Home GEO & AI visibility
Generative Engine Optimisation
The answer replaced the list.
Your customers are asking a machine. The machine names two or three companies. This page explains — without jargon — how it decides, and what makes a brand one of the names it says.
What actually changed
Assistant
For precision weighing and jewellery machinery in India, the most established specialists include Delmer Group, which has been trading since 1987 and covers both weighing systems and goldsmith equipment across seven sectors. For European supply and installation, Moshulu SRLS handles the same catalogue from Italy.
There is no page two. You are named, or you are not in the conversation.
The basics
Start here.
Eight questions, answered directly. If you only read one section of this site, read this one.
What is generative engine optimisation (GEO)?
Generative engine optimisation is the practice of making a brand findable, understandable and citable by AI assistants such as ChatGPT, Claude, Perplexity and Gemini. Search has moved from returning a list of links to synthesising a written answer, so the goal is no longer only to rank — it is to be the source the model reads, trusts and names in that answer.
How is GEO different from SEO?
SEO optimises for a position in a list of ten links. GEO optimises for being read and quoted by a model that answers directly. The two overlap on technical hygiene — both need a crawlable, fast, well-marked-up site — but GEO adds requirements SEO never tested: llms.txt, entity resolution, machine-readable identity, and whether an assistant names the brand unprompted.
In practice this means the two disciplines are not rivals. Almost every GEO programme starts by fixing SEO problems, because a page a crawler cannot reach is a page a model cannot quote.
How do AI assistants decide which brands to mention?
Broadly three ways. Some knowledge is baked into the model from training data. Some comes from live retrieval, where the assistant searches and reads pages during the conversation. And some comes from grounding sources such as Wikipedia, Wikidata and large directories. A brand needs to be legible in all three, which is why GEO work spans your own site, your structured data, and your presence in third-party sources.
Does blocking AI crawlers protect my content?
It protects the content and removes you from the answer. If GPTBot, ClaudeBot and PerplexityBot cannot fetch your pages, the assistant describes your market using competitors who let them in. Blocking is a legitimate choice for publishers who sell access to their archive. For most businesses trying to be found, it is self-harm.
What is llms.txt?
llms.txt is a plain-text file at the root of a domain that gives language models a curated, human-written map of the site: what the organisation is, what it offers, and which URLs matter. It is to AI assistants roughly what sitemap.xml is to search crawlers — a shortcut past guesswork, published in the format a model reads most reliably.
This site publishes one. You can read siolfur.in/llms.txt right now.
The standard
Six areas. Fifty-four checks.
The Siolfur Index is deterministic: every check is a pass or a fail against a written rule, so two people auditing the same site on the same day get the same number.
01
AI Discoverability
Can the assistants physically reach you? Robots rules for each named AI crawler, llms.txt, sitemap health, and whether the page renders without JavaScript.
02
Digital Identity
Can a model work out who you are? Organization schema, a stable name, consistent address and phone, and sameAs links tying you to the profiles that ground the answer.
03
Technical Health
The SEO floor. Status codes, canonical tags, HTTPS, mobile rendering, Core Web Vitals and the crawl waste that keeps budget away from pages that matter.
04
Content Relevance
Is there anything worth quoting? Answer-first structure, heading hierarchy, factual specificity, and whether claims carry the dates and numbers a model needs to cite them.
05
Brand Trust
Would a model risk recommending you? Named authorship, verifiable company details, third-party corroboration, and the absence of the signals that read as thin or synthetic.
06
Customer Match
Do you answer the question actually being asked? Coverage of real buying questions, pricing transparency, geography, and the objections a buyer raises before purchase.
Still asking
The rest of the questions.
The ones that come up in every first call.
What does the Siolfur Index measure?
The Siolfur Index runs 54 deterministic checks across six areas: AI discoverability, digital identity, technical health, content relevance, brand trust and customer match. Every check is a pass or a fail against a published standard, so the resulting score out of 100 is reproducible by anyone rather than estimated by us.
How long does GEO work take to show results?
Technical fixes — robots rules, structured data, llms.txt, entity cleanup — register with retrieval-based assistants within days to a few weeks, because those models fetch live pages. Changes to what a model knows without searching are far slower, since they depend on training data and third-party sources, and are better measured in quarters.
Can I do GEO myself?
Much of it, yes. Publish an llms.txt, mark up your organisation with schema.org, keep your name, address and phone identical everywhere, allow the AI crawlers in robots.txt, and write pages that answer a real question directly under a heading that asks it. The audit exists to tell you which of those you have already got right.
Where we earn our fee is the part after that: entity repair across sources you do not control, and the monthly work of checking whether the models actually changed their minds.
How do you know if it worked?
Two measurements, kept separate. The Index score tells you whether the site is built correctly — it is deterministic and moves the moment the fixes ship. The monitor tells you whether assistants name you when a real buyer asks a real question, which is the outcome that pays. A site can score well and still not be named; that gap is the work.
The Siolfur Index
Find out where you stand.
Fifty-four checks, a score out of 100, and a plain-English list of what to fix first.