LLM SEO is the work of making your brand and your pages easy for large language models such as ChatGPT and Gemini to find, trust and name in their answers. It has the same goal as GEO (generative engine optimization) and AEO (answer engine optimization): three names for one job.
A large language model, or LLM, is the type of AI that writes the reply in ChatGPT, Gemini, Perplexity and Google AI Overviews (the AI summary at the top of some Google results). Classic SEO gets a page ranked in a list of links. LLM SEO gets your brand written into the reply itself, usually as one name in a short list of recommendations, with a few source links underneath.
This guide covers what LLM SEO is called, how LLMs choose which brands to name, seven practical steps, what our study of 480 AI answers from India shows, and how to measure progress. The table below puts classic SEO and LLM SEO side by side.
| Aspect | SEO | LLM SEO |
|---|---|---|
| What you win | A position for your page in a list of links | Your brand named, or your page cited, inside a written answer |
| Where it shows | Google and Bing results pages | ChatGPT, Gemini, Perplexity and Google AI Overviews |
| What helps | Relevant pages, links from other sites, a site search bots can crawl | Pages AI bots can read, direct answers, mentions on the sites AI cites, the same brand facts everywhere |
| Same query, next try | One search returns one ranked list | The same question can return a different set of brands |
| How you measure | Rank, clicks and impressions, for example in Google Search Console | Share of answers that name you, your place in the list and the pages cited, across many runs |
| Typical tools | Search Console and rank trackers | Tools that ask AI engines your buyer questions on a schedule (compared in our AI visibility tools roundup) |
What is LLM SEO called?
LLM SEO goes by at least five names, and they describe the same work: getting AI answers to mention and cite you. Which label a team uses depends mostly on where they first read about it.
- GEO, generative engine optimization. The research name. It comes from a paper by Aggarwal et al., presented at KDD 2024, which tested which page changes make generative engines include a source. Read more in what is GEO.
- AEO, answer engine optimization. The name many marketing teams use. It treats ChatGPT, Perplexity and AI Overviews as answer engines. See what is AEO.
- LLM SEO or SEO for LLMs. The name that keeps the SEO label and makes clear that the reader is a language model.
- LLM optimization, or LLMO. The same idea with the word SEO dropped.
- AI SEO or AI search optimization. The broadest label. Some people also use it for SEO work done with AI tools, which is a different thing.
Is LLM SEO different from SEO?
LLM SEO builds on SEO and changes what counts as a win. SEO asks where your page ranks for a search. LLM SEO asks whether an AI answer to a buyer question names your brand, how high in the list, and which pages it cites as proof.
Three things change in practice. First, the answer is a summary, so the buyer may never click through, and being named becomes the result that counts. Second, the answer is built from several pages at once, so what other websites say about you counts alongside your own site. Third, the answer changes between runs, so one check tells you little and you need many runs to see a trend.
Plenty stays the same. A page search bots cannot reach will not show up in either place. Pages that answer the question early and clearly do well in both. For a deeper split of the labels, read AEO vs GEO vs SEO or the shorter GEO vs SEO comparison.
What does "LLM in SEO" mean?
The phrase "LLM in SEO" has two meanings. The first is the topic of this page: optimizing so that LLMs name and cite you. The second is using an LLM as a helper inside classic SEO work, for example to group keywords, draft outlines or write meta descriptions.
The two meanings meet in one place. If you use an LLM to draft pages, those pages still need facts, numbers and sources that a person has checked. The GEO research linked below found that adding statistics, quotations and citations helped pages appear in AI answers. Those have to come from real data.
How do LLMs choose which brands to name?
An LLM names a brand for one of two reasons: it learned about the brand during training, or it found the brand in web pages it fetched while writing the answer. For buying questions with web search switched on, the fetched pages play a large part, so what an engine can find today matters a lot.
Our India study shows how often search is involved. ChatGPT with web search and Perplexity attached at least one source to every one of their 320 answers. Gemini attached sources to 90.0% of English answers.
