LLM SEO agency
SEO Back Office is a GEO agency. This page is about optimizing for the large language models themselves, and about what a brand can and can't influence once one of them starts writing an answer.
What is LLM SEO?
LLM SEO is the work of getting a brand into the answers large language models write, by making sure the sources those models reach for say the right things about you. A model doesn’t rank you. When someone asks it a question it either retrieves pages and summarises them or it writes from what it absorbed in training, and usually it does some of each. You can influence the retrieval side directly, so that’s where the work sits.
The agency version of this is a GEO agency engagement under a different name. ChatGPT, Claude, Gemini and Perplexity each run their own crawler and their own index, and those differences matter for access, but the job reads the same everywhere. Rank, and be on the pages they pull from.
Buyers search for this under four or five names and the volume is spread across all of them. LLM SEO, AI SEO, GEO, AEO. The engagement underneath is one engagement, and we’d rather explain that than sell you four of them.
How retrieval works at answer time
OpenAI’s crawlers have separate jobs. GPTBot, announced in August 2023, collects training data. OAI-SearchBot builds the index ChatGPT search reads from. ChatGPT-User fetches a page live when someone’s question sends it there. Blocking one doesn’t block the others, and plenty of sites have blocked the wrong one.
Google works differently. AI Overviews and AI Mode read Googlebot’s index, so your ordinary Search performance carries straight into them. Google-Extended, added in September 2023, only controls whether your content trains Gemini. Switching it off doesn’t take you out of AI Overviews, and that confusion costs sites more than any other robots.txt mistake we see.
Perplexity has cited its sources inline since it started in 2022 and crawls with PerplexityBot. ClaudeBot is Anthropic’s. Applebot-Extended is Apple’s version of the training toggle. None of this is exotic. It’s a robots.txt review and a log check, and it takes an afternoon.
What you can influence and what you can’t
You can’t touch the weights. Whatever a model absorbed about your company during training is fixed until the next one trains, and no agency has a lever on it. You can’t make a model name you on a given run either. Ask the same question twice and you’ll get two different lists.
What’s in reach is the retrieval layer. That means the pages a model fetches for your category mention you and describe you correctly, your own pages stay crawlable and rank for the queries sitting behind the prompt, and the description of your company reads the same on your site, on LinkedIn, on the review sites your category uses and in the directories that rank.
The training cutoff sits outside your reach as well, and it’s the part people ask about most. If your company launched after a model’s cutoff and nothing on the live web has caught up, that model has to retrieve to know you exist at all, which makes the retrieval work most of the job rather than half of it.
Anyone selling a guaranteed slot in a ChatGPT answer is selling something they can’t deliver. What we can do is move the inputs and show you the log.
What we run for LLM SEO
The audit builds the baseline. Forty to eighty queries drawn from your buyers’ real questions, run across ChatGPT, Perplexity, Gemini, Claude and Google’s AI features over two to three weeks, with four fields logged every time. Whether you were named, where in the answer you appeared, which pages the answer cited, and how you were described.
The citation list that falls out of it becomes the outreach plan. We work it publisher by publisher, getting you added where you’re missing and corrected where the entry’s stale. While that runs, the pages on your own site get built to be quotable, the schema gets corrected, and crawler access gets confirmed rather than assumed.
We also check the failure that costs sites the most, which is content rendering only in JavaScript. Several of these crawlers fetch the HTML and move on, so a page whose main text arrives after a script runs can sit in the index as an empty shell while you wonder why nothing cites it.
Then the same query set runs every month, so movement means something. The rest of the service list is on the GEO services page, and ChatGPT SEO goes deeper on the platform most buyers ask about first.
Get on the pages the models read.
Send us your category and your three closest competitors. We’ll run the first round of queries before the call, so the conversation starts from the answers rather than from a pitch deck.
Request an AI visibility audit No inflated promises. A clear baseline and a practical roadmap.