In this article:
SHARE THIS
Somebody on the leadership team typed the company name into ChatGPT over the weekend. They didn’t love what came back, and now they’re knocking on marketing’s door for answers. Sometimes it’s the owner. Sometimes it’s a sales VP who got asked about it on a call and had nothing to say.
So you ran the same question through Perplexity and Google to see whether the picture was any better. Three answers came back, and the sources they cited barely overlapped.
The instinct at that point is to build three strategies, one per engine. That instinct will cost you months of work and hand you very little in return. Underneath the different answers, all three engines do the same thing at the step that decides everything.
We covered the writing side of this recently in how to write content that gets cited. This piece sits one layer below it: what happens before a word gets written, and why your page either makes that cut or never enters the conversation.
Every Engine Builds a Shortlist Before It Writes a Word
Picture a hiring manager filling one role. Two hundred applications come in. Somebody screens them down to six, and the interview happens with those six. Everyone outside that group is invisible to the decision, no matter how well they would have done in the room.
These engines run the same screening pass. Before generating an answer, each one assembles a candidate set of pages and then writes from what survived. Your content is competing to make the shortlist. Everything you have read about phrasing and schema only pays off after your page is in that pile.
That’s the shared mechanic. Where the three engines diverge is how the shortlist gets built.
| Google AI Overviews | ChatGPT | Perplexity | |
| Builds its shortlist from | Google’s index, plus its own expanded sub-queries | A retrieval pass, then synthesis | A live search pass built to cite as it answers |
| How much your Google ranking matters | Meaningful, and looser than it used to be | Retrieval position dominates everything else | Largely unmeasured |
| How visible the sourcing is to the reader | Links tucked beside the summary | Citations woven into the response | Sources on the surface of the answer |
| How much public research exists | Solid, and moving fast | The largest single study we have | Very little |
Google AI Overviews
AI Overviews draw on Google’s index, so your ranking matters more here than it does for the other two. It matters less than it did a year ago. Ahrefs found 76% of cited pages also ranked in the top ten in July 2025. When they ran it again in early 2026 across 863,000 keyword SERPs, that number came in at 38%. Ahrefs is candid that better detection in their own tooling accounts for part of the swing, so treat it as a range rather than a clean trend. Ranking still helps you here.
ChatGPT
ChatGPT retrieves first, then writes. The writing gets all the attention, and the retrieval step is what determines whether you’re even in the running. AirOps and Kevin Indig instrumented that pipeline across 16,851 queries and 353,799 pages, and retrieval position swamped every other factor: roughly a 58% citation rate at position one against 14% at position ten. One caveat worth carrying around with you, because it colors most of the advice you’ll read this year. That study measured ChatGPT and only ChatGPT, and its authors say so directly.
Perplexity
Perplexity shows its work, so the sourcing sits right on the surface of the answer instead of buried underneath it. It also has the smallest audience of the three on most measures and the thinnest body of public research behind it. If someone’s telling you confidently how to win in Perplexity, they’re extrapolating from ChatGPT data.
What the Differences Change, and What They Don’t
Reach differs. AI Overviews touch the widest audience and catch buyers early in research. ChatGPT gets the conversational category questions. Perplexity shows up later, during comparison. Measurement differs too, so Search Console tells you about one and GA4 referral filtering tells you about the others.
The work stays the same across all three. Every one of these engines has to find your page first. Then it has to decide the page answers the question well enough to quote. A page that gets shortlisted by two engines usually earns it the same way it earned the first one.
How You Get Into the Shortlist
Rank for the question.
Retrieval position outranked everything else the AirOps team measured, and the margin is wide. A page at position one got cited around 58% of the time against 14% at position ten. The SEO work you already do is the shortlist work.
Match your heading to the question.
Pages whose primary heading closely matched the query language got cited 41% of the time against 29% for headings that wandered. Say you’re the marketing director at a 200-person specialty coatings company, and your best service page is titled “Advanced Protective Solutions” while the plant engineer you want is asking which coating survives chemical processing. Rewriting that one line will do more for your AI visibility than months of link building.
Narrow the page.
One page, one question. Answer it all the way through. The AirOps study found pages covering 26% to 50% of the subtopics around a query got cited 38% of the time against 34% for pages covering everything, which cuts against a decade of comprehensive-guide advice. Broad coverage reads as generalist, and generalist loses.
Common Questions About How AI Engines Choose Sources
Do we need a separate strategy for each AI search engine?
No. All three build a candidate set of pages before generating anything, and the work that gets you into those sets is the same. Be findable first. Then match your heading to the question a buyer asked and answer it cleanly. Where the engines differ is reach and measurement, which shapes how you report on AI search rather than how you plan content. Engine-specific tactics are also the fastest-decaying advice in the field, given how far the numbers moved in a single year
Does domain authority get us into the shortlist?
The research is split, and we would rather say so than pick a side. The AirOps study of ChatGPT found no positive correlation between domain authority and citation, with always-cited pages averaging lower authority than never-cited ones. Other studies have found broader authority signals meaningful. Everyone agrees that authority still helps you rank, and ranking still helps you get retrieved. An engine still needs the page to answer the question before it will quote it.
The Shortlist Is the Whole Game
An engine can only cite what it retrieved. Retrieval decides which pages are in the running. Quality decides which of those gets quoted.
Find out where you stand. Take ten questions your buyers ask and run them through all three engines, then write down who got cited each time. The pages that show up in more than one place will have a few things in common. They were findable. Their headings matched the question somebody typed. They answered it without wandering into everything adjacent.
If you want a second set of eyes on where to start, we’d love to take a look with you.
Share this article with your network.