Social Citations Differ 26x Between ChatGPT and AI Overviews: What an Agent Should Watch

One number from Promptwatch's social media citations report should change how you brief an agent: in Google AI Overviews, social platforms made up 12.2% of citations. In ChatGPT they made up 0.46%. That is roughly a 26x gap between two engines your buyers may use on the same afternoon.

The window is one week, September 28 to October 4, 2026, covering ChatGPT, Perplexity and Google AI Overviews. It is one vendor's dataset, built from real-UI prompts. It does not tell you why the engines differ, and it says nothing about your market or language. Treat it as a snapshot of where things stood, and as a reason to stop planning "social" as a single channel.

What the report found

Averaged across the three engines, the share of all citations looked like this: YouTube 1.70%, Reddit 1.03%, LinkedIn 0.90%, Facebook 0.81%, Instagram 0.56%, TikTok 0.07%, X 0.04%. Those averages hide the interesting part, which is the split by engine.

ChatGPT: all seven platforms together are 0.46% of its citations, down from 5.8% in the earlier report. Reddit fell from 5.19% to 0.09% after the August collapse. LinkedIn is now its most cited social platform, at 0.3%.
Google AI Overviews: YouTube 4.28%, Facebook 2.38%, Instagram 1.61%, LinkedIn 1.55%. Facebook more than doubled and passed Reddit's 2.15%. TikTok shows up only here, at 0.22%.
Perplexity: 2.7% from social. LinkedIn and Reddit tie at 0.86%, and YouTube sits at 0.79%, down from 2.67%.
X: it barely registers anywhere and peaks at 0.1% on Perplexity.

The report's own suggestions follow from that: for Google-facing buyers, start with YouTube and keep public Facebook pages and Instagram profiles complete; do not lean on Reddit threads for ChatGPT; for Perplexity, add LinkedIn alongside Reddit; keep X out of an AI search plan.

What it does not mean

A low social share in ChatGPT does not mean social is irrelevant to your brand. It means that in this window, a small slice of ChatGPT's citations pointed at social platforms. The sharp fall from 5.8% also tells you the mix can move fast, and the report ties the Reddit drop to August. Whatever you build from this snapshot has a shelf life, and the right response to a volatile mix is a standing check, not a one-time plan.

What to do without any tool

Pick ten prompts your buyers really ask. Run each in ChatGPT, Perplexity and Google, open the sources, and mark every social link by platform. Do it on a Monday. Do it again in four weeks. That crude log already tells you whether your category follows the report's pattern or breaks it.

Then fix the cheap things the report points at. Complete the public Facebook page and the Instagram profile. Make sure the LinkedIn company page says what you sell in plain words. If you produce video, check that titles and descriptions match how people phrase the question. None of that guarantees a citation. It removes reasons to be skipped.

The agent loop: monitor, schedule, fix

This is where an agent earns its keep, because the work is repetitive and the mix keeps moving. Think of three jobs.

Monitor. The agent reads social citations per engine for your tracked prompts, not in aggregate. The question it answers each run: which platforms are cited for our prompts on each engine, and did that change since last week?

Schedule. The check has to recur. A weekly run is a reasonable default given that the report's window is a single week and the Reddit shift happened inside a month. Compare each run to the last and flag a platform that moved by a meaningful amount, with the threshold set by you.

Fix. The agent turns a flagged change into a small, reviewable to-do: refresh the Facebook page copy, brief a video, rewrite the LinkedIn company description. It does not post or reply in communities for you. A human approves anything public.

How Promptwatch fits this loop

Promptwatch maps onto each job. Its social citation views (Reddit, YouTube, LinkedIn and Instagram, under Sources, Social media) show what is being cited, and the "My channels" filter narrows the view to your own accounts, so the agent can separate "our LinkedIn page is cited" from "someone else's is." Reddit, YouTube and LinkedIn reports exist as separate views, and Instagram reporting covers posts and accounts with CSV or Excel export.

For the schedule, Scheduled Chats let Agent Chat run a prompt such as a weekly social citation briefing and deliver it in the app or in Slack. For the fix, the Action Board holds the to-do list: its Discovery agent investigates your data weekly and suggests up to five evidence-backed actions per run, which you can assign and comment on. If a gap turns out to be on-site content, the Content Agent (Business and Enterprise) can plan and draft the page against it, with a review inbox so a person approves before anything publishes.

On cost, the social views are not a reason to jump a tier on their own. The paid brand plans start at Essential ($95/mo), and agency plans start at Kick-off ($199/mo). The plan for Instagram citations is not stated in Promptwatch's changelog, so check the plan matrix before you buy on that feature alone.

Most monitoring-only trackers can tell you a platform was cited. The loop above also needs a schedule, a prioritised to-do list and a drafting step, and that combination is why Promptwatch is the one to evaluate first for agentic work.

A short checklist to hand your agent

List the prompts, and tag each with engine and country.
Pull social citations per engine, weekly, with "My channels" on and off.
Flag any platform whose share moved by your threshold.
Draft one fix per flag, in the Action Board, with a human owner.
Re-measure after four weeks before concluding anything worked.