Most advice about getting cited by ChatGPT is repeated on faith: write good content, get reviews, show up on Reddit. A recent independent study took a different approach. Instead of guessing from the outputs, researcher Suganthan Mohanadasan read the actual network traffic ChatGPT sends to the browser, the internal labels and metadata behind every answer, and published what he found in June 2026.
Two findings matter most for local businesses. First, being in the top 3-5 businesses an AI engine actually names is what determines whether you’re in a customer’s consideration set, a much smaller and more competitive window than the traditional top 10. Second, the model explicitly gives up on reading a business’s own pricing or facts when the page is built with JavaScript, and cites a third-party site like G2 or Reddit instead, even when the business’s own page had the correct information. This isn’t a niche concern: Cloudflare’s own crawler traffic data shows GPTBot’s request volume grew 305% in a single year, even as it became the most-blocked AI crawler in robots.txt files across the web, a sign that most site owners are actively managing access to these bots without fully understanding what each one controls.
Local Dominator's Site Audit is built to catch exactly this kind of problem: whether AI crawlers can actually read your site, and whether your key facts are sitting in plain text or hidden behind the rendering that just cost you a citation.
How Many Local Businesses Does ChatGPT Actually Show?
Traditional local SEO trains everyone to think in terms of the 3-pack or the top 10. The network traffic research references an internal field suggesting a hard limit of two results for a specific local UI component, so we tested this directly rather than taking that number at face value for the full answer a customer actually sees.
Across live queries we ran (a plumber search in Austin and a coffee shop search in Seattle), ChatGPT’s written answer consistently named five local businesses, each with a rating, review count, and a short description, not two. A separate, independent observational study by Sagapixel, based on 52 real users across 177 ChatGPT sessions searching for local healthcare services, found people considered an average of 3.7 businesses per search, and that appearing in the “top 3-5” results was enough to stay in serious consideration. Only 27% of users looked at just one option.
Put together, the honest picture is more nuanced than a single hard cutoff. A specific visual component, the map pin and card carousel that sometimes accompanies a local answer, does appear to render a small, limited set at a glance. But the written answer itself, the part a customer actually reads, regularly names four to five businesses with real detail. The practical takeaway lines up with the independent research either way: being in the top 3-5 is what determines whether you’re in a customer’s consideration set, not whether you’re first, and not being there at all is a much bigger problem than losing a percentage point of position within that group.
This is precisely the kind of gap AI visibility monitoring is built to close: not a single ranking number, but whether you’re consistently one of the small handful of businesses an AI engine actually names when someone asks.
Fetched, Cited, and Mentioned Are Three Different Things
The study draws a distinction most local business owners have never had a reason to think about. A page can be fetched (pulled into the model’s context), cited (attached as the source behind a specific claim), or mentioned (named in the answer without being the source), and a business can win one of these while losing the other two entirely.
In one batch of commercial queries, Reddit was fetched 278 times and cited 11 times. YouTube was fetched 201 times and cited zero times. The mechanical reason: a citation has to attach to actual text the model can read, and a YouTube search result hands the model metadata, not a transcript. Reddit threads are the page. A separate, larger analysis of 1.4 million ChatGPT prompts found the same pattern holds at scale, with Reddit earning meaningfully more citations than YouTube.
The practical version of this for a local business: showing up in an AI-generated answer at all is not the same as being the source that answer is built on. Those are two different things to track and two different problems to fix.
Why Your Pricing Page Might Be Handing Your Citation to a Competitor
The most direct, actionable finding involves what happens when ChatGPT tries to verify a price. On a comparison task, the model’s own saved reasoning, read directly from its saved trace rather than inferred, showed it going to a company’s official pricing page first, exactly the behavior you’d want. For one company, it found the numbers directly and cited the official page. For another in the same comparison, the reasoning was explicit: the pricing “isn’t showing up directly in the search result, possibly because it’s loaded with JavaScript,” so the model gave up and cited a third-party source instead.
To see how common this actually is, we ran our own test: checking whether pricing numbers appear in the raw, server-delivered HTML, the same thing a non-JavaScript crawler reads, before any script runs. We checked Local Dominator’s own pricing page along with two established SaaS products (Notion and Linear) known for strong technical SEO. All three passed: pricing was present in the raw HTML in every case, meaning a text-based crawler would find it without needing to execute anything. That’s reassuring, and it suggests the failure mode the original research documented is real but not universal, it shows up on sites that render critical facts entirely client-side rather than serving them in the initial HTML response, not on every site.
When that happens, the model doesn’t leave the answer blank. It falls back to a third-party source, in this case G2, and cites that instead. The business’s own numbers were sitting right there on their own page. The model just couldn’t read them, because the price was rendered by JavaScript rather than sitting in the page’s plain HTML.
