Blog

Notes on AI search visibility: how assistants pick the brands they recommend, and what moves the answer.

  • What changes when your customers stop searching

    Search rewarded the page that ranked. AI assistants reward the brand the model can describe: a different mechanism, and a different thing to measure.

  • Your brand is described by pages you do not own

    Assistants lean on retailers, reviews, forums and video for the sources behind a recommendation. Your own site is a minority of the evidence they use.

  • Should you let AI crawlers in?

    Training bots and search bots are different tokens with different consequences, and most robots.txt files block the wrong one. What each control actually does.

  • Structured data, llms.txt and the rest of the plumbing

    Which technical measures platforms actually say they use, which they say they ignore, and the one new control that genuinely changes whether you appear in AI answers.

  • Writing for retrieval, not for ranking

    What the research on position bias and retrieval actually supports about structure, and where the evidence runs out and vendor advice takes over.

  • What actually moves an AI answer

    The best evidence we have on changing how a generative engine describes you comes from one 2024 paper. Its findings are useful, specific, and smaller than the headlines.

  • Three numbers worth tracking before any others

    Presence, perception and displacement. What each one answers, how to compute it defensibly, and why a single AI visibility score usually hides the useful part.

  • Why testing ten prompts tells you nothing

    Same question, same settings, different answer. The research on variance says diversify phrasings, languages and engines rather than repeating one prompt.

  • Where AI answers get their sources

    Assistants cite about four sources where a search page lists ten, and roughly a third of the domains they cite never appear in the matching organic results.

  • How an assistant decides which brand to name

    Two different mechanisms decide whether your brand appears in an AI answer: what the model retrieved, and what it already believed. They need different work.