Probably not by default. A local business should add llms.txt only when it can name a documented consumer, keep the file accurate, and explain what useful job the file performs beyond the website’s normal pages.

llms.txt is a voluntary, evolving proposal for a Markdown file at /llms.txt that gives a concise introduction and links to useful resources. It is not a crawler-control, indexing, ranking, training, or citation command. The similar name to robots.txt causes much of the confusion.

Treat the two files as different controls

robots.txt is a crawler-access mechanism. Google and Bing both document it as a way to tell crawlers which URLs they may request; individual providers document their own user agents and access rules. OpenAI, Perplexity, and Anthropic likewise describe separate crawler roles and robots controls. Those are policy decisions about a named crawler.

llms.txt is a proposal for an optional content guide. The proposal does not create a directive language that tells a crawler what it may access, require a search engine to index a page, or require an assistant to cite it. Publishing one is not evidence that a particular system reads it.

That distinction matters because a business might reasonably choose a crawler policy while declining llms.txt, or publish a maintained guide while separately setting crawler rules. One choice does not settle the other.

Use the implement-or-decline decision tree

Start with the questions below in order. A “no” at any of the first three leaves is a sensible reason to decline for now.

  1. Can you identify a documented consumer? Do not rely on a vendor’s vague claim that “AI needs it.” Identify the specific product, documentation, or user workflow that says it can use the file.
  2. Is there a real reader job? A compact guide may be useful for a documentation-heavy site with stable reference material. A five-page local-service site with a clear homepage, service pages, contact path, and sitemap may have no separate job for it.
  3. Is there a maintenance owner? Someone must update the file when services, pricing, policies, locations, or key URLs change. A stale summary is worse than no extra summary.
  4. Can the file link to existing, maintained public pages? Do not make a private catalog, a keyword list, or a competing set of claims. The source pages remain the accountable pages.
  5. Does the opportunity cost make sense? If the website’s visible facts, internal links, crawlability, or service explanations are unfinished, fix those first.

If the answer reaches the fifth question with evidence and an owner, a small maintained llms.txt file may be reasonable. If it does not, record a decline decision and revisit only when a documented use case appears. Declining a speculative file is not a sign that the business has opted out of AI search.

What a careful implementation would look like

For a business that passes the decision tree, keep the file modest:

  • identify the business in plain language using already approved public facts;
  • link to the canonical homepage, service pages, contact path, and maintained articles that actually explain the business;
  • avoid claims that the linked pages cannot support;
  • name the person responsible for review; and
  • set a trigger such as a service, location, pricing, or URL change.

Do not use llms.txt to smuggle in hidden sales copy, duplicate pages, private operating notes, or crawler instructions. It should not become an unreviewed parallel version of the website.

Keep the boundary clear

llms.txt and robots.txt can both be plain-text files at a site root, but they do different jobs. llms.txt is not a crawler-access directive or a substitute for a crawler-specific policy. Making it publicly reachable does not command indexing, ranking, inclusion, training choices, or citations.

Providers publish their own crawler controls and roles; the proposal does not replace or combine those policies. Some sites may find a maintained guide useful for a documented consumer, but the proposal itself does not establish universal adoption or a universal local-business need.

Start with the pages that already carry the promise

The dependable work is still ordinary website maintenance: visible service facts, clear headings, useful pages, working links, crawl access where the business has chosen it, and current contact information. What AI Search and AI Assistants Need From a Local Service Website covers those broader retrieval foundations.

If a business cannot name the consumer and owner, write “declined: no documented use case or maintenance owner” in its site checklist. That is more honest than publishing a file simply because it sounds current.

Sources