Nine signals that decide
which sources AI models cite
“How do I get into ChatGPT?” is the question I hear most. There is no submission form and no switch. What there is, is a set of signals models use to select and trust sources. Here are nine, ordered by how much they weigh in practice.
By Manuel Cilurzo · AI consultant, Belgium · updated August 2026
Short version: models do not pick the prettiest source or the highest-ranked one. They pick the source they can understand, verify and retell without risk. Everything below is a variation on those three words.
1. Unambiguity
A model forced to guess what you mean will not name you. “We deliver end-to-end solutions for the construction sector” is unusable. “We install roof insulation and solar panels on private homes in the province of Antwerp, from €4,500” is quotable. The difference is not style, it is information density.
2. Facts existing as facts
Prices, opening hours, capacities and terms must exist as discrete data, not buried in prose or locked in an image. Structured data via schema.org is the strongest instrument here: it converts doubt into certainty. A model pulling a price from a JSON-LD block has nothing left to interpret.
3. Consistency across sources
Your hours on your site, your Google profile, your social pages and three directories must all say the same thing. Contradictions are the fastest way to make a model uncertain, and an uncertain model picks the source with no contradiction. This is dull work and probably the highest return per hour in this entire field.
4. Answering the question literally
People no longer type keywords into an assistant, they ask full sentences: “what does it cost to cover a terrace in Belgium and do I need a permit for it?” The page that literally answers that question — permit part included — gets cited. The page headed “Terrace canopies” with three beautiful photos does not.
5. A recognisable entity behind it
Models weigh who is making the claim. A name, a face, a LinkedIn profile, a consistent story about expertise, and a link between person and organisation — that turns an anonymous website into a source with an author. It is also the part your competitor cannot copy in an afternoon.
6. External corroboration
What others write about you weighs more than what you write about yourself. Industry platforms, press, reviews, mentions in articles, trade associations. Not because links count the way they do in classic SEO, but because repetition across independent sources gives a model confidence that it is true.
7. Recency
A page with a visible, recent update date beats an identical page without one. For questions where recency matters — prices, regulation, technology — it is decisive. So put the date on the page, and make sure it is honest.
8. Structure that can be summarised
Headings that say what sits under them. Lists where it is a list. Tables where it is a comparison. Short paragraphs. This is not aesthetics: a model that can cleanly segment a page can cite one segment from it with more confidence.
9. Crawler access
The most banal point and surprisingly often the decisive one. If your content only appears after three seconds of JavaScript, or your robots.txt blocks AI crawlers because a plugin once switched that on by default, none of the eight points above matter. Check this first.
| Signal | Effort | Effect | Speed |
|---|---|---|---|
| Verify crawler access | Very low | All or nothing | Immediate |
| Data consistency | Low but tedious | High | Weeks |
| Structured data | Medium | High | Weeks |
| Answering questions literally | Medium | High | Weeks |
| Rewriting unambiguously | Medium | Medium to high | Weeks |
| Entity building | High | Very high | Months |
| External corroboration | High | Very high | Months |
Where to start
Top to bottom in that table. First check whether you are allowed to be read at all, then align your data, then make your facts machine-readable. Only then start the slow work — because entity building is the part you cannot catch up on once it becomes urgent.
How these signals fit together is set out in my analysis of the agentic web and on the page about answer engine optimization.
Frequently asked questions
Can I submit my website to ChatGPT?
No, there is no submission process. AI assistants draw on their training data and, increasingly, on live web sources they fetch themselves. You influence that indirectly: by being present, clear and consistent across the sources they consult.
Do backlinks still matter for AI visibility?
Indirectly. Classic links do not count the way they do for Google rankings, but mentions in independent sources weigh heavily: repetition of the same facts in multiple places gives a model confidence. It is the mention that matters more than the link itself.
How do I know whether AI names my business?
By measuring it. Write a fixed set of ten to twenty realistic customer questions, put them to ChatGPT, Claude, Perplexity and Gemini monthly, and record who gets named. It is manual work, but it is the only reliable measurement available today.
See where you stand today
I put five customer questions to four AI assistants and send you the literal answers — including who is recommended in your place. Free, within 48 hours.