AI Visibility Audit - How to Measure Whether ChatGPT Recommends Your Brand (AI Share of Voice)

An AI visibility audit (a GEO audit) checks whether ChatGPT, Perplexity, Google AI Overviews and Copilot mention, cite and recommend your brand - and how you compare with competitors. The key metric is AI Share of Voice: the share of answers to your customers' questions in which your brand appears. A company ranking first in Google can have a zero share of AI answers, and classic SEO tools won't show it. An audit is the only way to find out before your sales team notices.

Do ChatGPT, Perplexity and AI Overviews recommend your brand or your competitors? Google's and Bing's 2026 reports, building a question set, measuring at three levels with a script, calculating and reading AI Share of Voice, and a one-day audit.

In this guide I show what Google's and Bing's official reports tell you today, what the audit measures, how to build a question set, how to measure at three levels (manual, semi-automated and automated), how to calculate and read the results and how to turn them into a plan. At the end you'll find a one-day version of the audit.

What official reports offer in 2026

Platforms have started sharing the first data on AI visibility, but only partly:

• Google Search Console. Since June 2026, a report shows impressions in AI Overviews, AI Mode and AI features in Discover, broken down by page, country and date. At the end of August 2026, Google made it available to all sites. But it shows impressions only - no clicks and no queries.
• Bing Webmaster Tools. Since February 2026, the AI Performance report shows how many times Copilot cited your site, which URLs and for which retrieval queries (grounding queries).

ChatGPT and Perplexity don't share any data with site owners. That's why an audit still requires actively asking the engines questions.

What the audit measures

You calculate AI Share of Voice simply: the number of answers mentioning the brand divided by the number of questions asked, times 100. Example: 100 questions in ChatGPT, the brand appears in 28 answers, so the share is 28%. The same questions in Perplexity produce 41 mentions, so 41%. Calculate the share for each engine separately and combined - the combined figure shows overall visibility, and the separate ones show where the gaps are.

How to build the question set

An audit is only as good as its questions. The set should reflect the customer's path to purchase, not your SEO keywords. Ask in full sentences, in the customer's language - the way people actually talk to AI.

The best sources of questions are conversations with salespeople ("what do customers ask before they buy?"), support ticket history and Search Console queries turned into full questions. Twenty questions are enough for a first audit; reliable measurement needs 30-100.

Measuring at three levels

Level 1 - manual

Ask each question in ChatGPT with search, Perplexity and Google AI Mode, and record in a spreadsheet whether the brand appeared, how it was described, which competitors appeared and whether your site is among the sources. Model answers vary, so ask each question 2-3 times. For 60 questions and three engines you need 3-4 hours.

Level 2 - semi-automated with the Perplexity API

The Perplexity API returns a list of sources along with the answer, so you can check programmatically whether your brand and domain appear:

import os, requestsAPI_KEY = os.environ["PERPLEXITY_API_KEY"]def check(question, brand, domain):    r = requests.post(        "https://api.perplexity.ai/chat/completions",        headers={"Authorization": f"Bearer {API_KEY}"},        json={"model": "sonar", "messages": [{"role": "user", "content": question}]},        timeout=60,    ).json()    answer = r["choices"][0]["message"]["content"]    sources = r.get("citations") or [s["url"] for s in r.get("search_results", [])]    return {        "question": question,        "brand_mentioned": brand.lower() in answer.lower(),        "domain_cited": any(domain in url for url in sources),    }questions = [    "Which companies do you recommend for AI automation projects in Poland?",    "How do I automate invoice processing in a small business?",]results = [check(q, "Your Brand", "yourbrand.com") for q in questions]sov = 100 * sum(r["brand_mentioned"] for r in results) / len(results)print(f"Share of Voice in Perplexity: {sov:.0f}%")

Run the script on the same question set every week or month and save the results to see the trend.

