Who Decides What AI Cites? Most Brands Are Watching the Wrong Signals
When I sit down with CMOs right now, the first question I usually get is some version of, “Where does AI get its answers from?” and, “Does my brand show up there?” That’s what our Q3 2026 AI Citation Trends Report is built to answer. Most brands are still working off last year’s assumptions rather than the new operating model taking shape.
Together with our partners at Profound, we tracked mid- to lower-funnel prompts across six AI platforms and nine commercial categories to see which domains are being cited, how often, and how that mix is evolving over time. What we found reinforces something we see consistently in Tinuiti’s client work: there isn’t a single “most important” source in AI search. Ask about beauty products in July and our Q3 data shows Ulta on ChatGPT, Amazon on Copilot, and Sephora on Perplexity. There are still clear patterns across platforms and verticals that marketers can use to make much smarter decisions about content, distribution, and measurement.
What AI Citation Patterns Reveal
When we talk about AI SEO or AEO with CMOs, we define it as understanding how AI platforms select and weight sources, then positioning your brand so those systems can confidently use you in their answers. That sounds straightforward until you look under the hood and see how differently the major assistants behave.
The six platforms in our testbed are ChatGPT, Perplexity, Google AI Mode, AI Overviews, Gemini, and Copilot. We kept the prompts steady from quarter to quarter so we could see how citation patterns shift over time, and where those shifts are large enough to change how a performance-minded marketer should act right now.
AI Citation Patterns in 2026: 4 Key Takeaways
Across four quarters of data, four themes stood out more than any individual chart:
- Social citations are settling into a new normal after Reddit’s retreat.
- There are persistent gaps between how different AI platforms cite the web.
- The ecommerce picture looks very different once you separate one outlier platform from the rest.
- A growing number of citation patterns are being decided in licensing deals and courtrooms rather than by algorithms alone.
Together, these patterns explain why your strategy has to start with your category, your buyers, and the specific assistants they rely on most.
What Happened to Reddit’s Grip on AI Citations?
Social media’s share of AI citations peaked at 11% in the first quarter of 2026, then settled back to 10% by July, and Reddit accounted for most of both the climb and the retreat. Our Q3 data puts Reddit at 4% of all citations in July, down from a 6% peak in Q1. YouTube moved the opposite direction, growing from 2% to 3% over the same stretch, and it’s now closer to Reddit than it has been at any point we’ve tracked. Instagram is moving too, climbing from 1% to 8% of AI Mode’s social citations since late 2025 and from 2% to 6% on AI Overviews.
I’ve watched brands stand up Reddit programs in January on the strength of a rising line, then they see it flatten by summer and assume they’d bet wrong. That isn’t the takeaway. When we unpack the data and combine it with our audit work, a different story emerges: AI systems aren’t rewarding brands on Reddit, they’re rewarding long-running, high-signal conversations that happen to live on Reddit. That was true when the line was going up. It’s still true now that it’s come back down.
From our analysis and external work, such as Profound’s deep dive into ChatGPT, roughly 99% of Reddit citations point to individual threads with substantive discussion, rather than subreddits, profiles, or brand-authored content. Large language models (LLMs) can’t have opinions, so when you ask “What’s the best moisturizer for sensitive skin?” they lean on places where people have already argued about “best” in public: listicles, detailed reviews, and multi-comment threads where people describe what worked, what failed, and why.
But averages hide the differences in how individual platforms behave. In mid-August, Promptwatch recorded Reddit’s share of ChatGPT citations falling roughly 86% in four days, from a steady 3.8% to about 0.5%, and our own Q3 analysis caught the same drop on August 14th. Over that window, Reddit’s share on AI Overviews slipped only slightly, from about 2.5% to 2.1%. No one has been able to explain this drop yet, and Promptwatch says to treat it as early. What’s not in question is that one platform can rewrite its citation profile in four days while the others barely move.
Where Reddit sits in your mix matters less than whether social shows up in your category at all. In July, apparel prompts pulled 14% of their citations from social platforms and electronics pulled 13%, while over-the-counter health prompts pulled just 4%. For an electronics or apparel brand, that means public opinion is shaping a real share of what buyers encounter. For an OTC health brand, social is a much smaller lever, and there are usually better places to spend the effort.
The bottom line for brands is that “show up on Reddit” was never a strategy on its own, and it’s less of one now. If your buyers lean on Google’s surfaces, YouTube deserves a harder look than it got last year. It grew across all three of them, and on AI Mode more than anywhere else. If they lean on Copilot, Reddit isn’t an option at all, and we’ll come back to why shortly.
