Why ChatGPT recommends your competitor instead of you
It usually isn't because their website is better than yours. It's because the model read about them somewhere it didn't read about you, and that somewhere is often not their website at all.
Here is the uncomfortable version, first, because it's the part that actually helps: your website is often not the source. When an answer engine names three dentists in Bangalore or four boutique hotels in Jaipur, the sentences it produces are frequently assembled from directories, roundup articles, review pages and forum threads: pages about those businesses rather than pages by them.
So the question "why does it recommend them and not me" is really two questions, and they have different answers.
Question one: can the model read your site at all?
This one is mechanical, and it's the one most often broken. Answer engines reach your site through crawlers: GPTBot and OAI-SearchBot for OpenAI, ClaudeBot for Claude, PerplexityBot for Perplexity. According to Vercel's AI crawler study, none of the major AI crawlers execute JavaScript. They fetch your HTML and move on. GPTBot downloaded JavaScript files in roughly 11.5% of requests and ran none of them; ClaudeBot downloaded them in about 23.84% of requests and ran none of them either.
Googlebot is different. It has a rendering service and will execute your JavaScript, eventually. AI crawlers won't. This isn't a gap anyone's racing to close; it's a cost decision. Rendering JavaScript at crawl scale is expensive, and skipping it keeps retrieval fast.
The consequence is blunt. If your treatments, your rooms, your practice areas or your menu only appear after a script runs, an AI crawler sees an empty page. Not a thin page. An empty one. I've measured sites where the entire visible body was under 400 bytes and contained zero words of text, while a human visitor saw a complete, working, rather nice website.
Question two: does anyone else say you exist?
Assume your page is perfectly readable. You can still lose, and this is the part the AEO industry tends to skip past, because there's nothing to sell you.
Models weight corroboration. A business mentioned on ten independent pages is easier to name confidently than one mentioned only on its own domain, where every claim is self-issued. Your competitor may be winning not because they out-marketed you but because they're on three directories, in one local listicle, and quoted in a trade piece from 2024, and you're on none.
You can't fix that with schema markup. You fix it by actually existing in the places that get cited, which is slower and less satisfying than adding a script tag.
How to check, honestly
- Ask the question a customer would ask: not "is my business good" but "best implant dentist in Indiranagar", in ChatGPT with search on, and in Perplexity. Do it in a fresh session so your history doesn't colour the answer.
- Read the citations, not the prose. The cited URLs are the evidence. They tell you which sources the answer was actually built from, and whether any of them are yours.
- Check whether your page is machine-readable: that's the one part you fully control, and it's what the free Report Card measures: whether there's structured data, whether the content is in the HTML, whether the title and description survive.
- Repeat it over a few days. These systems are not deterministic. One answer is an anecdote. A pattern across a week is a finding.
What you can actually change
In rough order of how much they matter:
- Put your facts in the HTML. If a crawler that never runs JavaScript can't read your services and your location, nothing else on this list matters.
- Say who and where you are, in machine-readable form. Schema.org structured data is a labelled statement of your entity's facts. It doesn't force anyone to recommend you; it removes the excuse of ambiguity.
- Don't block the crawlers. Plenty of sites disallow GPTBot in robots.txt without ever deciding to. That's a real choice with real consequences, so make it deliberately.
- Be mentioned somewhere other than your own site. Slow, unglamorous, and the highest-leverage item here.
The part I won't sell you
Nobody can guarantee that ChatGPT will recommend your business. There is no ranking dial. There is no submission form. The mechanism is retrieval plus training data plus a probabilistic model, and it changes without notice.
What's honest to claim: you can stop being unreadable, you can stop being uncorroborated, and both measurably raise your odds. Anyone promising more than that is selling you certainty they don't have. In a field this new, confident precision is the tell.
Frequently asked questions
Does ChatGPT recommend my competitor instead of me?
If your competitor is named and you aren't, it is usually because the model retrieved or remembered pages that mention them and didn't find comparable pages about you. That includes third-party pages (directories, listicles, press, forum threads), not just their own website. It is a sourcing gap, not a verdict on your business.
Can I make ChatGPT recommend my business?
No one can guarantee that, and you should be suspicious of anyone who says they can. There is no ranking dial to turn. What you can do is make your business easy to find, read, and corroborate, which raises the odds you're the one retrieved.
Why does ChatGPT name businesses that are worse than mine?
Answer engines optimise for retrievable, corroborated information, not for quality. A mediocre practice with a clear machine-readable site and a dozen third-party mentions is easier to name than an excellent one with a JavaScript-rendered site and no external footprint.
How do I check what ChatGPT says about my business?
Ask it the question a customer would ask, with web search enabled, and read the citations rather than the prose. The cited URLs tell you which sources the answer was actually built from. Then check whether your own page is even readable, which is what the free Report Card measures.