LLM SEO in one paragraph
An AI assistant answers a buyer's question in two steps. It runs one or more web searches, then writes an answer from the pages it fetched, naming brands and linking some of those pages. LLM SEO works on both steps: being fetchable and findable for the searches the engine runs, and being named on the pages it reads. Most of those pages are not yours, so LLM SEO is as much about where you are discussed as about your own site.
LLM SEO vs SEO: what changes
| SEO | LLM SEO | |
|---|---|---|
| Who reads the page | Googlebot, then a person on a results page | AI crawlers such as OAI-SearchBot, PerplexityBot and ClaudeBot, then a model writing an answer |
| What you win | A ranked link | A mention by name, a cited link, or both |
| The query | The words a person typed | The questions a person asked plus the searches the engine ran for them, often several per answer |
| Pages that do well | Pages that match intent and earn links | Pages with a direct answer up top, prices, comparisons and dated facts, plus third-party threads and lists |
| What you measure | Rankings, clicks, impressions | Named share and citation share per engine, AI referral visits, AI crawler requests |
| Where most of the work happens | Your site and its links | Your site and the sites the engines cite about your category |
What our own logs and scans show
We run CommunityMentions and measure it the same way we'd measure a client. Three numbers from the last two weeks:
- AI crawlers already outnumber Google. From October 31 to November 6, 2026, our Vercel logs recorded 1,602 requests from OAI-SearchBot, OpenAI's search crawler, against 324 from Googlebot. Claude-User, which fetches a page while someone is chatting with Claude, made 266 requests and ChatGPT-User made 150.
- The engines search before they answer. In our November 2, 2026 scan of 226 answers on ChatGPT, Claude, Gemini and Perplexity, the engines ran 622 web searches, about 2.75 per answer. Those searches, not only the question the user typed, decide which pages get read.
- Being named and being cited move apart. In the same scan CommunityMentions was named in 17 of 96 category answers, but only 30 of the 804 URLs those answers cited were our pages (3.7%). Engines often name a brand while citing someone else's page about it.
How to do LLM SEO: the steps
| Step | What to do | Done looks like |
|---|---|---|
| 1. Open the door | Check robots.txt and your firewall or bot protection for OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot and Claude-User. A crawler you block can't read the page it would cite. | Each one gets a 200 on your key pages; your server logs show their requests |
| 2. Write the buyer questions | List 20 to 30 questions buyers ask about your category without your brand name: "best X for Y", "X pricing", "X alternatives". | A fixed prompt list you will rerun each month |
| 3. Measure a baseline | Ask each question on each engine and record who is named and which URLs are cited. Our AI visibility guide has the full method. | Named share and citation share per engine |
| 4. Publish the pages the engines search for | One page per buyer question, titled in the buyer's words, with a 40 to 60 word answer first, a comparison table, real prices with the date read, and a short FAQ. | A page that answers the question in its first lines and is in your sitemap |
| 5. Get discussed where answers look | Find the threads, videos and lists the engines cite for your questions and take part where it's allowed: Reddit threads, YouTube comments, LinkedIn posts, review sites, roundups. | Your brand named, in context, on pages the engines already cite |
| 6. Re-measure monthly | Rerun the same prompts on the same engines and compare each engine with itself. | A monthly trend line per engine, not a one-off score |
Why Reddit sits at the centre of LLM SEO
Of 78,023 answers from OpenAI GPT models, Perplexity Sonar and Gemini models that we tracked between January 9 and October 30, 2026, 18,549 (23.8%) linked at least one Reddit thread. In our study of 41,592 AI search queries, reddit.com was cited in 18.9% of the answers that cited any source, ahead of YouTube at 14.4%. The subreddits that get cited are a short list: in our list of 1,313 cited subreddits, the top 10% took 58.6% of the citations. Use the subreddit finder to see which ones cover your category, and how to choose subreddits for AEO for the shortlist method.
What doesn't move it on its own
- llms.txt. It is cheap to add and we publish one, but we treat it as hygiene, not a lever: it tells engines where your pages are, not why to pick them.
- Schema markup alone. Structured data helps engines read a page; it doesn't make a page the one they pick.
- Branded prompts. Asking "what is <your brand>" tells you little: when the question had our name in it, the answers named us first 47 times out of 48. Measure on questions a buyer asks before they know you.
- One run. The same prompt can name different brands an hour later, so read a monthly trend, not a single answer.
Doing it yourself or having it done
Steps 1 to 4 are site work your team can do. Step 5 is the slow part: being discussed, in context, every week, on the threads the engines cite. CommunityMentions does that part. We post comments that name your brand on Reddit, YouTube, Medium, LinkedIn and X, and aim for 5 to 10 a day across those channels combined. In the 30 days from October 7 to November 5, 2026, the median of 55 client brands got 49 verified placements across those channels, 9 of them on Reddit. The price is $1,000 a month, month to month: a $1 setup charge today, and monthly billing starts only after you approve the voice in Slack. No service can promise that an engine will cite a given comment. Before you pay anything, get a free sample of the comments we would post for your site.
Sources
- CommunityMentions Vercel request logs, AI and search crawler user agents, October 31 to November 6, 2026, read November 7, 2026.
- CommunityMentions AI-answer scan, November 2, 2026: 226 answers on ChatGPT, Claude, Gemini and Perplexity, 622 web searches, 96 category answers.
- CommunityMentions subreddit citation data, January 9 to October 30, 2026, as published on the subreddit list.
- CommunityMentions citation study, 41,592 queries, January 9 to July 14, 2026, published July 20, 2026.
- Our price and terms: pricing section and company facts, read November 7, 2026.