Answer engine optimization · Research
Which AI is more likely to cite your own site?
Run the same home-services query through Gemini and ChatGPT and you'll often see the same businesses surface. What shifts is who the AI gives the credit to. A study that wrapped data collection on September 2 quantified that gap, and it's wider than most owners expect.
The number
Cheers tested 32 home-services prompts across 100 US and Canadian metro areas, running them on four engines for 28 days. They sorted 356,019 citations by domain. Here's how the engines broke down: Gemini directed 80.7% of citations to the contractor's own site. Perplexity hit 72.3%. ChatGPT landed at 48.8%, with 44.7% routing to directories and review platforms instead.
Google's AI Overviews sat in the middle, with 70.3% going to the business's own site plus 11.4% landing back on Google's own properties. That last piece fits with something we flagged a few days back: on Google specifically, Google takes its slice of the citation pie before anyone else gets a look.
What the gap actually means
If your customers lean on Gemini or Perplexity, your website's already carrying most of the weight. Eight or nine times out of ten, when the AI names you, it's pointing somewhere you own. Polish that page and the results show up straight away.
If your customers ask ChatGPT, that math falls apart. Nearly half the time you're mentioned, the "source" underneath isn't your site at all, it's Yelp, some niche directory, a review platform that happened to reference you. Your own page mattered less, good or bad.
That shifts where the leverage lives. On Gemini and Perplexity, the answer's mostly a stronger page on your domain. On ChatGPT, you have to factor in the directories too, because close to half the credit sits there regardless of what you do with your website.
Why this trips owners up
The reflex is to treat "AI visibility" as one problem with one solution. That reflex comes from SEO, where a better page climbs one ranking that every engine respects. AI citations don't play that game. The same business, the same question, run through four engines, produces four different verdicts on who gets credited and from where.
Take a med spa with a tight, detailed page on one specific treatment. They'll perform on Gemini and Perplexity because those engines already prefer citing the business directly. That same med spa, if its Yelp profile is sparse and its directory listings stale, keeps bleeding ground on ChatGPT no matter how sharp the website gets, because ChatGPT draws from those secondary sources almost as often as from the business itself.
Flip the script and take a chiropractor with solid Yelp traction and listings in every local directory, but a website that's basically a homepage and a phone number. ChatGPT keeps naming them through directories it already trusts. Gemini and Perplexity, hunting for the business's own site, find slim pickings and quietly elevate a competitor whose site actually answers the question.
What we see in our own data
We haven't sliced our citation database by engine yet, so I can't hand you a matching figure here, and I'd rather state that flatly than pretty up a guess. What we can say: across our broader dataset, just under a third of citations already route to sources we've tagged as known directories rather than the business's own page. That's under ChatGPT's 44.7% in the Cheers study, but it's the same contour, a meaningful chunk of credit parked outside pages you control.
What to do with this
Figure out which engine your customers actually use before you decide where to spend the next hour. A quick free check shows whether ChatGPT and Perplexity mention you at all, and for either one, note what the citation points to: your site or somewhere else.
If it hits your site, keep building that page out, since the credit is compounding where you can shape it. If it hits a directory, fixing your website alone won't touch that number. The listing itself, the review count, how you show up on Yelp and local platforms, that's the piece actually driving the answer.
The uncomfortable truth in this study is that no single fix works the same across every engine. The useful truth is that once you know which engine your customers favor and where its citations land, you know exactly which lever to pull.
Source: Cheers, "AI Search Engine Source Differences" — 32 home-services prompts across 100 US and Canadian metro areas, four engines, 28 days ending September 2, 2026, 356,019 citations classified by domain. Our own figures come from askchat.studio's ongoing GEO probe database.