Traditional SEO tools measure two things: where you rank and who clicks. AI answers break both.
There's no rank to track: An AI engine doesn't return ten blue links. It writes a paragraph, and your brand is either in it or not. The same question asked twice can produce two different answers, so there's no fixed position to monitor.
There's often no click: So for instance, if ChatGPT or Gemini recommends three tools (and lists the name of each tool), the user may directly click on one of the listed tools or later search for the brand name of one of the tools in a regular search. The visit from that user will show up in your analytics as either direct traffic or as branded traffic but not be tied back to the AI answer that triggered the recommendation in the first place. We covered the broader issue of zero-click search in a prior post.
Google's own AI features don't help much either. : Traffic from Google’s AI Overview and AI Mode features in Google Search is reported in your normal web traffic in Search Console and can’t be split out to report how many times you appeared in the AI-generated summary versus normal search results.
The upshot: if you want to know how AI engines talk about your brand, you have to ask them directly and record what they say. That's what an AI share of voice audit does.
The five numbers that actually matter
"Are we in ChatGPT?" is a yes/no question. It's a starting point, but it won't tell you much. Most single "AI visibility scores" also blend several different signals into one number, a problem Search Engine Journal flagged in its post on how brands measure AI visibility. Track these five separately instead.
Mention rate and citation rate are different things, and the gap between them tells you a lot. A brand with a high mention rate but low citation rate is being talked about, but the AI is getting its information from somewhere else, usually a review site, a Reddit thread or a competitor's comparison page. You don't control that narrative.
Share of voice is the number leadership will care about. The standard formula, as Semrush lays out in its guide to measuring AI share of voice, is your brand's mentions divided by total mentions across all brands in your category. It turns a pile of answers into one figure you can track month over month and compare against the two or three competitors that matter.
Step 1: Build a prompt set your buyers would actually type
Your audit is only as good as the questions you ask. The most common mistake we see is testing branded prompts like "What is [Brand]? " That tells you whether the AI knows you exist. It doesn't tell you whether it recommends you to someone who's never heard of you.
Build your prompt set around four types of question:
Category discovery. "Best [category] for [use case]." Example: "best accounting software for small manufacturers in India".
Problem-led. The buyer describes a pain, not a product. "How do I stop losing leads between my website and my sales team?"
Comparison: "[Your brand] vs [competitor]" and "alternatives to [competitor]". The second one matters more than most teams realise.
Decision detail. Pricing, integrations, eligibility, reviews. "Which credit cards give lounge access with no annual fee?"
The set should contain around 30-50 single prompts. Too few, and one odd answer throws off your averages. And if you have more than 100 single prompts, the manual work does not really have any value unless you are using a tracking tool as well.
Here are two insights from testing this for clients. First, prompt a chatbot much as one would converse with it – longer and more specific than a typical Google search query. Second, use real language from your sales calls, support tickets, and questions posed on subreddits and Quora in your industry. These are the types of questions that chatbot AI engines are currently being asked.
Step 2: Run the audit across engines
Don't test one engine and assume the rest behave the same way. They don't. ChatGPT, Perplexity, Gemini and Google's AI Overviews pull from different sources and weigh them differently. We saw this firsthand when we asked five AI engines the same question. A brand can do well in one and be invisible in another.
A few rules keep the results honest:
Use a clean session. Log out, or use a fresh account with memory and personalisation switched off. Otherwise the AI may lean on your own past chats.
Run each prompt more than once. AI answers vary between runs. Three runs per prompt per engine gives you a fair read without tripling your workload for nothing.
Set the location. If you sell in India, test from India. Recommendations for "best health insurance" look very different in Mumbai and Texas.
Log everything in one sheet. One row per prompt, per engine, per run. Columns for: mentioned (yes/no), position, cited URL, competitors named, sentiment, and any factual errors.
Screenshot the odd ones. When an answer gets your pricing wrong or recommends a discontinued product, save it. That's evidence you'll need when you go to fix it.
On cadence: run a full baseline once, then repeat monthly. AI answers shift as models update and new content gets indexed, so a single audit goes stale fast. If you're tracking more than 100 prompts, paid tools can take over the repetitive part.
Step 3: Read the citations, not just the mentions
This is where most of the useful findings come from. Engines like Perplexity, Gemini and ChatGPT with search turned on show the sources behind their answers. Sort those sources into buckets and a pattern usually jumps out.
In our audits for BFSI and fintech brands, the forum bucket is often bigger than the client expects. A three-year-old Reddit thread can outweigh a polished product page because it reads like an honest answer from a real person. The data backs this up: when Semrush analysed 248,000 Reddit posts cited in AI search, most of the cited posts had fewer than 20 upvotes. You don't need a viral thread to shape the answer, just a relevant one.
Also look at what's missing. If the AI never cites your pricing page, check whether it's crawlable, whether it uses JSON-LD structured data, and whether the key facts sit in plain text rather than inside an image or a script. Slow, script-heavy pages are a common culprit here, which we covered in our post on site speed and architecture as a ranking factor.
Step 4: Turn the findings into a fix list
An audit that ends in a spreadsheet hasn't done its job yet. Match each pattern you found to a specific action.
Prioritise by revenue, not by volume. A gap on a high-intent decision prompt ("which plan should I pick") is worth more than a gap on a broad awareness prompt. Fix those first, then re-run the same prompts next month to see what moved.
You can't improve what you've never checked
Search visibility for most brands has been measured for years with considerable accuracy. AI answers to questions have, however, proved to be beyond current precision. The method of measurement described here is simple to use. You can begin to measure your brand visibility AI answers within minutes.
These companies will have a huge competitive advantage over their peers and will have established a key performance indicator to measure against on an ongoing basis.
Get a head start on your competitors with MyAibo’s AI visibility audits for ChatGPT, Perplexity, Gemini and Google AI Overviews. Find out your share of voice against your competitors. Get a breakdown of the sources that matter most in your category. And get a prioritized fix list of things you need to do to improve.