AI Search Optimisation

Not for search visibility today. llms.txt is a proposed format for giving AI tools a summary of a site, not an adopted standard. Google has said it doesn’t use llms.txt in Search, and I haven’t seen evidence that the major AI search crawlers rely on it to choose sources. Adding one does little harm, but it shouldn’t take time away from the pages and facts those systems actually read.

Only by limiting snippets in normal search as well. Google applies the same controls to AI Overviews and AI Mode as to regular results: nosnippet, data-nosnippet, max-snippet and noindex. For most businesses the trade-off isn’t worth it, because those controls shrink or remove ordinary listings too. data-nosnippet is the most targeted option, since it can keep a specific passage out of snippets without hiding the page.

Systems that search the web live, such as ChatGPT search, Perplexity and Google’s AI features, can cite a page soon after it’s crawled and indexed, sometimes within days. What a model knows from training changes far more slowly, only when new versions are released. So page improvements show up first in answers that involve a live search, and later, if at all, in answers given from memory.

It helps understanding, but it isn’t a requirement. Google says no special markup is needed to appear in AI Overviews or AI Mode. Structured data still makes entities and facts explicit, which helps any system reading the page tell which business, product or attorney it describes. It only helps when it matches the visible content, so markup is never a substitute for clear passages on the page.

They can, especially for simple informational questions that the overview answers in full. Searches where people still need to compare, call or book are less likely to end inside the answer. Google says clicks from results pages with AI Overviews tend to be higher quality, which is hard to verify from outside. In your own data, the tell-tale sign is steady impressions and position with a falling click-through rate.

I can improve the odds, not guarantee the outcome. For local recommendations, assistants draw on business listings, reviews and mentions on other sites as well as your own pages. Accurate profiles, a steady flow of genuine reviews, clear service and location pages and coverage in local media all help. The question panel then shows whether you appear for those questions, and which competitors do instead.

It can, if it’s genuine. AI answers often cite forum threads, Reddit in particular, because they contain first-hand experience. Staff who answer questions in relevant communities, openly saying who they work for, can earn mentions that assistants pick up. Fake accounts, planted recommendations and undisclosed promotion break community rules, get removed and can damage trust in the brand far more than they help.

Judge it by influence as well as volume. AI referrals are still a small share of traffic for most sites, but people asking assistants detailed questions are often further into a decision, and many see a brand named in an answer before they ever search for it. Growth in branded searches and enquiries that mention an AI tool belong in the picture alongside referral visits.

That’s a business decision rather than an SEO one. Publishers with original research or paid content often block training to protect it. Businesses whose goal is to be known and recommended usually gain little from blocking, because a model that has never seen their pages knows less about them. I set out the trade-offs for your situation, and you decide which policy to run.