No, a small business website does not need an llms.txt file to appear in Google’s AI answers. Google’s own guide to optimising for generative AI features, last updated in July 2026, says Search does not use llms.txt or other AI-specific files, and that publishing one will neither help nor hurt your visibility there. The file is a plain Markdown summary of a site, placed at its root, proposed as a way for language models and agents to find the important pages quickly. It has been sold to business owners as a shortcut into ChatGPT and AI Overviews. We could not find evidence for that, and the one large dataset on whether anything reads these files points the same way Google does.
That dataset comes from Ahrefs, which looked at 137,210 domains that received traffic in May 2026. About 28 percent of them published an llms.txt file. Ahrefs notes its customers skew technical and SEO-aware, so it treats that figure as an upper bound for the wider web. Of the sites that did publish one, roughly 97 percent received no requests for it at all during the month. The remaining three percent, about 1,100 domains, accounted for around 22,000 requests, and 96 percent of those came from bots rather than people. The biggest named groups were SEO audit tools and general web crawlers. AI assistants made up 2.5 percent of requests and AI retrieval bots 1.1 percent.
Ahrefs is careful about what that shows. A request is not proof that a bot read or used the file, so the AI figures are ceilings, and the study did not check whether the files followed the specification. We would add a limit of our own, which is that server logs show whether a file was fetched and say nothing about whether a model absorbed the same information from the pages themselves. Even so, “almost nobody asks for it” is a hard place to start if you are trying to argue that a file is worth building.
The part we find more interesting is that Google’s two teams do not say quite the same thing. The Lighthouse documentation from Chrome, updated in May 2026, calls llms.txt an emerging convention, a machine-readable summary meant for LLMs and agents, and its audit flags a page if fetching the file returns a server error. A missing file, a plain 404, is marked not applicable, and the page says providing one is optional at the moment. Lighthouse is asking whether an agent can find its way around a site, which is a different question from whether Search will rank or cite you, and a site can be fine on both without resolving the difference.
Our read is that the file is harmless and mostly pointless for a local tradesperson or a ten-product Shopify store. If a client wants an llms.txt file, it takes about an hour, and we would rather spend that hour on what Google’s guide does say matters: pages that are crawlable and indexed, and content with something original in it. The same guide says you do not need special schema.org markup for generative AI features, though structured data still earns its place for rich results, and that cutting your content into small chunks for models is unnecessary. A fair number of “AI search” services are built on exactly those tactics, so reading the guide first is cheap insurance before paying for one.
Where the file might eventually matter is the agent end of things, software that browses and acts for a customer, which is the audience the Lighthouse page describes. We would treat that as a build question, closer to the work on our AI automation page than to SEO, and for now the case for it rests on a documentation page and not a measured result.
Google’s guide does point site owners to one concrete thing, which is to verify the site in Search Console and look at the Generative AI performance report there. We would read that before building anything for a model.
Of the sites that did publish one, roughly 97 percent received no requests for it at all during the month.
