Right now, an AI is deciding whether to recommend your business — and it’s probably not reading your website to decide. It’s reading what other people said about you, on platforms you may never have posted to.
For two years, one site sat at the centre of that decision more than any other: Reddit. Then, in a single week in August 2026, that changed dramatically on one of the biggest AI platforms — and most businesses never adjusted. If your AI-visibility plan is built on last year’s advice, there’s a good chance it’s now pointed at the wrong place. This post shows you where it’s actually pointed, and the engine split almost nobody explains.
The short version: Roughly three-quarters of what AI says about you is built from third-party sources, not your own site. But “get on Reddit” is no longer one strategy — the map splits by engine. Live-retrieval engines (Perplexity, Gemini, Google’s AI surfaces) reward being talked about in communities. ChatGPT and Claude increasingly reward being an authoritative source in your own right. You need both, and the map moves in weeks — so you monitor it, you don’t set it once.
The number that reframes everything
Start with the finding that reorganises the whole approach. Most of what AI says about you is not built from your website.
When Contently synthesised five separate 2026 citation studies, the headline was blunt: your own site accounts for roughly 25% of the citation equation; the other ~75% lives on third-party domains — the places other people talk about you. Most businesses have that ratio backwards, pouring effort into polishing the 25% while ignoring the 75%.
And the third-party side is extraordinarily concentrated. The 5W Citation Source Index, synthesising more than 680 million individual citations across ChatGPT, Claude, Gemini, Perplexity, and Google’s AI Overviews, found that just 15 domains absorb roughly 68% of every citation these engines produce. That’s a more extreme concentration than Google’s old PageRank ever created. To these systems, the web is not a level playing field — a small handful of sources compose most of the answers, and getting mentioned on them is the most direct lever you can pull.
(A note on these figures: each comes from a single vendor synthesis rather than independent replication. Treat them as strong directional evidence for how concentrated AI citations are — the exact percentages will vary by study, methodology, and query set.)
The Reddit trap: the engine split
The domain at the top of almost every one of those studies is Reddit. Peak AI’s analysis of 30 million cited sources ranked it the single most-cited domain across the major engines, with aggregate share estimates around 40%.
So the obvious advice writes itself: go all-in on Reddit. And that advice is a trap — because Reddit’s dominance is not spread evenly across the engines. It’s wildly lopsided, and on two of the biggest platforms it barely registers.
Here’s the split:
- Reddit dominates the live-retrieval engines — Perplexity, Gemini, Google AI Mode, and AI Overviews. On Perplexity, by some measures it approaches half of all citations.
- On ChatGPT and Claude, the story is completely different. An Otterly.AI study of over 370,000 Claude citations found Claude cited Reddit essentially zero times, and independent analysis suggested Reddit was never even supplied as a candidate source. And then ChatGPT changed too.
The August 2026 ChatGPT collapse
In the second week of August 2026, ChatGPT’s Reddit citations fell off a cliff. The scale is disputed by tracker — Promptwatch measured an 86% drop (from a 3.83% share down to 0.52% between 8 and 14 August), while Qwairy measured closer to 95% across its monitored brands. The event was real enough to be covered by Forbes, Axios, and Search Engine Land.
The mechanism is the important part, and it’s confirmed: on 8 August 2026, alongside the GPT-5.6 rollout, ChatGPT changed how it searches the web — shifting its query fan-out toward site: operators that interrogate specific official domains directly, rather than pulling from broad discussion. This wasn’t a penalty aimed at Reddit. It was a change in how ChatGPT decides which sites to ask, and Reddit was the most visible casualty of a much broader redistribution toward official, structured, authoritative sources.
One thing the breathless coverage left out, and it matters more than the number: this exact collapse happened once before, in September 2025, and Reddit’s ChatGPT share fully recovered. That’s not a reason to dismiss the shift — it’s the whole lesson. The map moves in weeks, in both directions, which is precisely why a strategy bet on any single platform is fragile. Don’t set it once. Monitor it.
Why the engines disagree
The split isn’t random. AI engines fall into two families based on how they get information, and once you see that, the whole map makes sense.
