AI Optimization: Is Your Website Ready for GEO?

We are approaching the era of Internet users relying primarily on AI for all information. Despite their increase in popularity, generative engines are young and imperfect. We saw this firsthand - tour guides are mistaken for sandwich shops, recommended beauty salons' websites are on domains ready to be claimed, and event planners can be supposedly found in the middle of the Pacific Ocean. Despite such chaos, there must be some rules for AI to follow. But what are those exactly? This and many other questions about GEO (Generative Engine Optimization)are riddles the SEO experts are still attempting to solve.

To find out what makes AI tick, we ran our own analysis by prompting ChatGPT in ways a real customer might. The results were sometimes accurate, often surprising, and occasionally bizarre. And with each of them we can learn how business websites can increase their visibility by editing their content for a better AI optimization.

How we tested for AI optimization

To evaluate how well real-world businesses are found and presented, we used three types of prompts in our ChatGPT optimization research:

Prompt 1️⃣: “Best [industry] company in [city, state]"

This tested how well AI handles general user queries and whether it would recommend the companies we selected. We also assessed if suggested businesses matched the correct industry and location, and verified the accuracy of any included contact info.

Prompt 2️⃣: “Who offers [service] near [city, state]?"

Here, we focused on niche or less-visible services. For each company we picked one such service from their website - usually not seen on the homepage. If no list of services was available, we checked team bios for specialties. We then searched for that service in the same area, evaluating both relevance and accuracy of AI results, as well as visibility of the researched companies.

Prompt 3️⃣: “What do you know about [business name]?"

With this prompt we checked if AI recognizes specific businesses by name alone. When a business had multiple name variations, we picked the most prominent one across the website – logo, website header, meta title, domain, text descriptions etc. Then it was a matter of reviewing the response for accuracy: background, services, reviews, contact info, and more. If ChatGPT showed multiple definitions, we also checked if they were industry-related and looked for possible hallucinations.

For this GEO research a 100 small business websites from across the U.S. were chosen. These companies came from all 50 states, represented a variety of city sizes, and used different CMS for their websites. We grouped them into four primary categories:

• Home & Field Services – appliance repair, handyman services, roadside assistance, lawn care & pest control
• Wellness & Personal Care – hair salon, spa, yoga classes, beauty salon & tattoo studio
• Health & Medical – dental office, medical practice, mental health, speech therapy & midwife
• Entertainment / Events – photography, event planning, DJ, tour guide & wait staff

After running all tests, we analyzed the patterns in LLM’s behavior. Were some industries more likely to yield detailed results? Can a business in a town as small as Whitefish, MT appear as easily as one in a metropolis? Here’s what we found across all four categories - all possibly vital for your SEO and AI optimization strategies.

What to watch out for when optimizing for GEO?

Here are frequently observed issues in AI-based search, organized by industry

🛻 Home & field services: distinctiveness issues

Businesses offering at-home or on-site services often include information about their service area on their websites. ChatGPT seems to take this into account. In about 8% of responses, it listed companies that were not based in the target location but explicitly stated on their websites that they serve it. These businesses were located up to 50 miles away.

Brand name confusion and hallucination risks

Businesses in this category had the highest chance of being confused with others. In 28% of prompt #3 responses, AI matched the business name with unrelated or similarly named companies from the same industry.

Take "Quiet Lawn" — ChatGPT not only listed a different company, but also interpreted the name as a trend of using electric lawn tools. That’s a stretch, but not as wild as with "Upright Brothers", where AI invented a fictional music duo by blending unrelated bands with “Brothers" in their names.

Why does this matter? When AI gets sidetracked by irrelevant or made-up content, it often skips key info like services, pricing, or contact details. In fact, 52% of responses in this category lacked any contact information. This shows how critical a name is in AI optimization — it should be unique enough to avoid confusion, but also clear enough to signal the right industry.

Smart AI optimization means keeping locations connected

It also should be ensured that the company doesn’t compete with… itself! We saw one case, where two different facilities belonging to the same company were listed as two separate positions, not connected to each other. This was most likely due to both of those offices having their websites on different domains. Other multi-location companies avoided this problem by using subpages for each location under a single domain.

Link accuracy issues

When prompting ChatGPT, we received two repeating types of outputs: a listing and a Mapbox embed (shown below). We consider Mapbox the better outcome — even for businesses that offer mobile services (where the physical address shouldn’t really matter). Why? Because it typically includes more information and offers a higher chance of getting the correct link.

With the listing format, the links always lead to Google Maps. Sometimes such links go straight to the business’s location, but other times they lead to a broader search results page — and the business we just clicked might not even appear. That means more clicking and lower chances of finding the right information.

Types of business website hyperlinks included in responses to prompts #1 and #2

Sparse contact info and crowded responses

Because ChatGPT tries to avoid clutter, it won't often add contact info when it thinks enough details have been provided. The frequent appearance of Google Maps links meant contact data was especially scarce – 42% of location-based replies had no address, e-mail, or phone number for any of the listed companies. For prompt #2, addresses showed up in only 12% of responses. This alone makes sense to some degree – the user will not necessarily need to know where the companies from this category are located. After all, these businesses typically come to them, not the other way around. It shows that following human intuition also plays a part in AI optimization.

