What we don’t sell you
Three items are routinely billed as growth levers in the GEO market although the published evidence points the other way: an llms.txt file, FAQPage markup, and unblocking the training crawlers. We will not put them on your invoice. Below is what the market sells, what the numbers say, and what we do instead.
An llms.txt as a deliverable
Providers list llms.txt as a named line item of a monthly retainer — a file that supposedly explains your website to the AI systems.
In a 90-day experiment by Otterly, exactly 84 of more than 62,100 AI-crawler visits went to /llms.txt — around 0.1 percent. No provider has confirmed support for the format.
We spend the same hours on what the crawlers do fetch: robots.txt access for the bots that answers are built from, and pages whose answer sits at the top. If you already have an llms.txt we do not delete it — we only make sure it does not contradict the site.
FAQPage markup as a growth lever
Structured data gets sold as a line that lifts your visibility in AI answers, and FAQPage markup most eagerly of all, because it is quick to add and easy to show.
The FAQ rich result stopped appearing in Google Search on 7 May 2026, and its documentation was removed on 15 June 2026. The one controlled measurement we know of — 1,885 pages that added JSON-LD against roughly 4,000 matched controls — found no uplift in citations: ChatGPT +2.2 percent, Google AI Mode +2.4 percent, AI Overviews minus 4.6 percent.
We write questions and answers as visible prose: a real question as the heading, the answer directly underneath. A human reads that, and an engine quotes from it.
For completeness: our own services page still emits FAQPage markup. It does no harm, and stripping existing markup would be a step backwards in ordinary search results. We simply never bill it to anyone as a growth lever.
Sources:Google Search Central — changelogAhrefs — schema and AI citations
Unblocking GPTBot, Google-Extended and ClaudeBot
“We’ll open your site to the AI crawlers” sounds like a piece of work and gets invoiced as one — usually without distinguishing which bot governs what.
The distinction is in the providers’ own documentation. What governs answers and citations today: OAI-SearchBot (“Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers”), Claude-SearchBot and Claude-User (“may reduce your site’s visibility for user-directed web search”), Googlebot and PerplexityBot. What does not: GPTBot (model training), Google-Extended (“does not impact a site’s inclusion in Google Search nor is it used as a ranking signal”) and ClaudeBot (training data).
We check the first group and open what is shut — that is the door into the answers. The second group we report as information: whether your content feeds model training is a matter of your own stance, not a visibility task. We do not invoice it.
Sources:OpenAI — crawlersAnthropic — crawler supportGoogle — common crawlers
How we back this up
Every line above has a source you can open yourself. That is exactly how we work on your case too: look first, claim second.
Want to see where you stand in AI answers today? free AI-visibility check.
Already paying someone for GEO and wondering which of it holds? independent GEO second opinion.