What websites does ChatGPT use when it recommends software?

ChatGPT uses websites as sources when it names software in an answer. That matters this year because a buyer who asks ChatGPT before visiting your site may never see a page ChatGPT cannot cite.

The short answer is Reddit is the largest named domain in our scan, followed by a long tail of sources that changes with the software category and buyer question. The useful task is to find which pages appear for your buyers, not to chase a universal list.

The figures come from CiteGraph's weekly research on who owns AI answers: 45,309 citations across 7,260 AI answers. Those answers came from 60 scans of 41 sites across 44 categories, using 442 distinct buyer questions. The engines cited 7,624 domains and 16,506 pages. You can also read how CiteGraph measures the sample.

This is a sample, not a map of every ChatGPT answer. We measure the questions and sites in the scans; we do not know what we have not measured. Our data reports AI answers as a group, so it does not isolate ChatGPT from every other engine. Engine by engine differences are in AI citation patterns by engine, and the full domain list is in The domains AI cites most.

Domains

Reddit is the largest named domain in the scan data, with 1,482 citations across 44 categories. GitHub followed with 481 citations across 26 categories, LinkedIn had 338 across 32 categories, and Apple had 323 across 21 categories. These figures show reach, not the role each domain performs in an answer.

The next named domains were Virlo.ai with 304 citations across 3 categories, Otterly.ai with 285 across 1 category, Medium with 267 across 31 categories, TikTok with 230 across 5 categories, G2 with 203 across 30 categories and YouTube with 201 across 31 categories. Further down were Tryprofound.com with 183 citations across 3 categories, HubSpot with 176 across 15 categories, Semrush with 166 across 11 categories, Ahrefs with 163 across 10 categories and dev.to with 148 across 18 categories.

The ledger does not confirm what role any cited domain plays in a particular answer. A domain can appear often overall and still be irrelevant to your category or buyer question.

The ten most-cited domains account for 9.1% of all citations. A further 885 domains have ten or more citations, and 84 have fifty or more, so the sample has a long tail rather than one universal source.

Page types

Listicles and roundups account for 67.5% of recorded citations, or 30,585 citations. Competitor product pages account for 24.4%, or 11,052 citations. Together, those page types account for 91.9% of citations in the sample.

That is the main structural fact. Software recommendations often rely on pages that compare products, explain alternatives or gather several options in one place.

Community threads account for 3.4% of citations, or 1,534. Social posts account for 0.8%, or 364, review platforms for 0.8%, or 348, and YouTube for 0.4%, or 201. News accounts for 0.1%, or 60, while Wikipedia accounts for 0%, with 21 citations recorded.

The figures do not make your own site unimportant. They show that your product page is one possible source among many, so it needs clear facts that comparison pages and other sources can use.

Use this checklist when you inspect the pages that appear for your category:

• Identify whether each cited page is a roundup, competitor page, community thread or another page type.
• Record the exact product facts the page uses when it recommends a tool.
• Check whether the page compares products on a buyer concern that your own site leaves vague.
• Look for domains that recur across the questions that matter to your category.
• Separate a domain with broad reach from a page that appears for your specific buyer question.
• Check whether your product has a clear alternative, comparison or use case page that supplies the missing facts.

Build for the page type first, then check whether the page fits your category.

Category spread

Domains do not spread evenly across categories. Reddit appears across the widest category range among the named domains, while GitHub, LinkedIn, G2, Medium and dev.to also cover multiple categories. CiteGraph does not have category-by-category winner data in this ledger, so we cannot name the leading page for a particular category from these figures alone.

The category counts still give you a useful starting point. A domain with broad reach is worth checking, while a domain with narrow reach needs a closer look at the exact buyer question.

The right target is a relevant page, not the biggest domain in the sample.

Checks

You can run a useful source check in under an hour:

1. Write the buyer questions that describe the software choice, including alternatives and use cases.
2. Run those questions through ChatGPT and save every cited domain and page.
3. Group the results by page type, then mark the domains that recur within your category.
4. Compare those pages with your own product, comparison and use case pages.
5. Fix the factual gaps first, then repeat the test with the same questions.

Keep the evidence tied to the exact question. A domain cited for one category does not automatically become a useful target for yours.

You can use the free AI visibility checker to inspect your citation footprint, or start with the free AI readiness check before changing pages. These checks do not turn a sample into a guarantee, but they can show where your pages are absent or unclear.

The practical work is narrower than getting cited everywhere. Find the pages that appear for your buyers, understand the facts they make easy to use, and publish the missing comparison or use case information in a form other pages can verify.

Reddit and comparison pages carry most of the citations in the scanned AI answers that recommend software, but which page appears still depends on the category and the question.