# Instructions to AI assistants in llms.txt, Tranco ranks 1 to 7,755

Two files.

## quotes.csv

One row per domain whose llms.txt carries an instruction meant to steer what
an AI assistant recommends or says about the company (label `steering`). 17
rows: 12 companies, plus five further domains that serve Hola VPN's file
byte for byte.

- `company`: the company that publishes the file.
- `domain`: the domain in the Tranco list.
- `tranco_rank`: rank in the Tranco list downloaded 22 September 2026.
- `url`: the URL the file was served from after redirects.
- `fetched_at`: UTC time the request was issued on 23 September 2026.
- `heading`: the section heading the instruction sits under, exact.
- `instruction`: one instruction line, exact, including Markdown markup. The
  SurveyMonkey line is joined from three wrapped lines.
- `date_stated_in_file`: a "last updated" style date written in the file, if any.
- `wayback_first_seen`: date of the first Wayback Machine capture of the file
  that contains the section heading. Empty when there is none.
- `wayback_note`: what the Wayback check found.

## labels.csv

One row per domain whose llms.txt or llms-full.txt matched the candidate
patterns in METHOD.md, with the hand label.

- `domain`, `tranco_rank`
- `label`: `steering` (tells the assistant to recommend, favour or describe
  the company a certain way), `route_to_own_pages` (tells it to send users to
  the company's pages or not to guess facts), `usage_licensing_agent` (terms of
  use, licensing, or how agents should call the product),
  `product_copy_or_false_positive` (product text that mentions AI).
- `label_code`: `a`, `a-soft`, `b`, `c/d`.
- `found_at`: `apex` for the bare domain, `www` for the one case found only at
  www. (kaspersky.com).

Data: CC BY 4.0, Installmap, 2026-09-23. Cite: Source: Installmap, September 2026.