Family one runs live web retrieval — Perplexity, Gemini, Google AI Mode, AI Overviews. They search the open web in real time for nearly every query, and Reddit is exactly what they want: fresh, specific, experience-based discussion, real people comparing real products. When a system is retrieving live and looking for “what do people actually think,” a karma-weighted forum thread is gold.
Family two behaves differently. When ChatGPT shifted in August toward directly interrogating official sites with precise queries, citations flowed away from third-party forums and toward official, structured, authoritative sources. Claude, by design, tends to bypass the social and encyclopedic layer and go straight to primary sources and brand-owned reference content — one study found it cites brand domains around 64% of the time.
So there’s your map. Live-retrieval engines reward you being talked about in communities. ChatGPT and Claude increasingly reward you being an authoritative, well-referenced source in your own right. Those are two different jobs, and you need both — the opposite of “just do Reddit.”
Why communities get cited at all
One quick “why,” because it tells you what good looks like everywhere else. Why do the live-retrieval engines trust Reddit so much?
Because a company saying it’s the best is marketing, and a hundred strangers independently comparing options is consensus. Reddit threads contain specific claims, real user comparisons, named products, and current discussion — exactly the extractable, quotable evidence these systems are built to lift. Most brand websites, by contrast, contain polished marketing language with nothing concrete to quote.
The signal isn’t link authority. It’s discussion depth and corroboration. Carry that into everything: be specific, be corroborated, be quotable. It’s also the principle that makes a genuine content strategy work for AI — reference-grade pages that answer real questions with concrete, citable substance, not marketing gloss.
How to show up — and the line you don’t cross
Here’s where people torch their own reputation, so be careful. You do not win these platforms by faking them. Astroturfing Reddit — fake accounts, planted recommendations, manufactured threads — is against the rules, increasingly detected by both moderators and the AI ranking signals, and a genuinely bad trade: you burn an account and risk the brand for a signal you could have earned honestly. We work across the full spectrum of technique and will always give you the honest risk maths — and here the maths is clearly negative. Promotional posts get filtered within hours.
The legitimate version splits by engine family.
For the live-retrieval engines (Reddit, Perplexity, the Google surfaces), the job is genuine community presence:
- Be genuinely present where your category lives. Find the subreddits, forums, and communities that map to your space, and participate as a real, useful contributor — answering questions, sharing specific experience, contributing frameworks. Earned reputation in a community is a durable asset and the only kind these systems reward for long.
- Be worth mentioning. The reason a stranger names a business in a thread is that it’s genuinely good and genuinely known. Your job is to be the answer people reach for, not to plant it. That’s downstream of a real product and real recognition.
For the brand-authority engines (ChatGPT, Claude), the job is different — and it’s the half the “just do Reddit” crowd misses:
- Become a citable authority in your own right. Because Claude leans heavily on brand domains and ChatGPT is shifting toward official sources, your own site’s reference content matters more here, not less. Clear, specific, well-sourced pages that directly answer the real questions — the reference-grade material these engines pull straight from the source.
- Earn third-party editorial and review coverage. An SE Ranking study of 129,000 domains found that domains listed across multiple review platforms (the G2s, Capterras, and Trustpilots of a category) earned roughly 4.5–6 citations on average, versus under 2 for domains absent from them. Review platforms are third-party validation these engines treat as authoritative for exactly the “which should I choose” questions that matter commercially.
This dual job — community presence and citable authority — is the core of what AI search visibility work actually is in 2026, and the earned-reputation and impersonation-monitoring side of it overlaps directly with brand protection.
YouTube: the most underrated lever
One source deserves singling out, and it’s where this very content lives. In Ahrefs’ December 2025 study of 75,000 brands, YouTube mentions were the single strongest correlation with AI visibility of anything measured — roughly 0.737 across ChatGPT, AI Mode, and AI Overviews. Nothing else came close as a single predictor.
The mechanism is the key insight: AI engines read the transcript and the description, not the pictures. The visual content is incidental. So a brand mention in a well-watched video — a review, a comparison, a tutorial — carries the same kind of signal as a mention in a top editorial article, and can generate citation lift for months. YouTube is also owned by Google, which makes it doubly valuable across both classic and AI search on Google’s own surfaces.
