Generative engine optimization · Original data
ChatGPT cited the business's own site in 2.5% of answers. Perplexity did it in 20%.
Every time someone runs our free check, we feed an AI engine a few customer-style questions about that person's industry and city, then log which sources the engine points to and whether the business's own website makes the cut. From June 23 through September 18, that gave us 160 completed answers across two engines. We pulled the full dataset this week to see how differently they operate, and the quick takeaway is that both engines always cite someone, but they leave very different amounts of space for you.
GPT-4o linked to the business's own site in 1 of 40 answers. Perplexity's Sonar did it in 23 of 115. That's 2.5% versus 20%, and the gap traces back to one mechanical quirk in how each engine assembles an answer.
What the data actually contains
This needs unpacking because it's not a controlled experiment. These rows are live checks that real people ran on our site over three months. The Perplexity answers all come from late June to early July, back when Sonar powered the check. The GPT-4o answers run late August to mid September, after we switched over. Different weeks, different businesses and different questions, in whatever mix visitors happened to bring. We're comparing two heaps of actual usage, and the numbers need reading with that in mind.
The check fired off 200 questions total. Forty never got answered because our upstream account hit its credit limit those days, so we excluded them rather than counting them as silence from the engine. Five more came from a preview version of GPT-4o, which is too few to mean anything, so we dropped those too. What remains is 40 GPT-4o answers covering 25 distinct questions, and 115 Perplexity answers covering 55 distinct questions.
One more caveat needs stating. Our tracker logs cited sources, meaning the actual web pages the engine links to, rather than business names that appear in the prose. A business can be named in the text without its site getting the citation, and our earlier Miami med spa post digs into exactly that split. Here we're only looking at which sites earn the link.
ChatGPT lists five sources. Perplexity lists fifteen.
GPT-4o cited between 1 and 10 sources per answer, averaging 5 with a median of 4.5. Perplexity returned exactly 15 citations in every single one of its 115 answers, which is how Sonar's API is built rather than something we configured.
That list length difference is the entire story. When an engine has five slots and fills them with "best med spas in Miami" listicles and directory pages, there's simply no room left for an individual business's site. When an engine has fifteen slots, the listicles and directories still come first, but after them there's space for the actual businesses being recommended.
You see it in who got cited. Across GPT-4o's 40 answers, the repeat winners were medsparankings.com (11 answers), medicalspaproviders.com (10), timeout.com (9), discovermedspa.com (9), and nobsmedspareviews.com (9), and every single one of those is a ranking, directory, or magazine page. Across Perplexity's 115 answers, the top two were also directories, medicalspalocator.com (26) and atly.com (19), but right behind them came individual businesses: birdiedayspa.com (18), laser-clinique.com (18), dermacaresandiego.com (13), and bellesirenesalon.com (12). Those are single-location spas and salons whose own homepages kept showing up.
GPT-4o cited 100 distinct domains across its 40 answers. Perplexity cited 380 across its 115. The longer list doesn't just mean more of the same directories. It means the engine digs deeper, all the way down to the businesses themselves.
Two engines, two different jobs for you
If a customer asks ChatGPT, your own site almost never makes the five sources it relies on, so your work is to be inside the pages it does rely on. In our data, that means the category ranking sites and city magazine roundups that keep appearing answer after answer. If medsparankings.com or your local Time Out list gets cited in a quarter of answers for your category and you're not on that page, you're missing from ChatGPT's answer for a reason that has nothing to do with your website.
If a customer asks Perplexity, your site has a real shot at direct citation, one in five in our numbers. That makes your own pages worth building in a way they simply aren't for ChatGPT. A page that clearly states what you do, where you are, and what you charge is something Perplexity can pull straight into its fifteen slots. It also means checking Perplexity flatters you a bit. Seeing your site in a Perplexity answer says nothing about whether ChatGPT knows you exist.
This matches the 356,019-citation study we covered separately, where Perplexity sent 72.3% of citations to the business's own site and ChatGPT sent 48.8%. Our table is way smaller and way messier, but it points the same direction, and it adds the mechanism, which is that a short list leaves no room for you and a long list leaves some.
What we couldn't measure
Google's AI Overviews aren't in this round. Our free check has never run against them, so there's no row to compare, and we'd rather omit them than speculate. Claude and Gemini are out for the same reason. The two engines here are the two we have actual usage data on.
The other honest limit is that the average of five sources on ChatGPT is an average across many different questions from many different businesses. Your category and your city have their own list, and the only way to know which listicles and directories dominate it is to ask. That's what the free check on our homepage does. Drop in your business and city, and it'll show you which sources each engine cites for the questions your customers ask, and whether you're among them.