LLM SEO starts with the searches the model runs
Assistants with web search do not answer from memory alone. ChatGPT, Perplexity and Claude rewrite the question into searches and read the results. A page that does not rank for those searches is rarely read, and a page that is never read is never cited. So the first job of any LLM SEO tool is to show those searches, and where you rank for them.
The content levers with evidence
The most cited controlled study of generative engine optimisation (Aggarwal et al., KDD 2024) tested nine ways of rewriting a page. Three raised visibility in generated answers; one lowered it:
| Change to the page | Effect on visibility in answers |
|---|---|
| Adding quotations from credible sources | about +41% |
| Adding statistics | about +32% |
| Citing sources | about +30% |
| Keyword stuffing | about −10% |
Source: P. Aggarwal et al., "GEO: Generative Engine Optimization", KDD 2024 (arXiv:2311.09735). Effects measured on position-adjusted word count in their benchmark; your results will differ.
What the tool does not do
- It never promises rankings or mentions. Engines change their models; we mark every change on your charts.
- It never recommends keyword stuffing, invented reviews or hidden text.
- It does not sell llms.txt or schema as levers. Both are reported, labelled as optional.
See the measurement protocol on the methodology page, or start with the free AI visibility check.