One honest caveat, which Ahrefs themselves flag: this is a correlation, not proven causation. Brands with high YouTube presence also tend to be brands with strong cross-platform recognition generally — the mentions are partly a proxy for “this is a real, known brand.” Opening a channel doesn’t trigger citations by itself. But the practical takeaway holds: get your brand into genuinely useful video content in your category — clearly titled, properly described, actually answering the question. If you’re absent from YouTube for your space, you’re absent from a source the AI is actively reading.
Measure it, because it moves in weeks
This is the discipline that separates the businesses winning here from the ones guessing. This landscape moves in weeks, not years — the August ChatGPT shift happened in days, and Reddit’s ChatGPT share has swung wildly and partially reverted before.
So don’t set a strategy once and walk away. Build a simple habit:
- Take your top 20 category questions — the ones a real customer would ask.
- Run them through ChatGPT, Perplexity, Claude, and Google’s AI answers.
- Note which competitors get named, and crucially, which source domains each engine cites.
- Re-run it quarterly at minimum.
That’s your map, per engine, and it costs nothing but time. It’s also the natural starting point for any serious engagement — the same diagnostic logic behind a site audit, extended to the AI layer.
The honest reframe
If roughly 75% of your AI mentions come from off your own site, then off-page isn’t a bolt-on to your AI strategy — it is most of the strategy.
But here’s the 2026 update most people are missing: off-page no longer means one thing. It means genuine community presence for the live-retrieval engines, and citable, authoritative, well-referenced brand presence for ChatGPT and Claude. Balance your effort across both. Keep your own site sharp and quotable as the anchor. And stop betting the whole strategy on a single platform that can change in a week.
Most businesses have this ratio backwards and this map out of date. Fix both, and you’re building on the right 75%.
Watch the video
This post accompanies our full video breakdown of off-page AI search and the engine split.
Where to go next
Most of what AI says about you comes from third-party sources; a tiny handful of domains compose most answers; and the map splits hard by engine — communities for the live-retrieval engines, authoritative brand and reference content for ChatGPT and Claude. The only durable way to win any of it is genuine participation and genuine reputation, never astroturfing. That’s a losing trade every time.
If you want help building that cross-platform presence properly — and figuring out which engines your buyers actually use — that’s the core of our AI search visibility work. See how it fits with the rest of the approach across the GrowifyLabs solutions catalogue, or start at growifylabs.com.
Seen the ChatGPT-Reddit shift in your own results? Did your citations move with it? That’s exactly the kind of change worth comparing notes on.
FAQ
Where do AI engines get the information they cite about a brand?
Mostly from third-party sources, not the brand’s own website. A Contently synthesis of five 2026 studies estimated that a brand’s own site accounts for only about 25% of the citation equation, with roughly 75% coming from third-party domains. Citations are also highly concentrated: the 5W Citation Source Index found around 15 domains absorb about 68% of all citations across the major engines.
Does Reddit still help with AI visibility in 2026?
It depends entirely on the engine. Reddit dominates live-retrieval engines like Perplexity, Gemini, and Google’s AI surfaces, but on ChatGPT and Claude its influence is minimal. In August 2026, ChatGPT’s Reddit citations dropped sharply (by 86–95% depending on the tracker) after a change in how ChatGPT searches the web. A similar collapse in September 2025 later recovered, which is why single-platform bets are fragile.
Why did ChatGPT stop citing Reddit in August 2026?
On 8 August 2026, alongside the GPT-5.6 rollout, ChatGPT changed its query fan-out to use site: operators that interrogate specific official domains directly, rather than pulling from broad web discussion. This redistributed citations away from forums and toward official, structured, authoritative sources. It was a retrieval change, not a penalty targeting Reddit specifically.
What’s the strongest single signal for AI search visibility?
In Ahrefs’ December 2025 study of 75,000 brands, YouTube mentions showed the strongest correlation with AI visibility (~0.737 across ChatGPT, AI Mode, and AI Overviews) — because AI engines read video transcripts and descriptions. Ahrefs caution that this is correlation, not proven causation: high-visibility brands also tend to have broad cross-platform presence.
Is posting on Reddit to get cited by AI against the rules?
Genuine participation is fine and encouraged. Astroturfing — fake accounts, planted recommendations, manufactured threads — is against platform rules, increasingly detected by both moderators and AI ranking signals, and a poor trade that risks the brand for a signal you could earn honestly. Promotional posts are typically filtered within hours.