🪷 Wellness & personal care: general query issues

When it came to the “Best company" prompt, wellness & personal care businesses performed the worst. Out of all AI responses, 56% of them were questionable - 44% were only marginally acceptable, and 12% were outright not acceptable. The main issues were:

• Incorrect phone numbers
• Incorrect website links
• The suggested businesses are no longer active

Website links, in particular, were all over the place. There were the usual 404 or nxdomain errors for sure, but also some cases of websites with domains ready to be claimed. Some links led to entirely different businesses, like an online shoe store. This makes wellness businesses especially vulnerable to outdated or misleading online data.

Accuracy of AI responses to prompts #1, #2 and #3

Niche prompts lead to much better results

That said, prompt #2 — “Who offers [specific service] near [city]?" — produced far better results. Just 16% of responses were marginally accurate, and only 4% were inaccurate. This is the best accuracy score across all categories for this prompt. Researched companies appeared more often in this category as well, albeit just slightly (56%). Across both location-based prompts it was also the least likely for the AI to omit contact information – this happened in only 16% of replies. Optimization for niches appears to be a clear case of effective AI optimization for companies in this category.

Tattoo studios stand out as a unique case

Tattoo studios stood out from the other industries. These were the only businesses for which ChatGPT included Instagram handles, belonging to the whole studios or individual artists. This makes sense - tattooists often rely on Instagram portfolios to showcase their creative work. Interestingly, ChatGPT was more likely to recommend individual tattoo artists than studios. Still, if an artist was tied to a specific business, that company was usually named too.

Limited website content works against AI optimization

When asking about details, the responses were generally decent, but with some oddities observed. Especially if websites meant primarily for booking and with minimal content are concerned. One salon, for example, had a simple booking page hosted on a square.site domain and a Facebook page. ChatGPT called it a “Facebook salon" and gave just two lines of info, since other similarly named companies were also mentioned. There were also two company websites with glossgenius.com domains. While the AI gave longer descriptions for those, its information on opening times didn’t match what was actually mentioned in their About pages. The possible culprit? Third-party domains may weaken a site's authority in the AI's eyes, leading to less reliable or less detailed responses.

🩺 Health & medical: competition-related issues

This most likely won’t come off as a surprise, but health and medical businesses operate in the most competitive space. Demand is constant, and it shows in the numbers: only 12% of the companies we researched were recommended as "best in the region", and 28% showed up when asking for specific services. While niche targeting is valuable, visibility in AI-generated results requires much stronger ChatGPT optimization due to the saturation of similar business names and services.

Visibility of businesses picked by us in responses to prompts #1 and #2

Website links are rare and often confusing

Prompt #2 responses rarely included any kind of website link. Mapbox appeared in only 12% of replies, and Google Maps links only once. However, in this category we noted the only cases of website link inclusions in the form of a bullet point in company descriptions. They weren't always intuitive though, as sometimes they appeared as source buttons placed right next to the actual sources. All told, 72% of responses had no link at all, making this category the least likely to help users click through to a business's website.

The largest risk of company omissions

Health & medical companies were also the most likely to be ignored entirely in prompt #3. In 12% of cases, the business we asked about wasn't acknowledged in any way – the highest omission rate across all four categories. Examples include:

• “Smile Solutions" – name for a dental office; seven others were discovered, including ones in Australia and UK
• “The Cypress Clinic" – name for a medical practice business; overshadowed by another medical center, two mental health centers and a retirement community
• “Mountain Midwife" – name for a midwife business; ChatGPT found three possible definitions, all related to midwife traditions or stories from Appalachian Mountains (not the business's region)

Naming conventions can make or break recognition

There is one case that highlighted AI's sensitivity to naming – “Speech-Language and Swallowing Specialist, LLC". The inclusion of “LLC" has proven to be very significant. Without it, it would be possible even for a human to think of it as another way to describe speech-language pathologists as a general profession. And AI agrees, since that’s exactly what it described! However, once we added the “LLC" part, it correctly identified the company. This example demonstrates how crucial AI optimization can be when dealing with complex or descriptive brand names.

The bright side: AI readily shares contact details

There is an upside though - health and medical businesses are the most likely to have contact info included in AI responses. We found that:

• 72% of companies had at least one contact detail
• 36% included both address and phone number
• 8% also included e-mail

Once again, this matches human intuition. Access to healthcare providers is considered a priority, so AI appears more likely to surface contact info where possible.

Combinations of contact information present in AI responses to prompts #1, #2 and #3

🎉 Entertainment businesses: mapping issues

The entertainment and event category produced some of the most interesting results, especially when it came to location data. In 24% of prompts, Mapbox markers pointed to the wrong place entirely. Some even landed in Lake Michigan or the Pacific Ocean.

