Why Perplexity and ChatGPT rarely cite the same sites

If you ran the same informational queries through ChatGPT Search and Perplexity and looked at which domains got cited in each, only a small share would appear in both lists. One published study compared the API versions of the two engines on two product categories and found a cited-domain overlap of roughly 9 to 10% (research finding). The engines also differed in how varied and how fresh their sources were, and in how much the wording of a question changed the answer. Treat that figure as a signal of direction, not a constant, and measure it on your own queries.

Low overlap is not the same as no spillover

It is tempting to read that sliver as "the two engines are separate games, so work for one does nothing for the other". The study does not say that. It measured which domains were cited, not whether the same pages were available to both engines or whether ranking work for one helped the other. The earlier version of this article leaned toward the "separate games" reading; this version corrects it.

What the platforms document points the other way on inputs:

• ChatGPT Search works from third-party search providers together with OpenAI's own crawler, OAI-SearchBot, and content from partners (official guidance). OpenAI's main search pages do not say which providers it uses today or how much each one contributes.
• Perplexity runs its own crawler, PerplexityBot, to find and link pages in its answers (official guidance). Its documentation does not say whether it also draws on other search indexes, or what it ranks on.
• Google's AI Overviews and AI Mode retrieve from Google's own Search index, often by running several related searches, and a supporting link must be indexed and eligible for a snippet (official guidance). So Google's index is the pool those features cite from, though Google does not say how it picks among eligible pages.

My reading of these documents is shared inputs, different selection: the engines draw on overlapping web indexes, then each picks and orders a few sources its own way. That is a hypothesis, and nobody has measured how big the shared part is. You can test it on your own queries: for every URL an engine cites, log where that URL ranks in Google and in Bing for the same query. If cited pages cluster near the top of an index, ranking work there is feeding citations. If they do not, selection is doing more of the work.

So the practical question is not "which engine do I pick?" It is "am I in the pools these engines retrieve from, and is my page the one worth citing once it is retrieved?"

What ChatGPT Search appears to select for

OpenAI does not name its current search providers on its main search pages, so don't build a plan on any single one. What you can control is being findable in the large web indexes, and not blocking OAI-SearchBot (OpenAI says sites that opt out of it are not shown in search answers, apart from navigational links). Many reports tie ChatGPT's results to Bing, so a cheap step most indie sites skip is to submit your sitemap to Bing Webmaster Tools and check that your key pages are indexed there. It costs an hour and helps any engine that draws on Bing; it is not a documented ChatGPT requirement.

Beyond indexing, these are my observations of what ChatGPT tends to cite, not documented ranking factors:

• Information density per paragraph. Long-winded intros seem to get skipped; cited passages tend to be the ones that state the most facts in the fewest words.
• Clear H2 hierarchy that matches likely question phrasings ("What is X", "How to Y").
• Who made it: named author + bio + visible publish/update dates, so readers can see who wrote the page, how, and when.
• Schema.org Article markup, plus FAQPage only for Q&A that is visible on the page. Mark up visible content accurately; no lift is guaranteed.

Some practitioners argue that ChatGPT gives little weight to raw backlink counts or domain age. Others argue the opposite: that authority carries through whatever search provider ChatGPT retrieves from, and that long-established brands get mentioned more often. Neither view is verified. Newer pages on smaller domains can win citations against older sites, but nothing guarantees it, and if ChatGPT draws on a search index, whatever that index rewards still shapes the pool it chooses from.

What Perplexity selects for

Perplexity runs its own crawler, PerplexityBot, and selects with different habits from ChatGPT. Whether it also draws on other indexes is not documented, so check that PerplexityBot is not blocked in your robots.txt or by your CDN's bot rules before anything else.

The preferences below are observations and practitioner reports, not documented ranking factors:

• Backlinks and topical authority: practitioners often report that Perplexity's picks look closer to traditional SEO than ChatGPT's. Perplexity has not said what it ranks on, so treat this as a hypothesis to check against your own citation log.
• Data-rich content: original statistics, charts, methodology. Generic "best practices" posts lose to specific case studies.
• Recency: recent stats and an honest dateModified. Change the date when the content changes in substance, and show the same date on the page. Changing dates just to look fresh is something Google's guidance warns against.
• Cleaner extraction: short, fact-dense paragraphs. Same as ChatGPT, but Perplexity is stricter — convoluted prose loses faster.

Perplexity also often cites discussion-platform content (Reddit threads, niche forums) when those threads contain specific testimony or technical detail (our observation, not a documented weighting). A useful forum answer can be cited ahead of your own site for the same query.

What this means tactically

You can't write a single "GEO-optimized" article and assume both engines will cite it. The citation overlap is small, because each engine selects differently. But you also don't need to write twice. Much of what gets a page into the retrieval pool, such as being crawlable, indexed and ranking for the query, is shared, and so is what makes it worth citing. The engine-specific part is finishing.

If the shared core is most of the work, two engines cost little more than one article. The ~70% split is an illustration; measure your own.

Shared (the core that helps both):

• Schema.org Article markup (FAQPage only for Q&A visible on the page; optional)
• Named author with bio + Person schema
• Visible publish + update dates
• Short, fact-dense paragraphs
• Clear H2 hierarchy matching real query phrasings
• Internal linking that disambiguates entities

ChatGPT-specific finishing:

• Make sure OAI-SearchBot is not blocked
• Submit your sitemap to Bing Webmaster Tools and check key pages with Bing URL Inspection (cheap cover for any engine that uses Bing; OpenAI does not name its providers)
• Clear, visible steps on procedural content (HowTo markup is optional; Google has shown no HowTo rich result since 2023)
• Audit content for "in this article" intros — strip them; in our observation ChatGPT tends to quote direct starts

Perplexity-specific finishing:

• Make sure PerplexityBot is not blocked
• Earn a few links from topical neighbors (helps the page rank anywhere; Perplexity's own weighting is unknown)
• Include at least one original number, chart, or methodology specific
• Update dateModified when the content changes in substance, and show the same date on the page
• Take part honestly where your buyers already discuss the topic (Reddit threads, niche forums)

How to measure

The only way to know what's actually working is to track citations. The cheap route is a Playwright script that runs each target query several times per engine every week, logs which URLs are cited, logs failed runs instead of dropping them, and reports per-query fractions (cited in k of N runs). One run is a snapshot, because answers vary from run to run. Add one column: where each cited URL ranks in Google and Bing for the same query. That column is how you find out how much of the citation pattern your rankings explain. The paid route is a tracker such as Otterly, Profound, Peec.ai or AthenaHQ. When we checked on 2026-10-06, published entry plans ran from about $29/month (Otterly) to about $295/month (AthenaHQ); Profound's and Peec.ai's pages didn't show prices to our fetcher. Prices change often, so recheck.

For Google, start with Search Console. The Generative AI performance report shows how often your URLs appeared in AI Overviews, AI Mode and Discover AI features. Google's help pages describe it as an impressions report and we saw no clicks metric in them, so check the report in your own account before computing a CTR from it. Treat any tool that claims Google's internal metrics with suspicion.

The expensive mistake is optimizing without measurement. The course's measurement module (planned) covers a DIY version of this script in more depth.

TL;DR

ChatGPT and Perplexity cite mostly different sites, but low overlap is not proof that work for one is wasted on the other. Both retrieve from the web, then select differently. Get into the indexes first, build the shared foundations once, then finish for each engine. Measure citations alongside the organic rank of every cited URL. Don't write twice; finish twice.