Training data: what the model already knows
A model is trained on a large snapshot of public text that stops at a cutoff date. If your brand was written about widely before that date, the model may name you even with search switched off. You cannot edit training data. You can only add to the public record that later models will learn from, which is one more reason to earn mentions on other websites.
Live search: what the engine reads right now
With web search on, the engine turns the question into one or more searches, reads some of the pages it finds, and writes an answer that quotes or links a few of them. This is where classic SEO still pays off, since a page that ranks for those searches is easier for the engine to find. Google AI Overviews are built from pages in Google's own index, which Googlebot crawls.
Cited pages can shape the brand list
An answer can repeat brands named in the pages the engine read, so the websites it cites are worth knowing. In our study, reddit.com was cited in 106 of the 480 answers, more than any other website, and youtube.com in 100. Forums, videos and review blogs feed AI answers alongside brand websites.
If ChatGPT names your rivals and skips you, our guide on why ChatGPT doesn't mention your brand walks through the common causes.
Crawler access decides what can be read
AI companies fetch pages with named bots. OpenAI, for example, runs GPTBot to collect training data and OAI-SearchBot to find pages for ChatGPT search. Your robots.txt file tells each bot which pages it may visit. If it blocks a search bot, that engine cannot read your pages when it answers. Pages behind a login, a paywall or a form are hidden for the same reason. Our guide to AI crawlers lists the bots and what each one does.
LLM SEO in practice: 7 steps
These seven steps cover the core of LLM SEO. Step 1 makes your site readable, steps 2 to 6 give AI engines better material to quote, and step 7 tells you whether the work is paying off.
1. Let AI search bots read your site
Open yourdomain.com/robots.txt and check the rules for the bots that fetch pages for live answers: OAI-SearchBot for ChatGPT search, PerplexityBot for Perplexity, and Googlebot for Google Search and AI Overviews. Blocking a training bot such as GPTBot is a separate decision. It limits future training on your pages and does not remove you from ChatGPT search.
Also check that your key pages load for these bots. A 403 error or a bot challenge screen shuts them out the same way a robots.txt block does. Depra's AI crawler check tests 41 AI crawlers from 26 vendors against your robots.txt and your competitors' files, and shows the exact rule behind each verdict.
2. Put the answer in the first two sentences
AI engines lift short passages that make sense on their own. Open each page, and each section, with a direct answer to the question in its heading, then add the detail. Use headings shaped like the questions buyers ask, tables for comparisons, and numbers with a named source.
Specific, sourced writing has research behind it: adding citations, quotations and statistics raised a page's visibility in AI answers by up to about 40% in a GPT-3.5 test setup (Aggarwal et al., KDD 2024).
3. Get listed on the sites AI already cites
Find the pages engines cite when they answer your category's buying questions. Expect "best X" lists, comparison articles, review blogs, marketplaces, Reddit threads and YouTube reviews. Then earn a place on them: pitch the author of the list, offer a product for review, claim your marketplace or directory listing, and answer relevant forum threads openly as your brand.
Depra's citation gap analysis lists the websites that recommend your competitors but never mention you, down to the exact page, with competitors' own sites filtered out.
4. Keep your brand facts the same everywhere
Engines combine what many pages say about you. When your price range, product line, category or founding story differ between your site, marketplace listings and press coverage, the model has less reason to state any of them with confidence. Write one short fact sheet (what you sell, who it is for, price range, where you ship) and use the same wording on your About page, marketplace listings, social profiles and press kit.
5. Publish the comparisons and lists buyers ask for
Many buying questions are comparisons: "X vs Y", "best X under ₹2,000", "X alternatives". Publish fair pages in those formats on your own site, and name rivals where they win. Engines can choose from many neutral pages, so a page that only praises you gives them little to quote.
Depra's AI action plan ranks up to 12 page formats by how often AI cites them in your category, such as listicles, comparisons and how-to guides, next to how many of each you already have.
6. Cover the languages your buyers use
In India many buyers type in Hinglish: Hindi-English written in English letters, such as "sabse accha protein powder kaunsa hai". In our study, Perplexity's brand lists changed far more between English and Hinglish than between two runs of the same question. If your buyers ask in Hinglish, check what engines say to the Hinglish version before you assume your English results carry over.
