists them three screens later is hard to use. A paragraph that answers the question in its first two sentences, names the place and the conditions, and then explains, is easy to use. This is a writing discipline, not a technical one: answer first, be specific, keep each section about one thing.

Mentions in the places assistants read

Assistants weigh what other sources say about you, and they read sources a traditional SEO program often ignored: comparison articles, industry association directories, local news, public forums, review platforms. A firm that appears only on its own website is a claim. A firm described consistently by an association listing, a news story and a body of reviews is evidence.

Checking what they say

Assistants can be confidently wrong: an old address, a service you dropped, a founder who retired, a competitor’s phone number. Asking the main assistants about your business on a schedule, and correcting the sources behind any errors, is now part of reputation management.

What is just SEO with a new label

Much of what gets sold as LLM optimization is search engine optimization that was always worth doing, which is fine, as long as you are not paying twice for it.

• Technical access: pages that load, render without heavy scripts, are indexed and are not locked in PDFs or images.
• Content that answers real questions: the questions buyers ask before they hire, answered by someone who knows the field. Google’s guidance has asked for helpful, people-first content for years.
• Clear service and location pages: what you do, for whom and where, stated plainly.
• Consistent business facts: name, address, phone, services and hours that agree across your site, your Business Profile and directories. See the business information AI search systems rely on.
• Structured data that matches the page: useful for helping machines read your facts accurately, as covered in our piece on structured data and AI citations. Useful, not decisive.

The practical point: assistants that search the web mostly read pages a search engine already considers relevant. If your site is invisible in ordinary search, no amount of AI-specific tuning will put it in front of the assistant.

Tactics with thin evidence

Some tactics circulate widely and deserve scepticism. Not all of them are harmful; most are simply unproven, and your budget is better spent elsewhere.

• Hidden instructions or text for AI. White text, hidden blocks or “notes to the AI” embedded in pages. This is manipulation, it can breach search engine spam policies, and it puts your reputation at risk if it is noticed.
• Special files that “talk to the AI”. Proposed conventions such as an llms.txt file are cheap to add, but we have not seen the major providers commit publicly to relying on them. Treat them as optional housekeeping, not a strategy.
• FAQ markup on every page. FAQ rich results are now limited to a small set of authoritative sites, and markup is not a citation switch. Write good answers in the visible page; mark up what is genuinely there.
• Statistics added for their own sake. It is often claimed that assistants prefer pages full of numbers. Invented or unsourced figures are a liability. Real data you own is valuable because it is original, not because it is numeric.
• Creating your own Wikipedia or Wikidata entry. These projects have notability and conflict-of-interest rules. A self-written entry for a regional firm is likely to be removed and can draw unwelcome attention.
• Volume content. Hundreds of machine-written pages targeting question variations. They tend to be thin, repetitive and easy for search systems to discount, and they bury the pages that matter.

How to evaluate an LLM optimization offer

You will be pitched. These questions separate a sound offer from a speculative one:

1. What exactly will you change, on which pages and which outside sources? A good answer names concrete work. A vague answer names a proprietary method.
2. What is the evidence that each tactic works? Platform documentation is good evidence. A vendor’s own case study of one client is weak evidence. “Everyone knows” is none.
3. How will you measure it? Be wary of a single “AI visibility score”. Assistant answers vary by phrasing, user and day. Honest measurement uses several imperfect signals, as we lay out in measuring AI search visibility without false precision.
4. Do you guarantee citations or placements? If yes, walk away. No one controls what an assistant says.
5. What happens to our ordinary search performance? Anything that would hurt your Google results to chase AI mentions is a bad trade for almost every business.

A worked example

Consider an illustrative environmental engineering consultancy with offices in Halifax and Moncton. When the partners ask assistants to recommend firms for contaminated site assessments in the Maritimes, they get larger national firms, a directory and occasionally one of their own former employees’ new companies.

A sensible LLM optimization program for them would look like this. First, check access: their hosting provider’s bot protection turns out to block several AI crawlers, so they decide which to allow. Second, rewrite the main service pages so each opens with a plain statement of what they assess, under which provincial frameworks, for which kinds of client. Third, publish three or four genuinely expert articles on questions their clients actually ask, with the regional detail a national firm would not bother with. Fourth, fix the association directory listings that still show a closed office, and make sure their Business Profiles agree with the website. Fifth, sample the same set of questions across the main assistants every month and note who is named.

Nothing on that list is exotic. The difference from a generic SEO plan is the attention to crawler access, quotable passages, and outside sources.

Where to start

Ask three or four assistants the questions your best clients would ask before hiring someone like you, and write down who is named and which sources are linked. Check whether those assistants can reach your site at all. Then fix, in order: access, the clarity of your main service pages, the consistency of your facts across directories, and the expert content that gives an assistant a reason to quote you. That sequence is the core of our SEO and AI visibility work, and it is the same sequence we would recommend to anyone doing it in-house. For the Google-specific side of the same question, see how AI Overviews are changing local search.