How Should I Build a Google AI Strategy?
Clients often come to us with a single “Google AI” line item in their roadmap, as if Gemini, AI Mode, and AI Overviews are three skins on the same system. This wasn’t a safe assumption in Q1 and our Q3 data has only widened the gaps between them. These experiences cite the web differently, evolve at different speeds, and reward different distribution choices.
Start with how much each surface leans on social sources. In July 2026, our Q3 data shows AI Overviews pulled 18% of its citations from social platforms and AI Mode pulled 15%, while Gemini pulled 5%. Those are significant differences. If your buyer asks the same question in AI Overviews and in Gemini, one of them is roughly three times more likely to answer using someone’s social media post.
Strategically, this is where Tinuiti’s “cluster” thinking comes in. We treat Google’s AI surfaces as a family that shares technical foundations but diverges in distribution bias:
- AI Mode behaves like a broad, exploratory layer, increasingly pulling from Google’s own ecosystem alongside a wide social mix.
- AI Overviews sits closer to classic SERPs, but leans on social sources harder than either sibling, especially video.
- Gemini is the most restrictive of the three about where it will pull from, which has real implications if your audience skews toward Gemini as their daily assistant.
YouTube is the clearest example here. It grew from roughly 1% of AI Mode citations in late 2025 to 5% in July, making it one of the biggest drivers of that surface’s shift. On AI Overviews it sits higher still at 7%. On Gemini it accounts for just 1%, and has barely moved all year.
Google is also increasingly citing itself. Google.com grew to 4% of all AI Mode citations by July, up from well under 1% at the end of 2025, largely through Business Profiles and product cards carrying price, photos, and reviews. It sits at 0.4% in AI Overviews and is absent from Gemini entirely. For local and retail brands, that makes your Google Business Profile and product feed a citation surface rather than just a listing.
Rather than writing three separate strategies, we help brands build a single core AI visibility framework with a clean information architecture, structured mid-funnel content, and multi-source coverage. We then layer on surface-specific tests informed by citation data.
Is Amazon Really Winning AI Citations for Ecommerce?
Yes, but less than it looks at first glance, and one platform is doing most of the work. Across the six platforms we track, the 100 most cited ecommerce sites accounted for 12% of all AI citations in July, down from 14% in Q2. Amazon alone held 3% on average. On Microsoft Copilot, however, ecommerce sites took 34% of citations and Amazon by itself took 14%.
Pull Copilot out and Amazon’s position changes shape entirely. In apparel, its citation share falls from 3.7% to 0.5%. In home and garden, from 5.0% to 1.6%. Only in OTC health does Amazon hold up without Copilot carrying it.
Here’s why: Reddit blocks Microsoft’s crawlers and Microsoft honors the block, while Google and OpenAI both pay Reddit for access. With one of the largest social sources missing from its index, Copilot fills the gap with retailer pages. Ecommerce citation share on Copilot runs more than three times higher than on AI Mode, the next closest platform, and more than eight times higher than on ChatGPT.
So if your customers are on ChatGPT and Google, Amazon’s presence in the answers they see is just a fraction of what the average would imply. The retailers showing up most are category specialists.
In electronics, Best Buy held a higher citation share than Amazon on every platform except Copilot in July. In home and garden, Home Depot outranked Amazon everywhere except Copilot, where Amazon took 22% of citations. Beauty splits further still, as mentioned earlier.
For mid-sized ecommerce brands, that’s an opening. Map which retailer pages reliably surface in the assistants your buyers rely on, and structure product content so it can be cited even when your Amazon listing can’t be seen.
Who Decides Which Sources AI Can Cite?
Increasingly, the answer isn’t up to an algorithm, but a contract or court docket. Several of the insights we found trace back to commercial agreements and active litigation rather than to how well anyone optimized their content.
Licensing explains part of it, as Copilot’s missing Reddit access shows. The disputes are ongoing too. Reddit sued Perplexity in October 2025 over alleged scraping of its content through Google search results, and Perplexity lost its bid to dismiss that case at the end of July 2026. Reddit citations on Perplexity fell to 2% by July, down from 20% in Q1, and social citations overall dropped from 26% to 10% across the same span.
Amazon and Perplexity have been through a version of the same fight. Amazon sued Perplexity in November 2025 and won a court order in March 2026 blocking Perplexity’s Comet browser from making agentic purchases for Amazon customers. In early August, the Ninth Circuit vacated that injunction. Amazon had already started appearing in Perplexity citations by July at roughly 2%, after three straight quarters at zero, even though Amazon still blocks Perplexity’s crawlers.
