How to get cited by ChatGPT, Perplexity, and Claude: breaking down the citation mechanism with a 12-item checklist

Last week a marketing director handed me a report. Their page ranks first on Google, but when they asked ChatGPT which industrial SaaS implementation firms operate in Taiwan, they weren't mentioned once. Instead, the model cited a forum thread and a competitor sitting on page two.

His confusion was clear: How can I rank first on Google but get completely invisible to AI?

Because the logic AI uses to cite you isn't the same logic Google uses to rank you. To get cited in an AI response, you need to understand the three stages between being crawled, being selected, and actually appearing in the answer. Most people focus only on the first stage and fail at the third.

How AI cites sources

When an AI generates an answer, three things happen in order, and the order matters.

Retrieval is the first stage. When a user asks a question, the AI doesn't answer from memory; it searches in real time. ChatGPT uses Bing, Perplexity runs its own crawlers across multiple engines, and Claude uses built-in search. This stage determines whether your page makes the candidate list at all. If your content isn't indexed, requires JavaScript to render, or keeps key information behind a login wall, you're not even in the game.

Ranking is the second stage. Once on the candidate list, the model has to pick from dozens of sources. It's not looking for the densest keyword load; it's looking for what semantically answers the question most directly. This is where traditional SEO and AI diverge: SEO rewards breadth of coverage; AI rewards answer density. A paragraph that opens with a clear conclusion beats an essay that takes three sections to get to the point.

Citation is the third stage. When assembling the answer, the model decides whether to tag you as a source and include a link. This stage is about attributability: can your sentence be cleanly extracted and mapped to a specific fact? Vague sentences that need surrounding context to make sense get absorbed into the model's answer but leave you unattributed. Your content gets used; your brand doesn't get credited.

Each stage has its own failure mode. Retrieval fails on technical issues, ranking fails on structure, citation fails on writing. The checklist below is organized around these three stages.

The 12-item checklist

These 12 items aren't equally weighted. If resources are tight, fix 1 to 4 first; technical blockers kill everything downstream. Next comes 5 to 9, which is what most people skip and where ROI peaks: take the same content, move the conclusion to the top, and your citation rate improves. Items 10 to 12 are long-term work. Item 12 especially: no amount of optimization matters if nobody else is talking about you.

What we learned from testing

When Tenten first wrote about FDM and GEO, we violated items 5 and 9. We buried the definition of front-line deployment marketing in the third paragraph, leading with industry background. Perplexity crawled our page but only cited a fragment of a sentence without including a link. Our content got absorbed, but we got no credit.

We moved the definition to the opening of each article: FDM is deploying marketing engineers into brand environments to make content citable in AI answers. One sentence, extractable, with concrete action. Two months later, the same pages started appearing as cited sources in Perplexity and ChatGPT results for What is FDM? and GEO Taiwan, with links included. We didn't add a single word; we just reorganized how the information was presented to make it extractable.

This convinced us of something: getting cited by AI isn't magic; it's an engineering problem you can deconstruct and fix item by item.

The difference between ranking and citations

Google ranking rewards breadth and authority. AI citations reward extractability: whether each sentence can stand alone and map to a fact. These two things are diverging.

At Tenten, when we work on FDM and GEO, we don't treat it as a writing technique; we treat it as a deployment engineering problem. Just like an FDE pushes an AI system to production, we work through this checklist item by item until your brand actually shows up in someone else's answer.

What counts is getting cited and having people click through. Beautiful demos don't matter.