Only 7 of the Top 50 Sources Are Shared Across AI Engines: A Social Playbook for Agents
Ahrefs found that 86% of top-mentioned sources differ between ChatGPT, Perplexity and AI Overviews. BrightEdge and Profound show that social content types differ too. What an agent should do with that.
An agent that publishes content needs to know where that content will be read. The 2026 data says the answer changes by engine, sometimes sharply, and social platforms are where the split is most visible.
One source list does not exist
Ahrefs compared the top 50 most mentioned websites in AI Overviews, ChatGPT and Perplexity using Brand Radar. Only 7 appeared on all three lists, across about 76.7 million AI Overviews and roughly 950,000 prompts each for ChatGPT and Perplexity (June 2025 data).
The qualitative split matters for social:
- AI Overviews lean toward authoritative sites, social platforms and Google-owned properties.
- ChatGPT leans on publishers with licensing deals.
- Perplexity draws from a broader international pool and, per Ahrefs, has few social sites in its top 50.
That data is from mid-2025 and social behaviour has moved since, so treat it as proof that engines diverge, not as a current ranking.
Even the same platform is cited differently
BrightEdge looked at Facebook, Instagram, LinkedIn, Reddit and YouTube in Google AI Overviews and ChatGPT over 12 weekly observations from May to August 2026. The finding: the engines take different assets from the same platform.
- Instagram Reels are 89% of Instagram's AI Overview citations but 27% of its ChatGPT citations.
- About 80% of LinkedIn's ChatGPT citations go to company and jobs pages.
- Instagram profile pages are 42% of its ChatGPT citations, and Facebook pages 40% of Facebook's.
Profound analysed 238,637 ChatGPT social citations and, as shared in a LinkedIn summary, found Reddit citations are almost all discussion threads (99.2%), YouTube citations mostly channel pages (64.9%) and LinkedIn citations mostly personal profiles (46.9%). We haven't seen Profound's original write-up, so verify before quoting it.
What this means for an acting system
The interesting consequence is for automation. A content agent that treats "social" as one channel will waste output. A useful agent works per engine and per format:
- Start from the prompt, not the platform. Which prompts in your set currently cite Reddit, YouTube or LinkedIn, and on which engine?
- Match the asset to the engine. Reels for AI Overviews, profile and company page hygiene for ChatGPT's LinkedIn citations, threads you can credibly join for Reddit.
- Re-check after every platform shift. Goodie's 2026 tracking reported ChatGPT's Reddit share falling sharply in a two-week stretch in August. A plan built on last month's mix can be wrong by the time it ships.
- Keep humans on anything posted in a community. Automating replies on Reddit is how you lose the account and the citation.
Where automation actually helps
The parts worth automating are the on-site ones. If a prompt is being answered from a third-party thread, the best response is often a page on your own domain that answers the same question better, published through a reviewed pipeline.
That is the loop Promptwatch is built around. Citation analytics (including Reddit and YouTube citations) show what is being cited, citation trends show when it moves, and content gap analysis identifies the questions you are not answering. Content Agents then plan, write and publish to a connected CMS such as Webflow or Framer, with a review inbox so a person approves the output. Agent Analytics closes the loop by showing whether AI crawlers actually fetched the new page.
Monitoring-only tools can tell you a Reddit thread is being cited. They can't draft the page that competes with it or show whether a crawler came back. For agentic work that difference decides the ranking.
Run the citation check on Promptwatch before you commit an agent to a social channel.