# LLM SEO: how do language models choose which brands to mention? Source: https://outname.now/guides/llm-seo Updated: 2026-10-07 By: Outnamed (https://outname.now) LLM SEO is the work of making your company easy for large language models to find, understand and name correctly. Models draw on what they learned in training and, in many assistants, on web pages fetched while answering. How they pick between brands is not public, differs per assistant and changes over time. ## Key points - LLM SEO means improving how large language models find, describe and name your company. - A brand can reach an answer through two routes: the model's training memory or pages retrieved at answer time. - The weighting the assistants apply is not public, so any claim to know the formula is a guess. - You can control your own facts, your pages and your consistency across sources, but not the outcome. - One answer proves nothing, because assistants rarely give the same list twice. ## What is LLM SEO? A large language model (LLM) is the kind of system behind ChatGPT, Claude, Gemini and Perplexity. It writes answers by predicting text, based on patterns learned from very large amounts of writing. LLM SEO is the practice of shaping what those systems find and say about your company, so that when a buyer asks for a recommendation, you are described accurately and have a chance of being named. It borrows the name from search engine optimization, but the goal differs. A search engine returns a list of links. An assistant writes one answer, and your company is either in it or not. ## How do language models decide which brands to name? There are two routes into an answer. The first is training memory. During training, the model absorbs text from a huge range of sources. If your company was described often, clearly and consistently in that text, the model may have a usable picture of it. That picture is fixed until the provider releases a new model version, and the timing is not public. The second route is retrieval. Many assistants can search the web while they answer, and then write from the pages they find. ChatGPT, Claude, Gemini and Perplexity all offer this in some form. Which searches they run, which pages they read and how much weight those pages get is not public. It differs per assistant, and it changes over time. So the honest answer is this: models name brands that they either remember or can find, and the exact choice between two similar brands is not something anyone outside the providers can state. ## Training memory or live search: what is the difference? The two routes call for different work. One is slow and indirect. The other responds to what is on the web now. ## What can you influence, and what can't you? You cannot edit a model, pay for a place in its answers or force a mention. You can control the material it has to work with. - Within your control: the facts on your own site, such as what you sell, for whom, in which regions, with which limits. - Within your control: whether each buying question has a page that answers it directly. - Partly in your control: whether your description matches across your site, review profiles and directories. - Partly in your control: the reviews and discussions about you on sites such as G2, Capterra, Trustpilot, Google reviews, Yelp and Reddit. You can ask happy customers to review. You cannot script what they say. - Not in your control: which model version is live, which pages an assistant retrieves, and how it weighs them. ## How do you start with LLM SEO? Start with measurement and your own facts. Tactics come after. 1. Write down ten to twenty questions a buyer would ask when choosing a company like yours, in their words, not yours. 2. Ask each question in several assistants, with web search on and off, and record which companies are named and what is said about yours. 3. Check what is wrong or missing: outdated prices of competitors do not matter, but your own wrong facts do. Fix them at the source first. 4. Publish one clear page per buying question on your own site, with concrete facts: who it is for, what it costs to start, what it does not do. 5. Align your profiles on review sites and directories with those facts, and repeat the check every month. ## Why does one answer tell you nothing? Assistants generate text, and the same question can produce a different list on the next try. A screenshot where you appear, or do not, is a single draw from a spread of possible answers. The useful number is frequency: how often your company is named across many runs of the same question. Outnamed measures this by asking each buying question thirty times across the assistants through their official programming interfaces, with web search on, and counting who is named. You can do a smaller version by hand, as long as you repeat each question several times and record every result. ## Is LLM SEO worth the effort? The evidence is early and mixed in what it shows. In the G2 Buyer Behavior Report from June 2026, covering more than 1,000 software buyers, 8 in 10 software buyers used an AI chatbot to find recommendations in the last two years. The same G2 report found that 37% say chatbots shape their shortlist most, level with review sites at 38%. Traffic numbers can look small while value is not. Ahrefs reported for its own traffic over 30 days that visitors from AI search were 0.5% of traffic and 12.1% of sign-ups. That is one company's data, not a rule for yours. It does suggest checking sign-ups, not only visits. **The two routes by which a brand can reach an assistant's answer** | Route | What it draws on | How it changes | What you can do | | --- | --- | --- | --- | | Training memory | Text the model learned from before its release | Only with a new model version; timing is not public | Keep your facts consistent and widely stated over time | | Live retrieval | Web pages found while answering, when search is on | Whenever pages or the assistant's search behavior change | Publish clear, factual pages that answer buying questions | ## Questions people also ask ### Is LLM SEO the same as regular SEO? No, though they overlap. Regular SEO aims at a position in a list of links. LLM SEO aims at being described correctly and named inside a written answer. Clear, factual pages help in both, but nobody can say how much the two overlap for a given assistant. ### Does ranking on Google make an assistant mention me? Not necessarily. Some assistants search the web while answering, but which search source they use is not always public and can change. A good Google position may help your pages get found. It does not guarantee a mention. ### How long does it take to appear in AI answers? Nobody knows. Retrieval can pick up new pages sooner than training memory can change, but neither timing is published, and no result is guaranteed. Measure monthly and judge the trend rather than a single day. ### Do reviews and Reddit threads matter? They can. When people ask for recommendations, review sites and community discussions are common places for an assistant to find opinions. How much weight any assistant gives them differs and is not public. Accurate, recent reviews are worth having anyway. ### Can I pay to be included in an assistant's answer? Not by changing the model's answer on request. The assistants decide what to name. Be wary of anyone who promises a mention or a position. ### How do I check how often my company is named? Ask the same buying question many times, in several assistants, and count how often your company appears. A handful of tries gives a rough picture. Thirty runs per question gives a steadier one. Sources for the figures on this page: G2 Buyer Behavior Report, June 2026 (1,000+ software buyers); Ahrefs, own traffic over 30 days. These are their figures, not a forecast for your company.