The questions, including the awkward ones

No, but it does not replace it either. SEO gets you into a results page; GEO works on whether an answer engine can extract, attribute and quote you once it is already reading. They share the foundation — a page nobody can crawl or render is invisible to both — and diverge above it.

Anyone selling you GEO while telling you technical SEO no longer matters is selling you a roof without walls.

Probably not for the reason you have been told. Ahrefs looked at 137,210 domains with traffic in May 2026: 28% published an llms.txt, and 97% of those files received zero requests that month. Their own line is that Slackbot fetched llms.txt more often than PerplexityBot did.

This site publishes one anyway, and says why: it costs nothing to generate from data that already exists, and a file that is wrong is worse than no file. What it will not do is sell you llms.txt as a visibility lever. That figure is Ahrefs', not mine.

No. Google has said machine-readable files like llms.txt are not needed to appear in generative AI search. What governs your appearance in AI Overviews is the robots.txt directives for Googlebot, plus the usual preview controls.

If you want to be quoted, yes: a crawler that cannot fetch your page has nothing of yours to quote. There is no third option where you are cited without being read.

The real decision is a different one, and it is yours: training and answering are not the same thing. Some tokens govern whether your content trains a model, others whether it can be retrieved to answer a question. Blocking the first does not stop the second, and blocking the second is what makes you invisible.

It is not a crawler. Google calls it a standalone product token you put in robots.txt to control whether your content trains future Gemini models and is used for grounding.

And in Google's own words, it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search". So blocking it is a decision about AI training, not a decision about your search visibility. People confuse these two constantly.

Because these systems are not deterministic. Ask an answer engine the same thing twice and the set of sources it cites changes — in published work that repeats identical queries, the overlap between two runs within a single day sits around a Jaccard index of 0.32 to 0.43. Roughly six of every ten sources change with nothing altered in between.

That is not a detail, it is the whole reason measurement here has to be repeated. The dataset behind that figure is public.

Two things get measured, and they are not the same thing, so they are not mixed.

The mechanical properties of your pages — structured data, entities that resolve, crawler access, content visible without JavaScript, claims tied to a source — are deterministic. A script measures them, you run the same script and get the same number, byte for byte. Those scripts are public.

Appearing in an answer is not deterministic, so it is measured by repetition: a fixed set of your customers' real queries, re-run and logged. A single run is an anecdote. A percentage with no run count and no spread is not a measurement.

No, and nobody honest can. The decision to cite belongs to the engine, changes between two identical runs, and is not exposed to anyone outside it.

What is deliverable is the engineering and the evidence: making your content extractable, attributable and quotable, and measuring what can be measured honestly. If someone guarantees you a position in an AI answer, ask them for the run count and the confidence interval.

First, write it down: the exact prompt, the engine, the date, and a screenshot. Without that you cannot tell a one-off from a pattern, and almost every complaint about this turns out to be a single run nobody repeated.

Then look at where it could have come from. A model does not invent a company's details out of nothing: usually there is an old page, an outdated profile, a directory entry or a competitor's comparison sitting somewhere. What you can act on is that source and your own pages — publishing the correct version in a form a machine can extract, with the fact stated plainly and tied to something verifiable.

What nobody can do is edit the model's answer. Anyone offering you that is selling something they do not control.

Usually not. Most of what matters here is structure and markup on the pages you already have: entities declared properly, content that survives without JavaScript, crawler access, claims that carry their source.

A rebuild only comes up when the site cannot render its content server-side at all. That is a real problem and it is visible in a minute — you do not need to take anyone's word for it.

The mechanical part is immediate: the moment the changes ship, the measurements of your pages change, and you can re-run the scripts yourself the same day.

Appearing in answers is another matter and does not follow a calendar: it depends on each engine re-crawling and on its own retrieval, which nobody outside it controls. That is exactly why the honest thing is a monthly panel instead of a promised date.

It depends on the size and the state of the site, and I would rather tell you a real number than a range that fits nobody. That is what the free mini-audit is for: with your sector and your URL I can see what is actually there, and then I quote.

What I will not do is quote before looking. A price given without seeing the site is either padded to cover the unknown or wrong.

That is the point. Every mechanical figure comes from a script in a public repository: clone it, run it against your own pages, and you get the same number. No API key, no model in the loop.

And it applies to this site too. What is claimed here is measured here first — including the parts that do not come out well.

Because what you get here is the person who does the work, and the work is published. The method, the experiments and the measurement scripts are all open: you can read what I would do to your site before hiring me, and you can check whether I did it afterwards.

The honest flip side: I am one person. If what you need is twenty people starting on Monday, this is not the right fit and I will tell you so.