Does ChatGPT Recommend Your Business? A 5-Step Test

If you have read our article on what AEO is, you know the theory: customers no longer pick from ten blue links, they ask an AI and get a single answer. But the question that follows is always the same: "So where does my business stand right now?"

That question is not guessed, it is measured. The test below exists for exactly that. You do not need to buy any tool; the free tier of ChatGPT, Gemini or Perplexity plus access to your own site files is enough.

Visibility you have not measured is visibility you only assume you have.

Before you start: two rules

• Use a private window. Your own account carries chat history and memory; if the model already knows you, the test is contaminated. Ask while logged out, in a clean window.
• Never judge from one attempt. AI answers are probabilistic — the same question can return different results twice. Ask each question at least three times and look for names that repeat.

Step 1 — Ask without naming your brand

This is the most important step. Your customer does not know you yet; all they know is their need. So phrase their sentence:

• "Can you recommend a good teeth-whitening clinic in Bursa?"
• "I'm looking for a reliable auto electrician in Kütahya, who would you suggest?"
• "I want to rent out a villa to foreign guests — which agency should I work with?"

City + service + intent — that is how customers ask. Write down the names in order. That list is your real competitor list — not the one on Google, the one inside the AI's head. The two are often not the same.

Step 2 — Now give your name

Type your business's full name and ask "what do you know about this business?" Put the answer in one of three boxes:

1. Accurate and complete: it knows your services, your location, what makes you different. Best case.
2. Empty: "I don't have information about this business." Bad but honest — and the easiest to fix.
3. Wrong: it thinks it knows you but gets the address, the service or the industry wrong. This is the dangerous one, because wrong information is delivered to customers confidently and you never hear about it.

Step 3 — Competitor check: why them?

Ask about each name from step one, then add: "Where did you get this from — what source are you relying on?"

On engines that cite sources, such as Perplexity, this gives you direct links. Look at what type of page gets cited: their own website, a directory listing, a map record, or a blog post? Whichever door your competitor came through is open to you too — and the most frequently cited item is usually a page on their own site that answers a question plainly.

Step 4 — Is the door to your own site open?

The first three steps looked from the outside. Now look at your side. Three files, three checks:

1. yoursite.com/robots.txt — open it in a browser. Is there a Disallow: / line for GPTBot, ClaudeBot, PerplexityBot or Google-Extended? If so, you closed the door yourself. This is a one-line mistake that sits unnoticed on many sites.
2. Structured data — on your homepage, right-click → "View page source" → search for application/ld+json. If there are no hits, your site is not describing itself to machines; even if the model reads the page, it has to guess what you are.
3. yoursite.com/llms.txt — if this returns 404, you have no summary prepared for AI models. It is not mandatory, but it is cheap and it works.

Step 5 — Has AI ever read your site?

This step gives the definitive answer. In your hosting panel's access logs (or your analytics tool, if it does not filter bots), search the last 30 days for these user agents: GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Bingbot.

If none appear, the picture is clear: AI has never visited your site. In that case ChatGPT not recommending you is not a ranking problem, it is an existence problem — the model does not know you because it never read you.

We track this traffic as its own signal on our own site; seeing which pages AI agents read is the most direct way to learn which content is working. The full measurement loop is described in our article on how we built the studio.

How to read the result

In practice the test produces three pictures:

• You are invisible (absent in step 1, empty in step 2, no bots in step 5): the problem is access, not content. Start with robots.txt and structured data. This is usually the fastest win.
• You are known but not recommended (step 2 full, absent in step 1): the model knows you but does not think of you when asked for "the best". What is missing is content that directly answers the customer's question, plus authority.
• You are known incorrectly (step 2 wrong): the most urgent case. Until you publish the correct information in marked-up form on your own site, the model keeps guessing.

Frequently Asked Questions