1 Why You Should Regularly Test Your AI Visibility
Most online retailers have no idea whether AI systems know their store. Yet this question is as relevant today as "Is my store on page 1 of Google?" was ten years ago.
A G2 survey from April 2026 shows just how relevant: 51% of B2B buyers now start their research in AI chatbots more often than in traditional search. If you don't show up there, you simply don't exist in these customers' buying decisions.
A regular AI visibility test shows you:
- Whether AI systems use your store as a source
- Which products are recommended – and which are not
- How you compare to competitors
- Whether your optimizations are having an effect
2 Prompt Strategy: Questions With and Without Your Brand Name
A reliable test separates two types of questions. Branded questions show what the models know about you and whether the details are correct. Unbranded questions show whether you appear at all when someone only describes a problem or a product category – which is the normal case with new customers. Tests 1 to 3 below apply this split.
Branded questions (awareness and accuracy):
→ "What does [your store] offer and for whom?"
→ "Is [your store] a reputable supplier for [product category]?"
→ "What are the delivery times and return conditions at [your store]?"
→ "How does [your store] differ from [competitor]?"
Unbranded questions (discoverability):
→ "Which online store is recommended for [product category] in Germany?"
→ "I'm looking for [product] with [feature]. Where can I get it?"
→ "Which suppliers deliver [product category] to business customers on invoice?"
→ "Who offers [product] with delivery to [region]?"
Write both lists down and do not change the wording afterwards. That is the only way to compare answers over months. For each prompt and AI system, record: mentioned yes or no, position in the answer, cited source, errors in the details.
3 Test 1: Direct Product Queries in ChatGPT
Start with ChatGPT (or another AI system) and ask typical purchase queries from your industry. Use the web search feature for this.
Example prompts for an equestrian supply store:
→ "Which online store has the best selection of riding helmets?"
→ "Recommend me a reputable store for equestrian supplies in Germany."
→ "Where can I best buy [product name] online?"
Evaluation criteria:
- ✓ Is your store mentioned by name?
- ✓ Are specific products with prices mentioned?
- ✓ Is a link to your store provided?
- ✗ Is only the competition recommended?
- ✗ Is your store not mentioned at all?
4 Test 2: Check Company Awareness
Ask AI systems directly about your company:
→ "What do you know about [company name]?"
→ "Is [company name] a reputable online store?"
→ "What products does [company name] offer?"
Pay attention to: Is the information correct? Are current products mentioned? Is the address right? Incorrect or outdated details usually come from outdated third-party sources or from gaps on your own website – check which source the model cites.
5 Test 3: Competitor Comparison
Compare how AI systems rate you versus your competitors:
→ "Compare [your store] with [competitor] for [product category]."
→ "Which stores for [your niche] in Germany would you recommend?"
If competitors are recommended and you're not, analyze their websites and the cited sources: Do their pages answer the question directly? Do they have more complete product data? Are they mentioned in review portals, trade media, or directories that the model lists as sources?
6 Test 4: Technical Check
In addition to AI queries, you should check the technical basics:
Are AI crawlers (GPTBot, ClaudeBot, Google-Extended) allowed?
Do bot protection, CDN, or firewall block the AI crawlers? Do the server logs show requests from GPTBot and others?
Is Product markup with price and availability present on product pages?
Is the XML sitemap up to date and linking all important pages?
Can AI crawlers load the pages quickly enough?
7 Test 5: Check the AI Surfaces Separately
Chatbot answers are only part of the picture. Google now shows AI Overviews on roughly half of all queries (studies vary) – for B2B tech queries, the share is 82% according to Semrush (2026). On top of that, Google AI Mode is a separate surface: only about 14% of its cited sources overlap with the sources cited in AI Overviews. If you only check one surface, you don't have the full picture.
Search your most important queries on Google and check whether an AI Overview appears – and whether your store is linked as a source.
Run the same queries in AI Mode. The cited sources differ significantly from AI Overviews – document both separately.
ChatGPT, Claude, and Perplexity with web search enabled – tests 1 to 3 from this article.
Why the effort pays off
AI Overviews reduce organic click-through rates by up to around 60%, depending on the query type (recent studies by Ahrefs and Seer) – branded searches can even benefit. At the same time, visitors arriving via AI search are, according to Semrush, about 4.4 times more valuable than average organic visitors. Visibility in the AI surfaces increasingly determines how much qualified traffic reaches you.
8 Document and Improve Results
Create a simple scorecard for your AI visibility test:
Tip: Run the test monthly and document the results. This way you can identify trends and measure the impact of your optimizations.
9 Measuring Through Self-Reporting: "How Did You Hear About Us?"
Chatbot tests show what the models answer – not whether that turns into customers. AI chatbots rarely pass on a usable referrer. Someone who reads a recommendation in ChatGPT often types the address directly or searches Google for the name. In your web analytics, that shows up as direct traffic or a branded search, not as AI traffic.
The simplest solution is to ask directly. Add a field "How did you hear about us?" to your contact form – as free text, not a fixed selection, so that answers like "ChatGPT recommended you" are possible at all. The same question belongs in the shipping confirmation email after an order or on the order confirmation page.
Evaluating the free-text answers:
- → Review monthly and count mentions of ChatGPT, Perplexity, Gemini, Claude, or simply "AI"
- → Record the share of all enquiries or orders – as a trend, not an exact metric
- → Compare with the prompt set: does what customers report match what the models answer in your tests?
- → Note the customers' wording – it shows which questions the models actually answered
10 What to Do With Your Test Results
Test results usually follow recurring patterns. This mapping helps you translate each finding into the right action:
Finding: Incorrect or incomplete product and price information
Typical cause: missing or faulty structured data. Implement Product, Offer, and Organization markup according to Schema.org and validate it with Google's Rich Results Test. Guide: Schema.org for AI recommendations →
Finding: Outdated information (old prices, discontinued products, old address)
Typical cause: missing freshness signals. Maintain a visible last-updated date on the page and a correct dateModified in your schema markup, and update key content regularly. AI systems demonstrably prefer fresh sources.
Finding: Your store isn't found at all
Typical cause: AI crawlers can't reach your content. Check your robots.txt for permissions for GPTBot, ClaudeBot, Google-Extended, and PerplexityBot, review bot blockers (firewall, CDN), and look at your server logs to see whether AI crawlers are fetching your pages at all.
Finding: Good Google rankings, but no AI citations
Not a contradiction: according to Ahrefs (March 2026), only 38% of AI Overview citations come from pages in the organic top 10 – 62% come from outside. What matters is the quality of individual passages, not the ranking. Add direct answers of 40–60 words, FAQ blocks, tables and lists, and author signals (Person schema).