LLM Optimization getting AI assistants to name you.

LLM optimization (LLMO) is the work of getting AI assistants built on large language models, such as ChatGPT, Gemini and Perplexity, to mention, recommend and cite your brand when people ask them questions. Prefer measures whether that work is paying off, tracking ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode and showing the sources behind each answer. The term is newer and less formal than its cousins; most people use it interchangeably with generative engine optimization.

How LLM optimization works#

A large language model can answer in two ways. It can answer from memory, using what it learned in training data, or it can search the web first and write the answer from what it finds. The second path is called grounding, and it is the one you can influence today.

When an assistant grounds an answer, it runs searches, reads the pages that come back, and quotes the passages that answer the question best. OpenAI says ChatGPT decides when to search the web and cites the sources it uses, and it can only reach pages its search crawler, OAI-SearchBot, is allowed to crawl. Google describes the same pattern for AI Overviews and AI Mode. So LLM optimization comes down to three questions:

• Can the engine reach your pages? Its search crawler must be allowed in robots.txt and able to read your content.
• Is your passage the best answer? Clear, self-contained, factual passages are easier to lift and quote.
• Do trusted sources name you? Engines lean on third-party pages, such as reviews, listicles and forums, when they recommend brands.

Why it matters for AI search visibility#

A buyer who asks ChatGPT “what is the best tool for X” gets a short list, not ten blue links. If you are not on that list, you are not in the running. LLM optimization is the work of earning a place on the list and keeping it.

Google’s own guidance says that, from its point of view, optimizing for generative AI search is “still SEO”. That is a useful anchor. Crawlability, helpful content and real authority still do the heavy lifting. What changes is the unit of competition: you compete for a passage and a mention inside one answer, not a position on a results page.

What the research says#

The best-known study is the 2023 paper “GEO: Generative Engine Optimization” by Aggarwal et al., accepted to KDD 2024. It tested content changes on a benchmark of queries and reported that the right methods could boost visibility in generative engine responses by up to 40%. Adding citations, quotations and statistics helped. Keyword stuffing did not.

What LLM optimization involves in practice#

The work splits into on-page and off-page moves. Both matter, because engines quote your pages and also quote what others say about you.

On your own site:

• Let the search crawlers in. Allow OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot and Bingbot in robots.txt. Training crawlers such as GPTBot are a separate choice.
• Lead each section with the answer. Put a short, self-contained answer at the top, then the detail. That is the passage a model is most likely to lift.
• State facts plainly. Prices, specs, dates and names in plain text, kept current, so a retrieved passage is right.
• Cover the questions buyers ask. Comparison, alternative, pricing and “best for” pages match how people prompt assistants.

Off your site:

• Get named where engines look. Find the listicles, reviews and threads already cited for your prompts and earn an accurate place on them.
• Keep your entity consistent. The same name, category and core facts everywhere help a model connect mentions to you.

How to measure it#

LLM optimization without measurement is guesswork. Answers vary by engine, by day and even between two runs of the same prompt. A useful setup:

• A fixed prompt set. The questions your buyers ask, from broad category prompts to direct comparisons.
• Several engines. ChatGPT, Gemini, Perplexity and Google’s AI features often disagree, so track each one.
• Repeat runs. Weekly checks show direction. One screenshot shows a moment.
• The right metrics. Whether you are mentioned, whether you are cited, your position, and your share of voice against named competitors.

Common confusions#

• LLMO vs GEO vs AEO. Same goal, different labels. GEO comes from the paper above; AEO is the older term from featured snippet work. Our SEO vs AEO vs GEO answer covers the differences.
• LLMO vs training the model. You cannot edit what a model learned. You can only shape what it retrieves and what the web says about you over time.
• LLMO vs llms.txt. An llms.txt file is one small tactic, not the whole practice. Google says Search ignores llms.txt files.

Example#

A project management tool is missing when buyers ask ChatGPT for “best project management tools for agencies”. The answer cites three listicles and a Reddit thread. LLM optimization here means publishing a clear page on agency use cases, getting included in the listicles ChatGPT already cites, and checking weekly whether the brand starts appearing.

How Prefer helps#

Prefer tracks your buyer prompts across five engines, shows which sources each answer cites, and measures your share of voice against named competitors. Every self-serve plan also writes AI articles built to be cited. Check your AI visibility for free to see where you stand today.