The first version of this site's llms.txt got the idea wrong by copying a more familiar, adjacent format — robots.txt syntax, in a file whose whole point was supposed to be prose. That mistake lived for ten minutes before a second commit fixed it. A week later, a Google Alert turned up the same mistake again, at industry scale, with a name attached to it: AEO, and its sibling GEO.
What showed up in one week of alerts
This site runs a Google Alert on the keyword "seo." In the week of September 5–12, 2026, it delivered 112 emails and 620 distinct results after deduplication. Seventy-seven of those 620 mention "AEO" (Answer Engine Optimization) by name. Ninety-three mention "GEO" (Generative Engine Optimization). Thirty-six mention both in the same result.
These aren't all agency blog spam dressing up a landing page. A sample of the exact job titles posted that same week, verbatim:
Senior Strategist, SEO and GEO— Verizon CareersTeam Lead, SEO and AEO— Atlassian, via Built InGlobal Digital Marketing Manager - SEO/AEO/GEO— BCG CareersLead Analyst, Technical Search (SEO/AEO/GEO)— CignaGlobal SEO/AEO Manager— Iron Mountain, via Built In
Five different enterprise employers, five job requisitions, all in the same seven-day window, all appending the same two new letters to a role that didn't need them a year earlier. Whatever AEO and GEO actually are, they'd crossed from agency marketing copy into corporate job titles by last week.
What the letters are supposed to mean, and what they don't change
Read at face value, the distinction is real: AEO is optimizing for getting cited inside an AI-generated answer rather than ranking a blue link; GEO is optimizing for being retrieved and quoted by a generative model at all. Both are genuinely different targets than classic keyword ranking. That's not nothing — it's the same kind of genuinely-new difference that llms.txt has from robots.txt: a new thing, responding to a new kind of consumer of the content, that happens to rhyme with an old and familiar one.
The same week's alerts also surfaced Search Engine Land's survey of 131 SEO professionals on what actually moves rankings. The top-rated factors in that survey: search intent match, backlinks, and original research outperforming scaled AI content. None of those are new. None of them are AEO- or GEO-specific. They're the same substance a search-ranking piece would have named five years ago, asked of the same working practitioners who are, this week, also relabeling their job titles and their agency service pages with the new initials.
The same distinguishing feature, missing in the same place
The llms.txt mistake wasn't writing a bad file — it was writing a file that satisfied the filename without engaging with what made the filename's format different from the one it was copying. User-agent, Allow, and Crawl-delay are real robots.txt syntax, correctly written, inside a file whose entire reason for existing was to hold something robots.txt can't: readable prose context for an agent, not machine directives for a crawler. It looked complete because the syntax was genuinely correct — just correct for the wrong thing.
A job posting titled "SEO/AEO/GEO Manager" has the same shape. It's not a lie — AEO and GEO are real, distinct optimization targets, the same way llms.txt is a real, distinct file format. But stapling the new initials onto the old title, without anything in the actual job description changing to reflect a genuinely different skill — citation tracking in AI answers, structured data an LLM can actually parse, prompt-level visibility testing instead of SERP-position tracking — is the vocabulary equivalent of writing real robots.txt directives into a file named llms.txt and calling it done. The new label is doing the work of signaling change, while the actual content underneath it, per 131 people who do this for a living, hasn't moved.
Why this is easy to miss at either scale
One file, wrong for ten minutes, was a low-stakes mistake anyone reading the raw text could catch immediately. An entire market relabeling itself over a single week is a slower, quieter version of the identical error, and harder to catch precisely because no one result looks wrong in isolation — a job posting, a LinkedIn hot-take, an agency service page each read as a single, reasonable-looking data point. It's only visible as a pattern once you count: 620 results, 93 of them reaching for a label that describes a real and useful distinction, most of them not doing anything differently underneath it. Checking whether the new name tracks a new substance, instead of assuming a name change implies one, is the same check either way — it's just much easier to run against one eight-line file than against ninety-three job postings and blog headlines scattered across a week's worth of search results.