We looked for an event planner in Philadelphia and got a suggestion for someone in the ocean, near West Coast

This is how that company's location appears in Google

Why does this happen? Likely because AI seems to rely on the company locations shown in Google Maps. The businesses from this category often don't display a specific location on Maps, and instead highlight a large service area (which may cover an entire state, or even the whole U.S.). But when AI tries to map them, it simply pins the center of that service area - sometimes hundreds of miles off. Therefore, listing a broad region as your "location" may negatively contribute to AI optimization and reduce your credibility in AI-generated responses.

Name confusion leads to mismatched listings

There were also two cases of mixing text descriptions of companies with a Mapbox position that appears completely unrelated, but partially shares a name:

• Instead of “Aria Melody DJ", who provide wedding and event DJ services, the Mapbox list has shown a wine bar named “Aria". Image, directions, website and phone number were all related to that company instead.
• Instead of “Taste Virginia Tours", providing food related tour guides, we got Mapbox showing us all information for “TASTE", which happens to be a local sandwich shop chain.

Inaccurate or irrelevant business suggestions are common

The issues mentioned in previous points contribute to only 52% of responses to prompts #1 and #2 getting an “acceptable" score from us. Other common issues included:

• Businesses being located far from the requested area
• Outdated or unverifiable contact info
• Full-on hallucinations, like recommendations for a nonexistent freelancer forum

In 20% of prompt #2 cases, AI suggested businesses that didn't even offer the specific service, or anything close to it. It shows that it still struggles with highly specific requests in this category and highlights the need for effective ChatGPT optimization.

AI optimization around niche has limited impact

Unlike other industries, entertainment and event companies saw little benefit from niche-focused prompts. Visibility didn't improve much between general and niche queries. That said, some smaller, specialized providers like DJs and wait stuff in New York still managed to surface in the results. It shows that such industries themselves might already be seen as a niche. Although keep in mind that overall visibility scores are on par with Home & Field Services and Wellness & Personal Care.

AI's knowledge on company details is the strongest

On the other hand, prompt #3 is where we saw the most promising results. When asked about a business by name, only 12% of replies fell into the marginally acceptable range. More importantly, 100% of researched companies were acknowledged, with 84% of replies focusing solely on the correct company.

Recognition of businesses and mentions of other definitions for their names in replies to prompt #3

This suggests name-based prompts offer a strong opportunity for AI optimization in this space. Although, those with not as unique names, such as “Peak Entertainment" or “Stargazer Productions", were listed primarily alongside businesses from other entertainment industries. Contact details were also hit or miss. In 52% of cases, no contact info was shown, but when it was, this category led the pack for full sets (address, phone, and e-mail). Tour guides performed particularly well in that regard.

Final GEO insights & how to make your business AI-ready

It's no surprise: the more specific the user's query, the more chances for an accurate response from ChatGPT. But as it also turns out - the higher chance for your business to appear in its answers.Based on our analysis, we can determine what's important (or not so important) for AI and share the most important takeaways for improving your AI optimization:

1. Businesses that clearly identify a niche perform better. Include niche-related keywords in your services list, team bios, or page content to help the AI recognize your specialties.
2. AI uses more than just your website. It pulls data from multiple sources - Google, LinkedIn, BBB, MapQuest, Yelp, and even industry-specific sites like WeddingWire (check our post about AI sources for more insight on what sites AI uses to answer service business queries). Keep your business info consistent and up to date everywhere to avoid conflicting details and user confusion.
3. Prioritize one clear phone number. it’s the most likely contact info to appear, and the only one that can show up in Mapbox results. List a single, primary number for clients. If you display multiple numbers (e.g. separate numbers to call different employees), AI may ignore them altogether.
4. E-mail does not matter as much as you could imagine. Across prompts #1 and #2, we didn’t get a single business e-mail. Even in prompt #3, only 13% of responses included one.
5. Ensure that AI finds a proper link to your home page. On top of that, check if appending ?utm_source=chatgpt.com to URL does not create a broken link, as websites accessible via hyperlinks in ChatGPT (including Mapbox buttons) have that automatically included.
6. Keep your branding unique, but also clear. Avoid business names that could be misspelled or mistaken for anything else. Do not consider your industry only – check if the name wouldn’t be more associated with an entirely different field from yours.
7. Do not worry about AI not reading your PDF files. If your services or details are listed in such file, ChatGPT can still access and understand content within any PDFs hosted on your site.
8. Phrases “in this location" and “near this location" are treated the same. That means your Mesa-based business might show up in Phoenix searches. Use nearby cities strategically in your content to expand your reach.

The generative engines are still evolving and we may need to wait a few years before they become refined. However, those who take effort to understand how they operate today will have easier times grasping their functionality in the future. Those who will prepare for such an upcoming pivotal change right now may be the ones to rank high in LLM responses sooner than later. After all, strategic ChatGPT optimization and adapting your website to AI’s quirks will be the key to being found and trusted. And with GEO just getting started, you still have a lot of room for manoeuvre. Starting with creating your SEO and GEO optimized website through an AI-powered generator could be the first step to your success.

The data and statistics presented in this blog post come from a research study conducted by IKOL in 2025. To learn more about IKOL research methodology and explore other findings, visit: ikol.com/research.