How to check the citation status of your brand in ChatGPT and Perplexity (with prompt word template)

I casually asked on ChatGPT "Recommend some SaaS consultants in Taiwan" and I breathed a sigh of relief when I saw my name. This is not an audit, but a lottery. If you change the account, turn off the memory, and ask the same sentence the next day, the answer will probably change. What the audit depends on is not "whether you have been mentioned", but under fixed conditions, how often you are mentioned, the order in which you are placed, and whether there is a clickable source link attached. Only when these three things can be quantified can it be called auditing.

Let’s first distinguish the three levels of “citation”

Many people confuse three completely different situations, leading to distorted audit results. The first is that the model mentions your name in the generated text, but there is no link; the second is that Perplexity or ChatGPT that opens the search lists the source card next to the answer, and the user can click to enter your website; the third is that the model directly puts you as the first choice when "recommending a company". The commercial values ​​of the three are quite different and should be recorded separately during the audit.

• Plain text mention: The name appears in the answer, with no link or ranking meaning. This is a basic offering, but it can easily be diluted by competing products.
• Source citation: Your URL card appears next to the answer, allowing users to click on it. This is where GEO really wants to compete.
• Preferred recommendation: The model will be the first to talk about you when "recommending only one company", or only talk about you in the entire paragraph. The rarest and most valuable signal in auditing.

Before starting to ask, first fix the four variables

Auditing is like conducting experiments. If the variables are not fixed, the data will be meaningless. If you ask using a login account today and ask using an incognito window tomorrow, the personalized memory for model calling is different, and the result will naturally be disappointing. Before going for an audit, write down the following four things.

• Account and memory: Always use logged out or incognito status, and turn off ChatGPT’s “Reference Chat History” to eliminate personalization contamination.
• Platform and mode: Run ChatGPT (including search mode), Perplexity, Google AI Overviews, and Gemini separately, and record them separately without mixing them up.
• Language and Region: Ask in the language of your target customers. Taiwan B2B uses traditional Chinese to ask questions, and the answers to English questions cannot be used to represent the Taiwan market.
• Question time: Try to finish the same batch of prompt words within the same day. Model indexing and instant retrieval change every day, and data across weeks is not comparable.

Prompt word template: five ways to ask questions, copy directly

One question cannot tell the whole story. Real users will approach the same need from different angles: some ask about categories, some name and compare, and some bring situations. The following five types of prompts cover the buyer’s journey from “I don’t know you yet” to “I’m already considering you.” Replace [category] [brand] [situation] with your own words.

1. Category exploration type: "I want to find [category, such as B2B GEO/SEO agencies in Taiwan]. Please recommend a few and explain what they are good at." See if you will appear in the unnamed list.
2. Name comparison type: "Compared with [main competing products], [your brand], what is the difference in [service items]?" See if the model recognizes you and whether it is correct.
3. Situational demand type: "I am an [industry] SaaS company. I want to increase my visibility in AI search. Who should I contact and how should I start?" See if you will be led into the solution.
4. Fact check: "What services does [your brand] provide? What is the price and scope of services?" Check whether the content of the model is correct and whether it is out of date.
5. Source request type: "Please recommend [category] and attach the source URL you refer to." Specifically force out the source citation level to see if your website has been linked out.

ChatGPT and Perplexity should be looked at separately

The reference mechanism of the two is different. Almost every answer in Perplexity will be accompanied by a source number. When auditing, you can directly count in which article your website appears and how many times it has been cited. The signal is very clean. ChatGPT needs to distinguish between pure model answers and search-enabled answers: the former relies on training and memory, and whether you can be mentioned depends on long-term brand reputation; the latter will instantly retrieve web pages and attach links, which is closer to Perplexity. When running ChatGPT, the two modes ask each question once. The gap in between is intelligence in itself.

Turn answers into scores: a table is enough

Don’t rely on impressions after asking. Open a table, put four platforms horizontally and five types of prompt words vertically. Only fill in three things in each box. When you run this table for the second or third time, you will have a trend instead of a single point.

• Mentioned or Not (0 or 1): Does your brand appear in the answers to this box?
• Citation level (1 to 3): plain text mention 1, source link attached 2, listed as first choice 3.
• Competitive product coverage: Which competing products appear in the same grid, and whether they are ranked in front of or behind you.

Deducing what needs to be fixed from the audit gap

The value of auditing is that it points to specific practices, rather than vague suggestions like "write more articles." If you can’t even get pure text mentions, the problem is probably too low in the brand’s discussion density, and what needs to be supplemented is external evidence and exposure opportunities. If there are mentions but almost no source citations, the questions are usually difficult to extract from the page itself. There is no clear question and answer structure, no schema, and key facts are hidden in pictures or lengthy paragraphs. The former is a sound volume issue, and the latter is a technical issue. The two roads need to be repaired with completely different things.

If you run a round of this audit yourself, you will be able to tell whether the gap lies in volume or structure in about half a day. If what you want is a complete version that organizes the twenty-grid results, competitive product comparison and specific repair list, Tenten GEO's 30-day GEO audit is to make this process in-depth and complete. If you want to confirm where your gaps are first, you can make an appointment for a 30-minute GEO diagnosis. We will use your brand to actually ask you a few rounds.