What We Publish So Answer Engines Can Quote Our Sites

What we publish for answer engines

An answer engine can only quote a page it is allowed to fetch and able to read without running our JavaScript. We work on both conditions on three sites: bbmmtech.com, flowoapp.com and textpack.app.

The set is small and each item is a plain file or a block of markup.

• robots.txt with a named group for each AI crawler we want to admit.
• /llms.txt, a short Markdown record of who we are and where the pages live.
• /llms-full.txt, the longer text in one request.
• FAQ answers written so that each one stands alone.
• FAQPage structured data generated from the same answers the page shows.

bbmmtech.com is a Next.js static export and flowoapp.com is prerendered at build time, so every page arrives as finished HTML.

How robots.txt names AI crawlers

The robots.txt on bbmmtech.com is generated by a route file that holds an explicit list of 17 user agents. A wildcard rule already allows every crawler. The list adds one group per agent, each with Allow set to the site root. It includes GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, Applebot-Extended, Amazonbot and CCBot.

The flowoapp.com file has the wildcard plus 19 named groups, sorted under three comments: search, AI retrieval and citation, AI training and grounding. Every group disallows /api/, because that path is the POST endpoint behind the Ask Ondi chat and holds nothing a crawler can index.

We name agents separately for two reasons that are written in the file's own comments. Google-Extended and Applebot-Extended are opt-out tokens, so an Allow line under their names is an explicit opt-in to grounding. OAI-SearchBot and ChatGPT-User are the fetchers that retrieve and cite pages, while GPTBot collects training data. A file that lists GPTBot alone says nothing to the fetchers that produce citations.

What llms.txt contains

The llms.txt on bbmmtech.com follows the llmstxt.org layout: a title, a quoted summary, then sections. The summary is two sentences that give the legal name, the city, the incorporation date and the two products with their platforms.

The sections that follow are lists of facts.

• Company facts: legal name, incorporation date and jurisdiction, registry name and number, registered office, officers, contact addresses.
• Services: the four kinds of client work we take on.
• Own products: name, platform, URL and a two-sentence summary for each.
• Pages: the home page, FAQ, blog, feed, legal documents, sitemap and llms-full.txt.
• Naming: the preferred name, the legal name, and a line saying BBMM has no public expansion.
• What this company does not claim.

The file also states its purpose and its licence to quote in two sentences: it exists so assistants can describe and cite the site accurately, and content may be quoted with attribution to BBMM Technologies.

The flowoapp.com file is a product record. It carries store identifiers, a description of each feature, a privacy section, a list of scheduler limits and eight questions in Q and A form. The textpack.app file lists the six source checks TextPack runs before export and the formats it writes.

What llms-full.txt adds

llms-full.txt lets an assistant read the substance of a site in one request. The two sites build it differently because their content lives in different forms.

On bbmmtech.com the file repeats the llms.txt header and then prints every FAQ answer under its category heading. After that it lists each published article with its URL, publication date, update date, author, summary and key points. The full article text stays at the URL. The current build of the file is about 106 KB.

On flowoapp.com the prerender script starts from llms.txt and appends the text of every page in the sitemap except the five language home pages, which llms.txt already describes. Each page starts with its title and a line that begins with URL: so a quoted passage can be traced to its address.

A small converter turns the prerendered HTML into text. It deletes script, style, nav, svg, noscript and template elements, because they repeat on every page. Headings become Markdown headings two levels deeper than in the page, and list items become bullets.

How FAQ answers are written and marked up

The FAQ on bbmmtech.com holds 18 questions in four categories, and the rule at the top of the file says each answer must be direct and self-contained. In practice this means the answer repeats its subject. The answer to the question about location begins with the full company name and ends with the street address, so it still reads correctly when a model lifts it without the question.

The page renders the questions as a definition list, with the question in a dt element and the answer in a dd element. The same array is passed to a builder that emits FAQPage structured data, with one Question and one acceptedAnswer per item. The text in the markup and the text on the page cannot differ, because both come from one array.

Blog articles follow the same pattern. Each article has a summary, a list of key points and its own questions. When an article has questions, the page emits a FAQPage block next to its BlogPosting block.

Freshness and discovery signals

We give crawlers dates that change only when content changes. One constant in the SEO module holds the date the site copy was last reviewed. It is printed in llms.txt as the last reviewed date, in a last-modified meta tag and in the home page's dateModified.

Each article in the sitemap carries its own lastModified taken from its update date. A single shared date on every URL tells a crawler nothing about what changed.

The robots meta tag sets max-snippet to -1 for Googlebot, which removes the length limit on quoted text. After each deploy, a job reads the live sitemap and submits its URLs to IndexNow.

Limits of this approach

llms.txt is a community proposal, and no answer engine has committed to reading it. We treat the two files as a low-cost addition. The HTML pages remain complete without them, and everything in the files is also stated on a page.

We publish no figures on how often our pages are cited, and this article makes no claim about ranking. It describes what the files contain and how they are built.