LLM optimization: the four reasons you are not cited

You ask an answer engine a question you should own. It writes a confident paragraph and cites two competitors. Nothing about your page was wrong — it simply was not chosen.

In our own classification, that outcome has four causes. They look similar from the outside and need completely different work.

1. The model cannot tell who you are

Naming a source is a small act of endorsement, and a model avoids endorsing something it cannot identify. If your pages never state plainly what the company is, what it does and where it operates, you are a page rather than an entity.

The fix is unglamorous: say it in plain sentences, say it the same way on every page, and make your structured data match the visible text rather than contradict it.

2. Your claims have nothing behind them

Models are tuned to avoid repeating assertions they cannot support. A statistic with no origin is risky to quote, so it gets dropped in favour of a source that shows its working.

Attach an origin to every number: who measured it, when, and how. Original measurement beats citing someone else's citation, which is why producing your own data is worth more than aggregating other people's.

3. The answer is there but cannot be lifted

Answers are assembled from fragments. A definition buried in the middle of a long paragraph is far harder to extract than the same sentence sitting alone under a heading that matches the question.

4. Nobody else says it about you

One site making a claim about itself is weak evidence. Models weight claims that appear in more than one independent place: a review, a forum thread, a video, a mention in someone else's article.

This is the slowest of the four to fix, because it does not depend on your website at all. It depends on other people having a reason to mention you — which usually means producing something worth mentioning.

Diagnosing which one you have

Ask the engine directly, then read what it did cite. If it cited a competitor's identical claim, your problem is corroboration or sourcing. If it paraphrased your content without naming you, your problem is identity or extractability. [internal: how citation misses are classified and fixed]