By John Wheeler, Founder of Wheeler Intelligent Systems LLC
Imagine a restaurant owner looking for a new supplier. She needs a frozen breakfast product that is easy to prepare, comes in a useful case size, and fits her kitchen’s needs. She also wants to know how to order it in her area.
She could search the web, open several sites, download product sheets, and call a sales rep. She could also ask an AI tool to help her build a short list.
Now put yourself on the supplier’s side of that search.
Your company may make a product that fits her needs. Your sales team may know exactly why it is a good fit. But can someone who has never heard of your business find that information? Can an AI search tool find enough clear evidence to include you in a useful answer?
That is the business problem I call AI Discovery.
At Wheeler Intelligent Systems, our focus is to study this problem, test what AI can find, and help businesses make their public information more useful. The goal is practical: help buyers find a business, understand what it offers, and take the next step with fewer gaps.
What AI Discovery means
AI Discovery is the process through which people find and learn about businesses, products, services, and experts with help from AI.
For a business, it raises three questions. Can the AI find you? Can it explain you correctly? Can it connect what you offer to what a buyer needs?
Those questions are related, but they are not the same.
An AI tool might know your company name yet have little useful information about your products. It might describe your products but get your service area wrong. It might link to your website without finding the details a buyer needs to decide whether to contact you.
That means a mention alone is not enough. A useful discovery process should lead to an accurate picture of the business.
Consider a fictional company called Cedar Valley Foods. An answer that says “Cedar Valley Foods sells food products” offers little help. A buyer wants to know what kinds of products it sells, who they are made for, how they are packed, and where current ordering details can be found.
Those details are where our work begins.
The technical term is Generative Engine Optimization
You may also hear this work described as Generative Engine Optimization, or GEO.
A research paper called GEO: Generative Engine Optimization, accepted at KDD 2024, introduced a framework for improving and measuring content visibility in answers from generative engines.
In plain language, GEO concerns how content shows up in AI-generated answers.
I use AI Discovery when talking about the business need. I use GEO when discussing the technical work and research around visibility in those answers.
The two terms help us speak to different readers. A business owner needs to understand the problem before learning a new acronym. Someone planning a website or content strategy may also need to know the term used in research and industry discussions.
Neither term should be treated as a promise that a business can control an AI answer.
What happens to traditional SEO
Search engine optimization, or SEO, still matters.
Google’s guidance on AI features and websites says that established SEO practices remain relevant to AI Overviews and AI Mode. It also says there are no extra requirements or special optimizations needed to appear in those features.
That is a useful check on the excitement around GEO. A new term does not mean a business should discard the basics.
My practical recommendation is to begin with pages that work, explain the business clearly, and help real buyers. Then test how well selected AI tools can use that public information.
Google’s guidance applies to Google’s features. We should not assume that every AI service finds sources, chooses links, or builds answers in exactly the same way.
For that reason, an assessment should name the service tested and describe how the test was run. “We tested this question in this tool on this date” is more useful than a broad claim that “AI knows your business.”
Start with the questions your buyers ask
The first step is to write realistic buyer questions.
Many businesses begin by asking an AI tool, “What do you know about our company?” That is a useful identity check. But it assumes the buyer already knows the company name.
Discovery often begins earlier.
For our fictional food supplier, a buyer might ask: “Which companies offer frozen breakfast products for hotel kitchens?” Another might ask: “What should I compare when choosing a packaged bakery item for a school food program?”
These are examples of questions to test, not claims about how often buyers use them.
They help us examine whether a company’s public information connects its products to real needs. A product page may name an item but never explain its intended use. A service page may describe a broad region without explaining how someone can check delivery or ordering options.
Sales teams hear these questions every day. Their knowledge can help shape a much better assessment than a random set of prompts.
The goal is to test the questions that matter to a buying decision.
Product information can become a discovery problem
In foodservice, a product name and a good photograph may leave many questions unanswered.
