If your strategy is “let’s optimize for ChatGPT” and that’s it, you’re leaving money on the table.
Every model has its biases: what it prefers, what it avoids, how it summarizes and when it cites. Claude, for example, tends to be more careful with sensitive claims and to favor clarity, context and consistency when the topic involves business decisions.
You don’t need to “hack” the model. You need to give it citable material.
What a good source looks like for Claude
Claude usually responds better when the source:
- defines terms without marketing language
- explains limits (“this applies if…”, “this doesn’t apply if…”)
- shows a process and decision criteria
- stays consistent across pages
When content reads like a brochure, the model has less to extract useful answers from.
Which business queries trigger citations (rather than mentions)
Citations appear more when the user asks for:
- comparisons (“X vs. Y”, “how to choose”)
- methodology (“how to measure”, “what to monitor”)
- risks (“what can go wrong”, “compliance”)
- implementation (“steps”, “requirements”, “checklist”)
If your website only has service pages without criteria, the model may recommend you, but it will struggle to cite you.
Practical implications for GEO in Colombia
1) Write for decisions, not definitions
Defining “GEO” is fine. What puts you on top is answering:
- which signals a manager would look at
- what evidence they would demand
- how they would measure progress
2) Use a format the model can “clip”
Short blocks, small lists, direct FAQs. If your page is a wall of text, the model loses precision.
3) Reduce internal contradictions
It’s common in Colombia: one page says “we do SEO”, another says “we’re a marketing agency”, another says “we’re a consultancy”. That confuses the knowledge graph.
4) Show governance
Especially in sensitive sectors, describe how you make decisions:
- human review
- traceability
- change control
That increases trust in generative answers.
A simple experiment to detect biases
Run 15 queries with commercial intent and 15 with technical intent.
In each answer, measure:
- whether it mentions you
- whether it cites you
- what type of source it cites (blogs, documentation, directories)
- whether the model’s language matches your positioning
When the model describes you incorrectly, the problem almost always lies in your signals. For a structured approach, see how to measure AI visibility.
Frequently asked questions
Does Claude always cite sources?
No. It depends on the experience and context of the product being used. That’s why it’s worth measuring mentions and citations separately.
What type of content increases the likelihood of being cited?
Practical guides, checklists, comparisons with criteria, pages with FAQs and definitions with clear limits.
Does this replace traditional SEO?
No. It feeds on the same things: good architecture, authority, consistency and useful content. Only the format of consumption changes. See GEO vs. SEO in Colombia.
Next step
If you want to audit how LLMs (including Claude) describe and cite you and turn that into an actionable GEO plan, contact us.