7. Measure across many runs
A single chat tells you little, because the same question can return different brands on the next try. Track a fixed set of buyer questions on each engine over several weeks, and watch three things: the share of answers that name you, where you appear in the list, and which pages get cited. The measurement section below explains how.
What our India study shows about LLM answers
LLM answers change from one run to the next, and the language of the question changes them further. That is the short result of a study Depra ran on 14 August 2026: 10 buying questions (5 skincare, 5 fashion), each asked in English and in Hinglish, 8 times, on ChatGPT with web search, Gemini and Perplexity, all from inside India. That made 480 answers.
To measure change, we compared the brand lists of two answers and took the share of brands they had in common, out of all brands either answer named. 100% means identical lists and 0% means no brand in common. The Rerun overlap column compares reruns of the identical question. The English vs Hinglish column compares the English answer with the Hinglish one. The Gap column subtracts the second from the first, in percentage points. For Perplexity, 68.3% minus 44.9% gives a gap of 23.4 points. That gap is the extra change that comes from language alone.
| Engine | Rerun overlap | English vs Hinglish | Gap |
|---|---|---|---|
| ChatGPT | 54.7% | 52.5% | 2.2 points |
| Gemini | 51.2% | 43.3% | 7.8 points |
| Perplexity | 68.3% | 44.9% | 23.4 points |
Gemini attached sources to 90.0% of English answers and 41.3% of Hinglish answers. (Depra English vs Hinglish study, 160 Gemini answers, 14 Aug 2026)
What the numbers show
- Reruns differ. On ChatGPT and Gemini, two answers to the identical question shared only about half of their brands (54.7% and 51.2%).
- Language moves some engines far more than others. Perplexity's English and Hinglish lists overlapped 23.4 percentage points less than its reruns did (44.9% against 68.3%). ChatGPT barely moved, at 2.2 points.
- Sources drop in Hinglish on Gemini. Gemini's Hinglish answers were also shorter and carried fewer source links on average (5.2 against 9.4 in English).
What this means for LLM SEO
Three lessons follow. Judge LLM SEO on many runs, because a single answer can differ from the next one. Check each engine on its own, because they react differently to the same change. And if your buyers ask in Hinglish, measure that version too, because English results do not carry over on every engine.
How to measure LLM SEO
Measure LLM SEO by asking a fixed set of buyer questions to each AI engine on a schedule and counting how often the answers name you. Four measures cover most of what you need.
- Visibility. The share of answers that name your brand. If 40 of 200 answers name you, your visibility is 20%.
- Share of voice. Your mentions as a share of all mentions of the brands you track, so you can see who AI names instead of you.
- Position. Where you appear when the answer lists options. 1 means you were named first.
- Cited pages. The websites and pages the engine links to, and whether the answers citing them named you.
Pick questions that do not contain your name
A question like "is [your brand] any good" names you by default and inflates the score. Use the questions a buyer asks before they know you: "best face serum for oily skin in India", "affordable running shoes for beginners". Keep the set fixed for a few weeks, so a change in the numbers is not caused by editing the questions.
For a manual method with a spreadsheet, read how to measure AI visibility.
Track it with Depra
Depra runs this for you. It asks your buyer questions to ChatGPT with web search, Gemini and Google AI Overviews every day, and to Perplexity every week, from inside your chosen country, in English and Hinglish. It stores every full answer and records whether you were named, in what position, in what tone, and which pages were cited. Questions that already contain your name are left out of the score.
See AI visibility tracking for how the numbers are built. Plans start at ₹1,999 a month plus GST, with 7 days free on any plan and no card.
What can LLM SEO not promise?
No one can guarantee that an AI engine names your brand. Answers vary from run to run, engines change their models and search sources without notice, and each engine weighs pages in its own way. Be wary of anyone who promises a fixed spot in ChatGPT.
What LLM SEO can do is raise the odds. Readable pages, clear answers, mentions on the sites engines cite and consistent facts give every engine more reasons to name you. Measuring over many runs shows whether those odds are moving.