Crawler policy explains a lot of what looks like platform preferences. Amazon blocks the crawlers behind ChatGPT, Gemini, Copilot, and Perplexity, but allows Googlebot and Bingbot, which power AI Mode, AI Overviews, and Copilot respectively. That’s why Amazon sits at zero on Gemini while holding its Copilot position. Platforms can also pick up citations indirectly, through blogs, forum threads, and publisher pages that reference a blocked site, which is part of why Amazon never disappears entirely.
None of this is something a marketing team can influence. What it does change is how you plan. When a single licensing deal or court ruling can strip eighteen points of Reddit citations off a platform in two quarters, concentration is the risk. Brands with coverage across their own site, retailer pages, publishers, and video are far less exposed to any one of these decisions than brands who bet on a single source.
How Should AI Citation Patterns Influence Your Strategy?
The vertical deltas matter more than the global averages. The gap isn’t only in how much social shows up in your category, it’s in which sources and which assistants decide the answer. OTC health and electronics sit at opposite ends of both: one where social barely registers and Amazon is strongest, one where social carries real weight and a specialist leads on the ecommerce side. If you’re in OTC health or B2B manufacturing and you treat “Reddit is down 86%” as your north star, you’re calibrating to someone else’s category.
At Tinuiti, this is where we push clients to slow down and ask:
- Which assistants do your buyers actually use, and do you know how differently each one answers your highest-value prompts?
- In your category, what type of source is writing the answer: social discussion, retailer pages, publisher reviews, or a specialist competitor?
- If one platform’s citation mix shifted sharply next quarter, how much of your visibility would move with it?
For some beauty and consumer tech brands, the answers point straight at social communities, and we build multi-quarter Reddit and social listening programs around that reality. For others, they point somewhere else entirely: opinion-rich formats like YouTube, influencer reviews, and detailed comparison content that already carry more weight in the AI surfaces their customers actually use.
How Tinuiti turns citation data into an AI visibility roadmap
The biggest risk I see right now isn’t that brands are ignoring AI search. It’s that they’re reacting to headlines instead of running a disciplined program. Our AI visibility programs plug into the same Love Growth, Hate Waste. framework that underpins our media planning, Tinuiti Audience Planning System (T.A.P.S.) process, and Bliss Point by Tinuiti measurement suite. When we build an AI visibility roadmap with clients, we usually start with three concrete steps:
1. Quantify where you stand today.
Using Profound and Tinuiti’s own AI visibility dashboards, we benchmark how often your brand is cited across key prompts, which domains carry your story (brand, retailer, publisher, social), and how that compares to your competitive set. That gives us a measurable baseline we can connect to business outcomes using Bliss Point tools like Rapid MMM and incrementality, rather than treating citations as vanity metrics.
2. Cluster platforms around real customer journeys.
Rather than chasing every AI product, we map platforms to audiences and roles in the funnel: for example, Google’s AI surfaces in high-reach discovery and refinement layers, ChatGPT and Perplexity serve as deeper research companions, and retailer or marketplace experiences serve as commerce adjacencies. That “next best customer” mindset mirrors how Tinuiti plans media more broadly – go where the incremental audience and influence actually are, not just where the headlines are.
3. Design tests that link AI visibility to measurable growth.
From there, we build focused experiments: improving coverage for a specific cluster of prompts, shifting content mix toward sources a given assistant leans on (e.g., YouTube, credible reviews, or retailer PDPs), and then reading the impact through both citation share and downstream KPIs like branded search, retailer conversion, and assisted revenue. Because Bliss Point centralizes measurement across channels, we can see whether a lift in AI citations is actually contributing to your overall media efficiency and revenue, not just your share of screenshots.
This approach keeps AI SEO grounded in the same disciplined, test-and-learn operating model we use for media and creative: start with data, tie every optimization back to the customer journey, and use unified measurement to decide what deserves more investment and what doesn’t.
Jen Cornwell
Senior Director of Innovation & Growth, Tinuiti
Jen Cornwell is Sr. Director of AI SEO Innovation at Tinuiti, helping brands navigate conversational search and large language models to grow organic visibility. With 10+ years of experience, she has led large SEO teams and been featured in outlets such as Search Engine Land, Ad Age, and Digiday. Originally from snowy Syracuse, NY, Jen now lives in San Diego, CA with her husband.
Want to learn more?
The findings in this article only scratch the surface of what we’re seeing across seven AI platforms, nine verticals, and thousands of prompts. If you want to dig into the charts by category, benchmark your brand, or share the data with your team, download the full Q3 2026 AI Citation Trends Report.