Level 3 - automated monitoring

With regular measurement, manual tests stop scaling. You have two options:

• Ready-made tools - Profound, Otterly, Peec AI, AthenaHQ, Scrunch (acquired by Sitecore in 2026) and modules in SEO tools such as the Semrush AI Visibility Toolkit and Ahrefs Brand Radar. They differ in method: some generate questions synthetically, others rely on real queries. I compare prices and selection criteria in the post on AI visibility monitoring tools.
• Your own system - I built AnswerLyzer for this. It queries Gemini, OpenAI models and Perplexity at set intervals, a separate model assesses each answer (recommendation, neutral mention or negative comparison), and the system shows the share trend against competitors and sends an alert when the share drops by more than 5 percentage points.

Where engines get their sources

Audit results are easier to understand when you know where each engine retrieves pages from:

• Google AI Overviews and AI Mode use Google's index and split a question into many sub-questions (query fan-out).
• ChatGPT has its own search crawler but also relies heavily on Bing. A Seer Interactive study found that 87% of citations in ChatGPT search matched Bing's top results.
• Copilot uses the Bing index.
• Perplexity has its own crawler and its own search.

The takeaway: you can't do GEO while ignoring Bing. Indexing in Bing Webmaster Tools unlocks ChatGPT and Copilot at the same time, yet most companies only check Google. Details in the post on Bing, Copilot and IndexNow.

How to read the results

Presence isn't everything. A negative mention is worse than none, because it sends the customer straight to a competitor. So record the sentiment for every answer.

Then set priorities. Draw a simple matrix: one axis for the value of the question (how close it is to a purchase decision), the other for how hard the gap is to close. Start with high-value, low-difficulty questions - usually technical issues such as missing Bing indexing, a blocked crawler or a page with no direct answer.

Why AI cites competitors instead of you

When an engine cites a competitor for a question, open their page and read it the way a model would. The most common reasons for their advantage, from easiest to fix:

• The competitor is in Bing's index and you're not.
• Their content is easy to cite - a direct answer, data and a table, while yours is generic.
• Others write about them - media, rankings, forums, Wikipedia.
• Their content is fresher for date-sensitive questions.

That's your ready-made task list. I show how to write fragments engines cite in the post on how to write content AI cites.

The most dangerous audit result is AI saying something untrue about you - an invented offer, a wrong price or confusion with another company. I cover where such errors come from and how to correct them in the post on brand reputation in AI.

How often to measure

Outside the schedule, measure after a major model update, after an important competitor publication, after your own publication, and when salespeople hear from customers that "AI recommended someone else".

A one-day audit

You don't need to buy a tool right away. You can do your first audit by hand in one day:

1. 1.Write down 20 questions your customers ask, from all six categories.
2. 2.Ask them in ChatGPT, Perplexity and Google AI Mode, each 2-3 times.
3. 3.Record in a spreadsheet these columns: question, engine, whether the brand appeared, sentiment, whether your site is among the sources, which competitors appeared and which domains the engine cited.
4. 4.Calculate the share for yourself and two or three competitors.
5. 5.Pick three gaps with the highest value and lowest difficulty and start with them.

The column of cited domains is the most valuable part of the spreadsheet - it shows where to earn a mention or what content to prepare.

---

I run AI visibility audits with a concrete improvement plan as part of AI optimization (GEO), and for regular measurement I use my own tool, AnswerLyzer. I also teach it in the SEO & GEO course. Get in touch - I'll check whether AI recommends you or your competitors.

Worth reading next:

• How to Get Cited in ChatGPT and Perplexity - How AI Engines Choose Sources and What to Do
• Brand Reputation in AI - What to Do When ChatGPT Says Something Untrue About Your Company
• How to Measure Traffic from ChatGPT, Perplexity and AI Mode - Analytics for the AI Era (GA4 + Server Logs)
• AI Visibility Monitoring Tools - Profound, Peec, Otterly, Semrush and the "Build Your Own" Option (Comparison)