Generative engines have changed how people find local businesses. When someone asks ChatGPT, Perplexity, or Google for a recommendation, they often get a direct answer with a few named sources instead of ten blue links. If your business is not cited, you are invisible for that question, no matter how well your old rank report looks.
Why AI citations need their own measurement
Traditional search rankings tell you where a page sits for a keyword. Generative engines do not work that way. They synthesize answers from multiple pages and may cite a source in one response but not the next, even for the same question. A citation is not a stable position. It is an event: the model selected your page as a useful reference for a specific prompt.
That means you cannot buy a dashboard that shows your universal "AI rank." The honest signal is a log of prompts that actually produced a citation for your business. That log is messier than a rank checker, but it matches how customers ask questions in real life. You want evidence that your content is being used, not a vanity number that looks good in a report.
The tracking sheet that takes ten minutes a week
Use a simple spreadsheet or a page in a notebook. The goal is not to build a complex system. You need a repeatable set of prompts and a consistent way to record what happens.
Create columns for:
- Date
- Platform (ChatGPT, Perplexity, Google AI Overview)
- Prompt used
- Cited? (yes/no)
- URL or page cited
- Quote snippet
- Notes
Keep 10 to 15 buyer prompts that fit your business. For example, a local accountant might test prompts like "how much does a small business tax return cost in [city]," "best accountant for ecommerce near me," and "what should I ask before hiring a bookkeeper." Run the same prompts once a week. That stability is what lets you see change over time. If you change the prompts every week, you are collecting anecdotes, not data.
Log only clear citations: your business name, a direct link to your site, or a quoted phrase that clearly refers to you. Do not log "my competitor appeared." That noise will waste your time and make the log feel like a scoreboard.
What to log for each citation
The prompt matters more than the platform. Generative systems respond to exact wording, including location, intent, and conversational context. Record the prompt exactly as you typed it, not a cleaned-up version. If you later see a citation, you need to know what question triggered it.
For each citation, capture four facts:
- The platform where it appeared.
- The exact prompt.
- The page on your site that was cited, if shown.
- The type of citation: a brand mention, a direct link, a source card, or a quoted text block.
Add a short quote snippet. This is the most useful part of the log. When ChatGPT quotes three lines from your pricing page, you learn that your pricing page is clear enough to be used as a source. When Perplexity paraphrases your service area description, you know the model trusts that piece of structured information. Later, you can write more of what works and fix what does not.
How to turn logs into a monthly action list
After four weekly logs, you will have a small sample. Do not overreact to a single week. Look for patterns across prompts and platforms.
For example, if your contact page or service page gets cited in five different prompts about pricing, that page is doing real work. Add a concise pricing section to the other main service pages. If you never appear for a local prompt, check whether your business name, address, and service area are stated plainly on your site. If a competitor is cited again and again for a specific question, open the response and read what the model says about them. Then answer that same question more directly on your own page.
This monthly review should produce one or two concrete content tasks, not a long abstract to-do list. If you are unsure how to prioritize the gaps, our services page describes how we work with small businesses on this kind of content and visibility work.
When to ignore the numbers
Do not treat total citations as a score. One citation for a high-intent prompt like "emergency AC repair available now in my area" is worth more than a dozen citations for broad research questions that never lead to a call. The context of the prompt tells you whether the citation matters.
Also accept that generative output is probabilistic. The same prompt can produce a citation on Monday and nothing on Thursday. That is not a failure. It is how these systems work. If you track daily or worry about every change, you will chase noise. Weekly or twice weekly is enough. Direction over four to six weeks is the signal.
Finally, do not try to appear in every AI answer. The goal is to be the named source when a buyer is actively comparing or ready to hire. A small log kept honestly will show you whether that is happening more often over time. For more on how we think about AI visibility and practical marketing, see our blog.