Which Sources Do AI Engines Actually Cite?

If you knew the ten domains ChatGPT trusts in your category, link building would stop being a guessing game. No more chasing "authority" in general — you'd have a target list: get present, reviewed and mentioned on the specific sites the engine looks at when a buyer asks about your market.

That list exists, and it has been measured. A handful of published studies have logged hundreds of millions of AI citations and reported which domains keep showing up. This post pulls their findings into one picture.

Quick disambiguation, because this keyword leads a double life: this is about which websites AI engines cite as sources in their answers, not how to cite ChatGPT in an essay.

Why should I care?

Because citations are the currency of AI search. When ChatGPT, Perplexity or Google AI Overviews answers a buying question, it names a few brands and links the sources it built that answer from. Citations are to answer engines what backlinks were to SEO: the visible trace of trust, and the part you can influence.

Here's the core principle this whole post hangs off:

AI engines don't cite the whole web — each one leans on a short, knowable list of domains it already trusts.

Short: on Perplexity, 64% of URLs never get cited at all, while roughly 6% of URLs produce just under half of all citations. Knowable: the studies below converge on the same names, over and over. That's the whole opportunity: small enough to matter, public enough to find.

What the studies agree on: the top domains cited by AI

The widest-angle view is Peec AI's March 2026 analysis of 30 million sources across ChatGPT, Google AI Mode, Gemini, Perplexity and AI Overviews (US data). The top ten most-cited domains, in order: Reddit, YouTube, LinkedIn, Wikipedia, Forbes, G2, Yelp, Facebook, Medium, Techradar. Reddit came in #1 or #2 for every LLM tested. One honesty note: Peec publishes rank order only, no percentages — a leaderboard, not a share chart.

Look at what's on that list: not brand homepages, but places where many voices compare things — forums, video, a professional network, an encyclopedia, publishers, review aggregators. AI engines lean on sources that have already done the comparing for them.

Each engine has its own flavour of the pattern. Peec's per-engine top fives:

Percentages do exist — they come from a different study with a different window. Profound's analysis of 680 million citations (August 2024–June 2025) puts Wikipedia at 7.8% of all ChatGPT citations — a huge 47.9% of its top-10 share — and Reddit at 6.6% of everything Perplexity cites. Google AI Overviews spread theirs far more evenly: Reddit 2.2%, YouTube 1.9%, Quora 1.5%, LinkedIn 1.3%. On ChatGPT, .com domains take 80.41% of citations and .org another 11.29%.

One non-vendor anchor, since most of this research comes from vendors: an academic study from the AI Disclosures Project analysed roughly 14,000 real conversation logs across 11 commercial LLMs and found 30% of answers provided no citations at all, Gemini offered no clickable citation in 92% of its answers, and Perplexity visits roughly 10 relevant pages per query but cites only 3–4. Citations aren't just concentrated — on some engines they're scarce.

How citation patterns differ by engine

Before you lump the engines into one "AI" bucket: the most useful finding in this whole research pile is that they mostly disagree. Ahrefs compared ~76.7 million AI Overviews with ~957,000 ChatGPT and ~953,000 Perplexity prompts and found 86% of the top-50 mentioned sources are not shared across the three — only 7 of the top 50 appear in all of them. There is no single "AI citation list" — there are several, overlapping far less than most people assume.

ChatGPT: encyclopedic, publisher-heavy — and volatile

So which sources does ChatGPT cite, and how does it choose them? Its lean is encyclopedic: Peec ranks Wikipedia first for ChatGPT, ahead of Reddit, Forbes, Techradar and LinkedIn, and in Ahrefs' June 2025 data Wikipedia's mention share on ChatGPT (16.3%) beat Perplexity (12.5%) and AI Overviews (8.4%).

It's also the engine most likely to change its mind. Semrush tracked 230,000+ prompts across 13 weekly snapshots (14 July–12 October 2025) and watched ChatGPT cite Reddit in close to 60% of responses in early August — then around 10% by mid-September. Wikipedia fell from roughly 55% of responses to under 20% in the same stretch, while holding steady at about 3% on AI Mode and 0.8% on Perplexity: the crash was ChatGPT-only. The gap was filled by PRNewswire, Forbes (which doubled its cited-response count) and Medium. If ranking in ChatGPT is your priority, expect its source diet to be rewritten month to month.

One more trait: ChatGPT is the most generous citer. In Peec's separate one-million-citation benchmark, 31% of its URLs sat in the top citation-rate bucket (2.0+) and produced 59% of citations, and 52% of cited listicles were cited more than twice. Once ChatGPT likes a page, it keeps coming back.

Google AI Overviews and AI Mode: balanced, conservative, very Google

Google's AI surfaces are the most balanced citers — in Profound's August 2024–June 2025 dataset, the top-10 shares run Reddit 21.0%, YouTube 18.8%, Quora 14.3%, LinkedIn 13.0%, with no Wikipedia-style dependency. They're also the most Google-flavoured: in Semrush's data, AI Mode's top-cited set is made up entirely of properties Google owns or has major partnerships with; it cited LinkedIn in nearly 15% of responses and Wikipedia in only about 2%. And they're the most conservative: Peec's benchmark found more than 9 in 10 URLs on AI Mode hold a citation rate below 1.0, with a slight preference for brand-owned category and product pages — one of the few places your own site gets cited directly. Getting into AI Overviews is its own playbook.

Perplexity: the most concentrated

Perplexity is Reddit's biggest customer: Reddit takes 46.7% of its top-10 citation share and 6.6% of all its citations (Profound, over the same August 2024–June 2025 window). It's also winner-take-all — 64% of URLs never get cited, while roughly 6% produce just under half of all citations (Peec). And per the arXiv study above, it reads about 10 pages per query and cites 3–4 of them. A hard editor with strong favourites.

