Adding an llms.txt file does not make a company understandable to an AI assistant.
It can provide a useful map, but it is one signal among many. If the pages it points to disagree about the company name, services, people, or evidence, the map only leads a machine into inconsistent material.
We investigated this while building DandelionTech’s own site as both a company website and a working example of the SEO and GEO service we sell. The central lesson was simple: machine readability begins with factual discipline, not a special file.
Start with one stable identity
A company can accidentally publish several identities at once: a short brand name in navigation, a legal name in the footer, an older description in social profiles, and a different service list in structured data.
Humans can often infer that these refer to the same organization. Machines are less forgiving, especially when the name itself is shared by unrelated businesses or common words.
The site therefore needs a consistent identity layer:
- one preferred brand name;
- the legal name where relevant;
- one canonical domain;
- a crawlable logo;
- the same concise description across important surfaces;
- explicit links between the organization, its founder, services, work, and articles.
Structured data can state those relationships, but the visible page must support them. Markup is not a substitute for content.
Give every important claim a destination
AI assistants are more likely to use facts they can locate, interpret, and reconcile. A homepage claim with no supporting page is weak evidence.
For DandelionTech, that meant creating distinct destinations for:
- each service and its typical engagements;
- verified work and delivery scope;
- the founder’s experience;
- the FLASH Consensus case study;
- the way an engagement operates;
- practitioner articles that explain how decisions are made.
This structure helps people too. A buyer can move from a broad claim to the underlying detail without searching the site manually.
Use structured data to describe relationships
Useful structured data is modest and accurate. An organization can identify its name, legal name, URL, logo, contact route, founder, service area, and areas of expertise. Service pages can identify the service being offered and its provider. Articles can identify their author, publisher, publication date, and canonical page.
The goal is not to add every schema type available. It is to remove ambiguity around the entities already visible on the page.
Unsupported ratings, invented locations, fake frequently asked questions, and fabricated client outcomes create risk rather than authority.
Canonical URLs and internal links still matter
GEO does not replace technical SEO. A machine still needs to find the preferred version of a page and understand how it relates to the rest of the site.
The basic controls remain important:
- descriptive, stable URLs;
- a canonical URL on every indexable page;
- a sitemap containing the public routes;
- a permissive but deliberate robots policy;
- internal links using descriptive language;
- unique titles and descriptions;
- semantic headings and readable body content.
An llms.txt file becomes useful after this foundation exists. It can summarize the organization and point to the authoritative pages, but those pages remain the evidence.
Consistency must extend beyond the website
An assistant rarely relies on one domain alone. It compares what the company says with profiles, repositories, directories, client references, and other public sources.
That makes off-site consistency essential. The preferred company name, domain, founder name, service descriptions, and contact information should match wherever the organization appears. Real client references and attributable profiles are more valuable than repeating the same claim across more pages on the company’s own site.
This is also why ranking for a generic term is different from establishing a brand entity. “Dandelion” can describe a plant, a protocol, or many unrelated companies. The achievable first goal is making “DandelionTech” and “Dandelion Tech Solutions” resolve consistently to one organization, then earning visibility for specific services and problems.
Measure what the machine can actually retrieve
The final check is not whether the source code contains the right tag. It is whether the built page exposes the intended facts without requiring fragile client-side behaviour.
We verify the generated HTML, internal links, sitemap, canonical URLs, structured data, performance, and accessible reading order. We also inspect what search tools currently return for branded queries. Indexing and citations take time, and no implementation can guarantee the first position or an AI mention.
The work is to make the evidence unambiguous, technically available, and worth citing.
Our SEO and GEO service covers this foundation. The engagement guide explains how the work is scoped, reviewed, and handed over.