By John Wheeler, Founder of Wheeler Intelligent Systems
A business owner asks an AI assistant, “What does our company do?” The answer sounds good. It gets the name right, finds the website, and repeats a few lines from the home page. Does that mean the business will show up when a buyer asks for help?
That is encouraging. It is also a very easy test.
A buyer may ask something quite different: “Which company can help us solve this problem?” “Does this product work for our needs?” “What happens after we place an order?” “How does this firm compare with another one?”
Your name may never appear in the question. The buyer may not even know you exist. To be useful in that moment, an AI system needs enough reliable information to recognize where your business fits and explain why. That is a much harder test.
The way to find out is to test the questions buyers really ask. Try some with your company name and some without it. Check whether the answers are accurate, useful, and supported by sources. Then fix the gaps you can control and run the same questions again.
Yesterday, I wrote about whether AI can find your business when a buyer does not know its name. Here is a practical way to test what it knows today.
One good answer can hide many gaps
Most companies have spent years writing for people who already have some idea what they sell. A website might introduce the company, show attractive photos, and invite visitors to call. A salesperson fills in the rest.
That approach can work well when the buyer reaches the salesperson. But suppose the buyer first asks an AI assistant to compare possible suppliers. If the public information stops at “high quality products and exceptional service,” there may be little basis for a useful comparison.
The same issue applies outside sales. A job candidate may ask what it is like to work for you. A potential partner may ask where you operate. A customer may want to understand your support process. A buyer may care about a detail your marketing team has never put on the website.
AI can sometimes fill a gap with an educated guess. That is not the same as finding a fact. A smooth answer can be wrong, incomplete, or based on an old source. The danger is assuming that because the first answer sounds confident, your business is well represented.
Start with the questions people really ask
The best test begins outside the AI tool. Ask your salespeople, customer service staff, and managers what people need to know before they choose you. Look at the questions that slow down a sale or lead to a follow-up call.
Some will be basic: What do you offer? Where do you work? Who is your ideal customer? Others will reveal what makes a decision hard: Does the product meet a specific requirement? Can the service work with an existing system? What information does a buyer need to get started?
Include questions that do not contain your company name. A named question tests whether the assistant can describe you. An unnamed question tests whether it might connect you to a buyer’s need. Both matter, but they tell you different things.
You do not need a huge test to begin. Ten thoughtful questions drawn from real customer conversations can teach you more than a hundred vague questions written to make the company look good.
Check more than whether you appeared
When you run each question, save the exact wording, the answer, the date, and any sources shown. Then judge the answer on several points.
Was your business identified correctly? Were its products or services described accurately? Did the answer address the buyer’s actual question? Did it give enough detail to help someone decide what to do next? Did it cite a useful, current source? If it named competitors, were they relevant to the same need?
An answer can fail in more than one way. Your company might be absent. It might appear but be described as doing work it does not do. It might be described accurately but without the one detail a buyer needs. These call for different fixes.
I would also mark “cannot tell” when the evidence is weak. If an assistant says something plausible without a source, do not count it as a verified fact just because it happens to be right. Separate what the system actually found from what it may have inferred.
Follow the trail back to the source
Once you find a weak answer, look for the reason. Is the needed information missing from your public pages? Is it present but hidden in an image or a file that is hard to navigate? Does an older page disagree with the current one? Is another business with a similar name getting mixed into the answer?
Sometimes the information is clear and the assistant still misses it. That matters too. One test cannot prove exactly why an AI system selected a source or produced an answer. Record the outcome, try a second wording or another service, and avoid claiming a cause you have not shown.
Google says its AI search features draw on the same core search practices used for regular Search. Its guidance calls for useful text, accessible pages, working internal links, and structured data that matches visible content. Google also says that meeting its requirements does not guarantee that a page will be indexed or shown. OpenAI describes how website owners can allow its search crawler to access public pages for ChatGPT search. These are practical checks, not a recipe for guaranteed placement. [1][2]
Fix what helps a buyer decide
Once you see the gaps, it is tempting to publish a page for every possible question. That can create a pile of thin pages that repeat the same claims. A better approach is to improve the pages where a buyer should naturally find the answer.
If buyers ask which businesses you serve, explain that on your service page. If they need specifications, put accurate details on the product page. If a prospective client needs to understand your process, describe the steps in plain language. Say when an offering is a good fit and, where helpful, when it is not.
Use the terms customers actually use. A buyer often describes a problem before learning your industry’s preferred label for it. Your page can use both, with a clear explanation of how they relate.
Some facts should stay private. You do not need to expose customer records, confidential methods, or a price you can only quote after review. A public page can still say what information a buyer should provide and how your team will respond.
Make the test repeatable
AI answers can change. The assistant may use a different source, a search index may update, or a revised question may lead somewhere else. Run a small set of the same questions again after you improve your pages. Keep the original wording so you can compare like with like.
Do not reduce everything to one score. A total can be useful for tracking, but it can conceal what matters most. Missing from a minor question and being wrongly described on a key product question are not equal problems. Show the examples alongside any count.
You can also look at the tools the search platforms provide. Google Search Console shows search performance, while Bing Webmaster Tools has an AI Performance report that can show which pages are cited in some Microsoft AI experiences. Neither gives a full picture of every AI answer or every buyer. [1][3]
The point is to learn, improve, and test again. It is much easier to act on “buyers cannot find our service area” than on “our AI visibility score is low.”
A question every manager can own
AI discoverability may sound like a marketing topic. I think it belongs on a manager’s list because it crosses the whole business. Sales knows the objections. Operations knows what the company can deliver. Product teams know the details. Customer service hears what people misunderstood. Marketing brings those facts together for the public.
If those teams do not agree on the answers, an AI system has little chance of giving a dependable one.
Start small. Choose ten real questions. Ask them with and without your name. Save the answers and sources. Find the three gaps most likely to affect a buyer’s decision, correct the information you control, and run the same test again.
The most useful question may be the one AI cannot answer yet. It tells you what your next customer may be looking for—and what your business needs to explain.
Sources
[1] Google Search Central, AI features and your website.
[2] OpenAI Help Center, Publishers and Developers FAQ.
[3] Bing Webmaster Tools, AI Performance.