How AI search engines decide which maritime brands to cite

When a buyer asks ChatGPT “who are the leading bunker suppliers in Rotterdam” or pastes a charter party clause into Claude and asks for context, something specific happens behind the scenes. The model decides what evidence it has, weights that evidence and constructs an answer that cites particular brands and not others. The decision is not random and it is not a black box. It runs on identifiable signals, and once you understand them you can move yourself up the citation list.

The five signals that matter most

1. Retrieval-time presence

For models that retrieve live web content, the question is whether your page appears in the retrieval set the model pulls back when answering the query. This is essentially classical search ranking, but with the AI engine’s own retrieval index rather than Google’s. Bing-backed systems (Copilot, Perplexity for some queries) reward what Bing rewards. Google-backed systems weight what Google weights.

2. Source authority

Not all retrieved pages count equally. A page on tradewindsnews.com or lloydslist.com that mentions your brand contributes more confidence than a page on a low-authority directory. Maritime LLM traces consistently show TradeWinds, Lloyd’s List, Splash, The Loadstar, gCaptain and Marine Insight punching above their weight relative to general web authority metrics.

3. Structural extractability

If your page has clean H2 and H3 headings, schema markup, FAQ blocks and explicit factual statements, the model can pull a quotable claim with confidence. If your page is a wall of marketing prose with no factual anchors, the model often skips it even when it ranks well in retrieval.

4. Cross-source consistency

Models prefer claims they can corroborate. If your website says you operate 84 vessels and a TradeWinds article says 84 vessels and your DNV listing says 84 vessels, that fact is high-confidence and quotable. If your numbers contradict each other across sources, the model often hedges or omits.

5. Training-data presence

For older or more general queries, the model leans on what it learned during training. Brands that were heavily mentioned in maritime trade media, classification society pages, port authority directories and Wikipedia at the time of training cutoff carry an advantage that retrieval cannot fully erase.

What this implies for your work

The work is not glamorous but it is tractable.

• Audit your top ten service pages for structural extractability. If a paragraph cannot be excerpted as a standalone factual claim, rewrite it.
• Standardise your factual base across your website, your LinkedIn company page, your industry directory listings and your press releases. Pick one set of numbers and use them everywhere.
• Pursue tier-one trade press placements specifically. One TradeWinds feature outweighs ten guest posts on lower-authority sites.
• Implement Organization, Service and FAQ schema with real, verifiable data inside, not placeholder content.
• Track where you appear through quarterly LLM-visibility audits using realistic buyer prompts.

A note on prompt sensitivity

Citations are sensitive to phrasing. “Who are the top ship managers” returns a different list to “which ship managers are most reputable” and a different one again to “I need a ship manager for a fleet of LR2 tankers operating in Asia, who should I consider”. The third is the kind of prompt a real buyer writes, and it is the one where structural specificity on your service pages pays the biggest dividend.

The mechanics are not magic. They are a search system with an additional generation layer on top, and the same disciplines that earned visibility on Google ten years ago will earn visibility in chat-based search this year, with the dial turned up on specificity and structure.