You searched your company name on ChatGPT. Nothing. You asked it about the best B2B SEO agencies in Indonesia. Your competitors came up. You didn’t. Now you’re wondering what GEO strategy you need to fix it.
Before you go down that path, you need to understand something more fundamental: how LLMs actually know things. There are two mechanisms, and both of them lead back to the same root cause.
How LLMs get their information
Large language models have two distinct ways of knowing things about the world.
The first is training data. Before a model like ChatGPT is released, it is trained on an enormous snapshot of text from the internet — billions of web pages, articles, forums, documentation, and more. This is a one-time process with a cutoff date. Everything the model learned about the world came from what was on the web up to that point. If your company was not represented in that snapshot — because you didn’t exist yet, or because your web presence was too thin to be captured — the model simply does not know you exist. You were not part of its education.
The second is real-time retrieval. For current queries, most AI tools — ChatGPT with web browsing, Perplexity, Google AI Overviews — do not rely purely on training data. They call a search API, retrieve live web pages, and use those as the source for their answer. We covered exactly how this works in What is AI SEO, exactly? — the short version is that they are pulling from Google and Bing, which means your search ranking determines whether you get retrieved.
Your company is invisible on ChatGPT for one or both of these reasons. Let’s look at each.
The training data problem
Training datasets for large language models are built primarily from large web crawls — Common Crawl being the most significant. Common Crawl is essentially a snapshot of a large portion of the indexed web, taken periodically.
Two things determine whether your company made it into that snapshot.
Whether your site was properly indexed. Common Crawl pulls from the web the same way search engines do — it follows links, crawls pages, and captures content. If your site had poor technical SEO, thin content, weak authority, or was simply hard to crawl, it may have been skipped or captured poorly. The crawl quality of your training data representation is a direct function of your SEO. Companies with strong web presences and well-structured sites are over-represented in training data. Companies with poor SEO are under-represented or absent.
Whether you existed before the training cutoff. Every LLM has a knowledge cutoff — a date after which it has no training data. GPT-5.5’s cutoff is early 2026. Anything that happened after that date, any company that was founded after that date, any content published after that date — the model has no knowledge of it from training alone. If your company is relatively new or significantly rebranded recently, you may simply not be in the training data regardless of your SEO.
The fix for the first problem is obvious: better SEO means better web presence means better representation in future training datasets. The fix for the second problem is the retrieval layer — which brings us back to ranking.
The retrieval problem
Even if your company is not in the training data, you can still show up in AI-generated answers — if you rank well enough to be retrieved in real time. As explained in What is AI SEO, exactly?, AI tools with web browsing call search APIs and pass the retrieved pages to the model as context.
If you rank for the queries a user’s question generates, your page gets passed to the model. If you don’t rank, it doesn’t. The retrieval layer does not care about your training data representation — it cares about your current search ranking.
The two paths to appearing on ChatGPT both require the same thing: a strong enough web presence to be indexed, ranked, and found.
This is why companies that do good SEO tend to appear in AI answers, and companies that neglect it tend not to — regardless of which mechanism you look at.
What “too deep undercover” actually means
There is a third scenario that is less obvious. Some B2B companies have decent websites but almost no external web presence. No press coverage. No third-party mentions. No backlinks from relevant industry sources. No presence in industry forums, directories, or databases.
These companies may be indexed and even ranking for branded queries. But for general industry or category queries — the kind a B2B buyer would ask ChatGPT — they do not surface. The training data captured very little about them because there was very little to capture. The retrieval layer cannot find them for non-branded queries because they do not rank for anything beyond their own name.
This is the “deep undercover” problem. The solution is not a GEO tactic. It is building the web presence that should have been built anyway — content that earns links, coverage that generates mentions, and a B2B SEO strategy built around the queries your buyers actually use.
How to get your website on ChatGPT
There is no submission form. There is no “ChatGPT index” you can register with. The path is the same as it has always been — build a site Google wants to rank, produce content that earns authority, and make sure your pages are crawlable and properly indexed. Do that consistently and you will show up in AI-generated answers, because AI tools are pulling from the same search infrastructure you are already trying to rank in.
The only thing that is genuinely new is the query fan-out behaviour we covered in What is AI SEO, exactly? — writing niche content that covers the specific sub-questions an LLM generates when decomposing a broader topic. That is worth doing on top of your core SEO work. But it is not a replacement for it.
The practical diagnosis
If your company is not showing up on ChatGPT, run through this list before doing anything else:
- Does your site rank on Google for any non-branded queries in your category?
- Is your site properly indexed — are your key pages appearing in Google Search Console?
- Does your company have any third-party mentions, coverage, or backlinks from relevant sources?
- When did your company launch or rebrand — was it after the model’s training cutoff?
If the answers to the first three are no, the problem is not GEO. The problem is SEO. Fix that first. The AI visibility follows.
If the problem is recency — your company is simply too new to be in the training data — the retrieval layer is your path to visibility, which means ranking needs to be your priority now.
Either way, the answer is the same.
B2B SEO practitioner specialising in search strategy for the AI era. Working directly with marketing managers at mid-size companies — no account managers, no handoffs.