15 Domains Supply 68% of Every AI Answer's Citations

Fifteen domains supply roughly 68% of every citation that ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews produce. If you publish developer content, that number decides where your effort is worth spending.

The figure comes from the AI Citation Source Index 2026, which ranks the fifty most-cited domains in AI answers and is synthesized from six independent studies covering 680M+ citations (Everything-PR, updated August 2026).

Who the super-sources actually are

The index's load-bearing rows are not the ones most content teams plan for:

• Reddit leads the consolidated index at roughly 40% of citations — the single most-cited domain family across the engines studied.
• Wikipedia is second, at 26-48% of ChatGPT's top-10 citations. In some categories nearly half of ChatGPT's top ten comes from one nonprofit encyclopedia.
• YouTube is ~19% of Google AI Overviews' top-source share, meaning Google's AI answers lean on video transcripts far harder than text-first teams assume.
• LinkedIn and Forbes complete the top five.

Format matters as much as domain. A separate 2026 analysis from Evertune found roughly 50% of ChatGPT citations are listicles, and 58% of those are ranked lists.

The cliff the aggregate hides

Consolidated indices average across five engines with different retrieval systems, and that average can conceal a collapse.

Reddit's share of ChatGPT Search citations held a steady 3.8% average from July 18 to August 7, 2026. On August 14 it fell below 1%, and the August 14-17 average was 0.52% — an 86% relative drop, tracked day by day by Promptwatch. Google's AI Overviews declined far more slowly in the same window, which is exactly why the consolidated index still shows Reddit on top.

The author's proposed explanation is that ChatGPT now compiles a shortlist of known brands before running its search, reaching for recognized entities instead of scraping forums. That is a hypothesis fitted to the data, not a confirmed product change — but the measured drop stands on its own.

The practical read: both popular takes on Reddit are wrong. It is not dying, because it leads the cross-engine index. It is not king, because one engine just walked away from it. Per-engine behaviour diverges faster than any annual index can track.

What to change this quarter

The write-up's recommendations translate cleanly into work a docs or devrel team can ship:

• Fix your entity definition first. Wikipedia's citation share is an entity-definition win, not a content win — neutral, structured statements of what something is. Write one sentence describing what your product does and repeat it verbatim on your site, GitHub org, docs, LinkedIn and directories. Models deduplicate across sources; give them the same sentence everywhere.
• Publish the ranked list your category deserves. With half of ChatGPT citations being listicles, comparison pages, alternatives pages and benchmark posts with real sourced rows are the highest-leverage formats. Each entry should stand alone: name, metric, one-line rationale, source.
• Put FAQ blocks on docs and product pages. Question-answer structure is the most quotable format there is. If your docs answer "how do I migrate from X" in the first paragraph, humans and models both win.
• Measure your own citation mix per engine, quarterly. The index is a market snapshot; yours is different and it shifts when engines change.

The Reddit item is the one to think about rather than copy. Treating it as a consistency asset — posting value, not links — compounds across the engines that still cite it heavily, even while ChatGPT does not.

What to watch next: whether ChatGPT's Reddit share recovers or holds below 1%. A sustained floor would confirm that engine-level source policy, not content quality, is now the variable that moves citation volume.