This isn’t a ChatGPT-specific quirk. It’s a direct window into how these models actually read a page: the same way a text search would, checking the rendered HTML for the facts it needs. A price, a phone number, or a set of hours that only appears after a script runs is invisible to that process. A price sitting in plain text is not.
For a local business, this applies to more than pricing. Hours of operation, service areas, phone numbers, and address details that load dynamically rather than sitting in the initial HTML carry the same risk. If an AI engine can’t read it, it either skips the claim entirely or cites whichever directory or review site happens to have the same information in a format it can parse.
What This Means for Local Businesses Specifically
Three practical implications follow directly from how the traffic behaves:
Put your core facts in plain HTML, not behind JavaScript. Business hours, pricing, service areas, and contact details should be readable in the page's raw HTML without requiring a script to run first. This is precisely what an AEO-focused technical audit checks for: whether AI crawlers can actually access and parse the content that matters most.
Third-party coverage is doing more work than most businesses realize. Since citations often go to review sites and community threads rather than a business's own page, especially when that page is technically hard to parse, getting more reviews and building citation consistency across 70+ directories isn't just a ranking signal anymore. It's frequently the actual source an AI engine cites when it talks about a business.
Being one of two matters more than being on a list. With local results capped at two, the businesses that consistently show up are the ones with the clearest, most consistently verifiable information across the sources AI engines can actually read, not necessarily the ones with the most content.
How to Check Whether This Is Happening to Your Business
You don’t need to read raw network traffic to get a useful signal. A few practical checks:
Ask the AI engines directly. Query ChatGPT, Perplexity, and Gemini with the exact questions a customer would ask, including pricing and hours questions. Note whether your business is named, and if so, whether the answer sounds like it came from your site or from a review platform.
Check whether your key facts render without JavaScript. Right-click your pricing page or hours section and select "View Page Source," or use your browser's "Inspect" tool and search the raw HTML for the number you expect to see. If it's not there, it's likely being rendered by JavaScript after the fact, the same gap that cost a real company its citation in the study referenced above, and the same test we ran on Local Dominator's own pricing page before publishing this article.
Run a technical audit that checks AI crawler access specifically. Local Dominator's Site Audit scores whether GPTBot, ClaudeBot, and similar AI crawlers can access a site, alongside whether critical content is structured in a way these models can actually parse, distinct from a traditional SEO crawl that only checks Google's crawlers.
Frequently Asked Questions
Why won't ChatGPT show my business when I search for my own services?
A few explanations are common: the query may be answered from ChatGPT’s training data without triggering a live search at all, your business may not be one of the two results ChatGPT surfaces for local queries, or your site’s key information may be rendered in a way the model can’t read. Checking your page’s raw HTML for critical facts like pricing and hours is a fast way to rule out the third possibility.
Does ChatGPT actually crawl websites like Google does?
Not in the same way. The research found ChatGPT fetches most open-web pages through third-party scraping services rather than its own crawler, and a small allowlist of major publishers get pulled through a separate, more thorough pipeline. Most businesses are only reachable through the scraped tier, which makes clean, plain-HTML content more important, not less.
If I block GPTBot, does that stop my site from appearing in ChatGPT answers?
No, and this is a common mix-up. According to OpenAI’s own documentation, GPTBot is used only to collect training data for its models; blocking it opts a site out of that training but has no effect on whether it appears in ChatGPT’s live search answers. The crawler that actually controls search visibility is a separate one called OAI-SearchBot. A site that disallows GPTBot for training-data privacy reasons while leaving OAI-SearchBot allowed can still be fully visible in ChatGPT search results, and vice versa.
How many local businesses does ChatGPT show for a "near me" search?
Live testing found ChatGPT’s written answers typically name four to five local businesses with ratings and descriptions, not just one or two. An independent study of 52 real users found people considered an average of 3.7 businesses per search and that appearing in the top 3-5 results was enough to remain a serious option. A specific visual widget that sometimes accompanies these answers may display fewer at a glance, but the text answer itself is where most of the real estate is.
Is this the same as SEO, or something different?
It overlaps but isn’t identical. Traditional SEO focuses on ranking in a list of links. This is about whether an AI model can read your page well enough to cite it as the source behind a specific claim, which depends more on clean, parseable HTML and third-party verification than on backlinks alone.
What's the single most useful fix for showing up in ChatGPT?
Make sure pricing, hours, service areas, and other core facts appear in your page’s plain HTML rather than being loaded in by JavaScript after the page renders. A real company lost a citation to a third-party review site for exactly this reason, and Local Dominator’s own pricing page passed this check when we tested it. It’s one of the most common technical issues an AI-focused site audit surfaces.
AI engines aren’t reading websites the way search engines do, and the businesses getting cited aren’t necessarily the ones with the best content. Often, they’re simply the ones whose facts happen to be in a format the model can actually parse. That’s a fixable, testable problem, not a mystery.
