AnswerShare founder Robert Maynard Jr published a guide on Search Engine Journal to changing how AI answers describe a brand, using a Cloudflare-style worker to serve a llms-full.txt file only to AI crawlers

A brand's reputation on ChatGPT, Perplexity or Google's AI mode is decided by a small set of pages an AI crawler read months ago. On Monday, Search Engine Journal ran a piece by Robert Maynard Jr, the founder of an AI-visibility company called AnswerShare, describing how his firm gets AI answers to change their tone within days: publish a file at /llms-full.txt with the full brand story, then serve it to AI crawlers through an edge worker at the CDN, byte-for-byte identical to what a human would read.

The article was published on 2026-09-29 at 20:40 UTC and carries an "SPA" tag, marking it as sponsored partner content. Maynard is writing about the technique his own company sells, so every measurement in the piece is AnswerShare's own account of its own client.

What the machine layer is

Two facts sit behind the design. AI crawlers read a page differently from a person: they lift a passage, drop the surrounding page, and rank it against thousands of other passages on the same claim. And the AI systems overweight the passages that carry a number or a complaint, without the denominator that would put the complaint in context. So a business with 35,000 customers over 13 years and two Better Business Bureau complaints, one of Maynard's client examples, reads to an AI model like a business defined by two complaints.

The proposed fix is a plain-text file at the root of the site, llms-full.txt, that carries the full brand context: what the business does, credentials, reviews and their denominators, the response to concerns, and the sourced record. The file is what the AI crawler is meant to reach. An edge worker at the CDN routes requests: AI crawlers to the machine layer, users and Googlebot to the human site. The two layers are byte-identical, Maynard writes, so nothing about the routing hides anything from a search engine that also indexes the page.

llms-full.txt is not an internet standard. There is no RFC behind it. Maynard's own site publishes one at answershare.com/llms-full.txt, and his article says his client's file was crawled 154 times in two months. Whether AI providers keep reading a file at that path, without a formal convention, is what a team trying this would be betting on.

The numbers, as AnswerShare reports them

The main case study in the article is one unnamed AnswerShare client. On 42 prompts a prospective customer would type, AI answers about the client initially warned readers off in 27 of them, and only 10 of 42 answers recommended the client. The concerns cited by the AI models appeared with proper context in only 1 of 42 answers.

After the client published llms-full.txt and routed AI crawlers to it, Maynard reports that the first shift showed up in three days, and stabilised in about two weeks. Warnings fell to zero of 40 answers, recommendations rose to 40 of 40, and every concern cited by an AI model appeared with the client's own denominator or response next to it.

Those figures were measured by AnswerShare on one client, on a set of prompts AnswerShare chose. There is no independent replication, no comparison to a control, and the client industry is not named. They are the company's report on its own service.

Source

• Primary: Yes, You Can Change AI's Opinion. Here's How., Search Engine Journal, 2026-09-29, by Robert Maynard Jr (AnswerShare). Marked as sponsored partner content on the page.