A buyer might need to know the case pack, unit size, storage needs, preparation steps, ingredients, or allergens. They may also need current information about where and how to order.
Imagine that Cedar Valley Foods posts a beautiful product image. The case pack is buried in an old PDF. Preparation steps are missing. Two public pages list different weights.
A human buyer may struggle with that information. In an AI Discovery assessment, we would test whether the same gaps appear in answers and record which sources were used.
This is also where product information management, or PIM, connects to AI Discovery.
My working view is that clear, accurate product records give a business a stronger foundation for useful public pages. That is a reason to improve the records and test the result, not proof that a particular PIM system will cause an AI recommendation.
Any claim about ingredients, allergens, certifications, or availability should come from verified business records. AI should not fill those gaps by guessing.
Test discovery and answer quality separately
A useful assessment should distinguish between being found and being understood.
One question is whether the company appears when the buyer has not named it. Another is whether an answer about the company gets the facts right. A third is whether the answer supplies sources that support those facts.
These checks reveal different problems.
If a business never appears, we should examine the questions, competing sources, and available public information. If it appears with wrong details, we should look for conflicting or outdated records. If it appears accurately but the next step is unclear, the business may need a better path to a product sheet, ordering page, or contact.
I would also record unanswered questions. A clear admission that information could not be found is different from a confident but false answer.
A useful score should preserve those differences rather than hide them inside one large number.
Keep a record before making changes
A baseline is a record of the starting point.
For each test, save the exact question, the date, the service used, whether web search was enabled, the answer, and the cited links. Note whether the question named the business or asked more broadly about a need.
Repeat key questions under comparable conditions. If we change the wording, switch tools, and update the website at the same time, it becomes harder to explain a different result.
Even a careful before-and-after test has limits. It can show what happened in the recorded runs. It does not, by itself, prove that a website edit caused the change.
That is why I favor language such as “the company appeared in more of our repeated tests” over “we solved AI visibility.”
We can make useful progress while staying honest about what the evidence shows.
Improve the information buyers need most
My recommended first changes are often simple.
State clearly what the business does and who it serves. Give important products their own useful pages. Correct old information. Explain how a buyer can confirm availability. Make it easy to reach the right person.
Where several public records describe the same business or product, check that they agree.
Original knowledge can help make those pages worth reading. Explain a real product use, answer a common buying question, or describe a tradeoff that customers often miss.
For example, a manufacturer could explain how to compare portion size and preparation needs for a breakfast program. A distributor could explain how a new customer checks whether a product is available in its market.
These are proposed content approaches. We should measure whether they improve the tested answers instead of assuming that more content will produce better discovery.
A practical first test for your business
You can begin without buying a large software package.
Write a small set of questions that a new customer might ask. Include questions about your business by name and questions about needs you can meet.
Run them in an AI tool with web search available. Save the answers and open the cited sources. Compare what you see with facts you can verify.
Look for one useful gap to fix. Perhaps your service area is unclear. Perhaps a product sheet is out of date. Perhaps your website never answers a question your sales team hears each week.
Correct that gap, allow time for public changes to become available, and repeat the recorded test.
This will not give you a complete view of the market. It will give you a concrete starting point.
What Wheeler Intelligent Systems is building
At Wheeler Intelligent Systems, we are developing a research-backed approach to AI Discovery and GEO, with an initial focus on foodservice manufacturers and distributors.
Our aim is to connect buyer questions, public information, recorded tests, and practical improvements.
We want to help businesses see where they are missing, where their information is unclear, and where an answer gets the facts wrong. We also want to show the limits of each test so that a promising result does not become an unsupported claim.
The business goal reaches beyond appearing in an answer. It is to help the right buyer understand an offer and reach a useful next step.
If you want to start a conversation about your business’s AI Discovery, visit WheelerIS.com.
Bring one question your customers ask often. That question can tell us more about where to begin than a page full of general marketing claims.