Lesson 3 of 5 · 3 clips · 14 min
Content that gets pulled into answers
What does a page need before an AI answer will use it?
Write the facts a machine can use, not the pitch a person skims. Two guests make that case from different directions, and the third clip takes on llms.txt, the file every team is being asked about.
Clip 1: Say who you serve, and who you do not
Who is speaking: Solomon Thimothy until about 1:54, then Matt Bertram.
Solomon Thimothy's case: people type messy prompts and the engine still works out what they meant, so the website has to do more of the explaining. Answer real customer questions plainly, and be explicit about which industries you serve and which you do not, even as a simple table, because that is the level of detail people now ask about. Matt ties it to Google's mission of organizing the world's information and adds one instruction: pick a focus, because a brand cannot be the expert on everything.
Where our evidence differs: Schema markup comes up early in this clip as part of what lets ChatGPT read a site. The best controlled test we know of found that adding structured data did not raise AI citations. That is narrower than saying schema is useless; the lesson linked here has the study and its limits. SEO 101, lesson 4: what actually moves the needle.
Clip 2: Machines want the facts, not the hook
Who is speaking: Duane Forrester, then Matt Bertram from about 3:51.
Duane Forrester's coffee maker story is the most concrete example in the course. He needed to know whether a machine would fit under his cabinets, and the only dimensions published were the shipping box's, because that is what the factory supplied and what flowed into every product feed. The emotional hooks we write for people are ignored by these systems; they want plain facts. Each platform also weights things differently. Matt adds personalization: an assistant that remembers a user can give two people different answers to the same question.
Clip 3: Does llms.txt matter?
Who is speaking: Matt Bertram asks; Duane Forrester answers from about 1:07.
Matt asks the question every team is fielding. Duane Forrester's answer: the trouble with llms.txt is not the idea but adoption and trust. The robots.txt file, XML sitemaps and schema.org all launched with the search engines behind them, and not a single AI platform has committed to llms.txt. He sees little value in it today. If he worked at an AI company, he says, he would collect every llms.txt file published and keep the signals that help. His closing point corrects an old myth: robots.txt is strictly a blocking mechanism. It never told a crawler what to crawl.
Try this before the next lesson: Open your most important service or product page and list every fact a buyer needs before deciding: who it is for, who it is not for, sizes, pricing basis, locations, timelines. Mark each one as on this page, on another page, or nowhere. Everything marked nowhere is a gap an answer has to fill from somewhere else.
All 5 lessons · SEO 101, the companion course for business owners