How to get your brand cited by ChatGPT: a 7-step implementation checklist

Last week we did something straightforward with a B2B SaaS customer. We opened ChatGPT, asked it 'What vendors in Taiwan provide this kind of service?' and watched it list five names with confidence, each backed by reasoning. The customer wasn't on the list.

They had been in this category longer than most. Their website ranks second on page two for SEO. But in ChatGPT's answer, they don't exist. It wasn't a product problem. The issue was simpler: ChatGPT's answer isn't a blue link you click through. It's rewritten text. And you haven't given it the material it needs to rewrite.

What does being cited actually mean?

Here's a working definition. Getting your brand cited by ChatGPT means making large language models treat your content as a trustworthy, extractable source when they generate answers.

Traditional SEO and what's now called generative engine optimization compete on different grounds. SEO wins on ranking positions. GEO wins on being written into the answer. With SEO, you get clicks. With GEO, you get mentions. Users who get their answer often leave without clicking through, which is why GEO can feel awkward at first. You're optimizing for a position that may not drive traffic but determines whether the AI mentions your company at all.

Breaking it into 7 steps: the implementation checklist for getting cited by ChatGPT

We built this into a checklist you can execute. The key isn't just what to do, but that every step is measurable. Without visibility checks, you won't know if you're progressing.

Step	What to do	How to verify (visibility check)
1. Establish baseline	Ask ChatGPT, Perplexity, and Gemini 20-30 real questions from your category. Record whether you appear and who they cite instead.	Calculate your "mention rate" and how far behind top three competitors you are.
2. Make content extractable	Lead paragraphs with conclusions, use Q&A headlines, provide one-line definitions, enable LLMs to grab complete sections.	Pick a random passage, ask an LLM "what's this section about?" and check if it extracts right.
3. Unify your entity	Make your brand, founder, and products described consistently across your website, Wikipedia, LinkedIn, and industry directories.	Ask an AI "who is this company?" and see if it pieces together a complete, accurate picture.
4. Get into the source pool	Win mentions from third-party authority sites, industry media, G2, Reddit, Wikipedia.	Check Perplexity's citation list. Do you appear, or pages that mention you?
5. Add the machine-readable layer	Use schema.org markup, set up llms.txt, allow GPTBot/OAI-SearchBot/PerplexityBot, provide clean markdown.	Crawl your own pages with a bot UA string and verify you're getting clean text.
6. Stack named facts for value	Use original data, explicit numbers, verifiable claims, replace adjectives.	For every key claim, check if there's one sentence a model would "dare copy verbatim."
7. Build the monitoring loop	Re-run your question set monthly. Track mention rates and source shifts, fill content gaps.	Month-over-month mention rate changes. Are new sources entering the list?

A few easy-to-skip steps that are actually critical

Step 2 is the dividing line. LLMs don't extract whole pages. They extract paragraph-level snippets. If your key point is buried in the second half of paragraph three, the model will skip the whole thing. Put your conclusion at the start. A single sentence can stand alone and get cited. That alone lifts your hit rate noticeably.

Step 3, entity consistency, is underestimated. Models stitch scattered fragments into recognizing a single company. If your name, positioning, and focus differ across platforms, the model won't connect them and won't cite you.

Step 5 is pure mechanics, but skip it and everything before it falls apart. If your site blocks AI crawlers or your content only renders via JavaScript, the first six steps don't matter. The model can't read it.

We used ourselves as test subjects

Last year we ran this checklist on Tenten. We started by asking Perplexity, 'forward deployed engineering consultants in Taiwan.' Tenten wasn't in the answer. It cited a few English vendor blogs instead.

We made three changes. We rewrote our FDE and FDM definition pages to lead with conclusions and use Q&A structure (Step 2). We added consistent entity descriptions in Chinese and English so models understand Tenten as 'forward deployed enterprise AI consultants who embed engineers on-site,' not an AI product studio (Step 3). We added llms.txt and negotiated clean markdown so crawlers get plain text directly (Step 5).

About six to eight weeks later, Tenten started appearing in the candidate list for that same set of questions and got marked as a citation source. We didn't win every question. Chinese-category questions performed better than English ones. We're still filling English long-tail gaps. That's why we're sharing this: being cited isn't a one-time switch. It's a curve you feed continuously.

This isn't a one-time project

There's a clear tradeoff: AI citations dilute your clicks. Users get their answer and leave without visiting your site. But at the front end of a purchase decision, the AI saying your name unprompted carries real weight.

Don't treat this as a project you complete and shelve. Models update every few weeks. The source pool shifts constantly. Being cited today doesn't guarantee you're cited next quarter. At Tenten, we treat GEO as a production line we monitor continuously. Engineers set up the monitoring loop, we check mention rates monthly, and if a question drops, we patch it. What counts is ChatGPT actually writing you into its answer and your staying there next month.