The citation half-life: your spot on the list expires

Here's the finding almost nobody covers: citations decay. Scrunch AI and Stacker tracked 3.5 million citation events from September 2025 to March 2026 and found citation activity drops by half in roughly 4.5 weeks. By engine: ChatGPT is fastest at 3.4 weeks, Google's surfaces sit at 4.3–4.8, and Perplexity is stickiest at 5.8 — its citations last nearly 70% longer than ChatGPT's.

Industry matters too, just less than platform. In the same Scrunch data, insurance (5.0 weeks) and financial services (4.8) hold citations longest; healthcare (4.1) and retail (4.3) churn fastest. Real, but narrower than the engine spread — hence Scrunch's verdict:

"Platform strategy has more leverage than vertical."

— Scrunch AI × Stacker, The Half-Life of AI Citations

Domains inside Stacker's 4,000+-publisher partner network held citations about twice as long as average — distribution reach appears to buy persistence. One caveat, and it's Scrunch's own: these are aggregate persistence estimates across many prompts and many weeks, not what happens when you rerun one exact prompt. Treat the half-life as a planning number, not a countdown on any single answer.

Why the numbers seem to contradict each other

If you've read other roundups of AI citation research, you'll have noticed the stats refuse to match. Take Reddit — the most-studied domain in AI search — across four published findings:

• #1 or #2 by rank for every LLM in Peec's 30-million-source study.
• 1.8% of all ChatGPT citations in Profound's 680-million-citation dataset (August 2024–June 2025).
• Close to 60% of ChatGPT responses in early August 2025, around 10% six weeks later, in Semrush's weekly tracking.
• Outside the top 10 on ChatGPT and Perplexity in Ahrefs' June 2025 read — where Reddit and Quora ranked high only in AI Overviews (7.4% and 3.6% mention share).

All four are true at once — which is the kind of thing that makes people quit marketing and take up woodworking. Each study just divides by a different thing. Rank order counts raw appearances. "Percent of all citations" divides by every citation in a dataset, where millions of long-tail URLs drag every big name down to single digits. "Percent of responses" divides by answers, so one Reddit link among ten citations still counts the whole answer. "Share of the top 10" divides by only the ten biggest sources. Different denominators, different time windows — same web.

So here's the one test that beats every misleading citation stat, vendor decks included: ask "percent of what?" If the piece can't answer, the number can't help you.

How to find the sources AI trusts in your niche

Everything above is aggregate, mostly US data. Your buyers ask narrower questions, and the trusted list shifts with the niche. Peec's own caveat says it best:

"In the hotel space, platforms like Booking.com and Airbnb often outrank Reddit" as AI sources.

— Peec AI, Top Domains Cited by AI (March 2026)

The practitioner method is simple, slightly tedious, and doable this week:

1. Write 20–30 real buyer prompts — the questions customers actually ask, from "best X for Y" to "is [brand] worth it".
2. Run them across ChatGPT, Perplexity and Google AI Overviews. Three engines minimum, because of that 86% non-overlap.
3. Log every cited domain — one spreadsheet row per citation.
4. Frequency-rank the domains. Your top ten is your niche's trusted list.
5. Rerun monthly. Semrush watched ChatGPT's favourite source lose most of its response share inside six weeks; a quarterly check is too slow.

This is exactly the grunt work an AI visibility platform automates — Apex's citations report runs your prompt set across the engines on a schedule and keeps the frequency table current, alongside your share of voice. But the manual version costs nothing except an afternoon.

What to do once you have the list

A trusted-source list turns "how to get cited by AI" from theory into placement work: get present, reviewed and mentioned on the specific domains your engines already lean on. Where that usually points:

• Reddit — build a genuine presence in the subreddits where your category gets discussed; it ranked #1 or #2 for every LLM Peec tested.
• YouTube — publish comparison and how-to videos; it's a top-two ranked source on every Google AI surface.
• Wikipedia — if you have real notability, make sure your page exists and is accurate; it's ChatGPT's #1-ranked domain.
• LinkedIn — post actual expertise under real names; AI Mode cited it in nearly 15% of responses.
• G2, Yelp and your category's review sites — claim and complete your profiles and keep reviews flowing; both sit in the global top ten.
• Forbes, Techradar and your trade press — pitch expert commentary and original data; publishers gained the most when ChatGPT's Reddit reliance collapsed.
• Medium — republish your best thinking; it was a post-September winner on ChatGPT.
• Your own site — publish the statistics others will cite and keep category pages sharp; AI Mode slightly prefers brand-owned pages, and being the primary source is the one spot nobody can take from you.

FAQ

Which websites does ChatGPT cite most?

In Peec AI's 30-million-source analysis, ChatGPT's top five cited domains were Wikipedia, Reddit, Forbes, Techradar and LinkedIn, in that order. But ChatGPT is the most volatile engine: Semrush watched its Reddit citations fall from close to 60% of responses to around 10% in about six weeks in late 2025, so any snapshot ages fast.

How do I get my site cited by AI?

Map the domains AI engines already trust in your niche by running 20–30 real buyer prompts across ChatGPT, Perplexity and Google AI Overviews and logging every citation. Then get present on those domains — review profiles, community threads, expert contributions — and publish original, citable data on your own site so publishers and engines have a reason to reference you.

Do AI engines cite Reddit?

Heavily. Reddit ranked #1 or #2 for every LLM in Peec AI's 30-million-source study, and it takes 46.7% of Perplexity's top-10 citation share according to Profound (measured August 2024–June 2025). But the level varies wildly by engine and time window: ChatGPT's Reddit citations collapsed in September 2025, and Ahrefs' June 2025 data found Reddit ranking high only in Google AI Overviews.