Key takeaways
- Product pages overtook listicles as the most-cited content format in Google AI Overviews in late July 2026, ending the month at 17.9% of citations versus 16.2% for listicles, per Promptwatch Data.
- How-to content has held steady at roughly 15% of citations all year, while video has more than doubled since January (2.7% to around 6%).
- Google AI Mode's domain-level data shows google.com itself taking 7.31% of citations in June 2026, more than YouTube and Reddit combined, and a strong preference for long-form LinkedIn articles over feed posts.
- Reddit's citation share is declining across all Google AI surfaces, gradually rather than the abrupt collapse seen in ChatGPT.
- Markdown files account for just 0.05% of AI search citations. HTML pages win. The "publish .md files for AI" advice is aimed at coding agents, not search.
One caveat before we dig in: Google tracks two separate AI surfaces, AI Overviews (the summary that appears at the top of regular search results) and AI Mode (the full conversational, Gemini-powered search experience). Promptwatch classifies content types (listicle, product page, how-to, and so on) for AI Overviews and ChatGPT, while AI Mode gets domain-level and page-type tracking. The two surfaces behave similarly, but they are not identical, so this guide uses the AI Overviews format data as the closest first-party proxy and pulls AI Mode's own numbers where they exist. I'll flag which is which throughout.
The headline shift: product pages are winning
For most of 2026, the story of AI Overviews citations was a listicle story. In Q1 2026, listicles took 26.1% of all classified citations in Promptwatch's AI Overviews data, built on a sample of more than 20 million classified citations from January through early April. Product pages sat at 9.7%. It wasn't close.
By July, it was close. And then it flipped.
July's daily breakdown shows product pages climbing all month and overtaking listicles as the single most-cited format from July 28 onward, finishing at 17.9% versus 16.2%. Product pages have nearly doubled since January. Listicles have fallen from ~26% to 18% over the same stretch.
| Format | Q1 2026 share | July 2026 share | Trend |
|---|---|---|---|
| Listicle | 26.1% | 16.2% (18.0% monthly avg) | Falling steadily |
| Product page | 9.7% | 17.9% | Nearly doubled, now #1 |
| How-to | 12.1% | ~15.1% | Stable, slightly up |
| News article | 15.3% | ~13.5% | Slipping |
| Video | 3.4% | ~6.3% | More than doubled |
| Landing page | 6.2% | ~4.7% | Declining |
| Social post | 4.1% | ~5.1% | Up modestly |
| Comparison | 2.7% | ~3.6% | Flat |
I find this shift genuinely interesting, because it runs against a lot of the early GEO advice. The 2025-era playbook said AI search rewards third-party listicles, review roundups, and Reddit threads, and that your own commercial pages barely register. That was true then. It's becoming less true now. Google's AI surfaces are increasingly willing to cite brand-owned product pages directly, which changes where you should invest content effort.
And it's not just a Google quirk. ChatGPT's July data shows product pages at 32.8% of citations, nearly a third of everything ChatGPT Search cites, roughly double their March share. Whatever is driving this, it's happening across AI search broadly.
What AI Mode's own data says
AI Mode doesn't have a published content-type breakdown the way AI Overviews does, but its domain-level data tells a compatible story with a few twists of its own.
In June 2026, google.com itself led all domains at 7.31% of AI Mode citations, up from 4.24% in May, a 72% jump on top of an earlier sixfold increase. Add YouTube at 2.88% and Google-owned properties captured over 10% of the entire citation mix. Reddit sat third at 2.52%. Everything below the top three fell under 1%.
That self-preference trend is uncomfortable if you're a publisher competing with Google's own properties, but it's worth knowing. It also wasn't always this way: in March 2026, YouTube led at 3.30%, google.com had slipped to 0.58%, and six of the top seven cited domains were video, social, or community platforms. Forbes, representing traditional publishing, managed 0.17%. AI Mode's retrieval preferences swing hard month to month, which is exactly why point-in-time snapshots of "what AI cites" age badly.
The other striking AI Mode finding is about LinkedIn. Promptwatch's LinkedIn page-type report shows AI Mode is the most article-hungry engine it tracks: long-form Pulse articles take 44.85% of its LinkedIn citations, with posts a distant second at 27.37% and company pages at 10.71%. ChatGPT, by contrast, treats LinkedIn mostly as a directory, favoring company pages and homepages over articles. If you're publishing B2B thought leadership, the long-form article with a question-style headline and a direct answer up front is the format AI Mode reaches for.
Video is the quiet overperformer
Video deserves its own section because the numbers moved so much and because almost nobody optimized for it.
Video citations in AI Overviews roughly doubled over Q1 (2.7% to 5.4%) and climbed to around 6.3% by end of July. Compare that with ChatGPT, where video sits at roughly 0.1% of citations, essentially zero. Google's AI surfaces reward video; ChatGPT mostly ignores it. If your video strategy was built on ChatGPT assumptions, you've been leaving Google traffic on the table.
The more surprising part is which videos get cited. Promptwatch's YouTube citation research found that mid-tier videos dominate: the 10K–100K view bucket was the largest cited group at 36%, about 80% of cited videos had under 100K views, and only ~7% had passed a million views. Same pattern with channels: 80% of citations went to channels under 100K subscribers. Engagement signals like likes barely factor in.
The practical takeaway: retrieval matches content to the query, not popularity. Put the exact question in the video title, summarize the answer in the first lines of the description, and add chapters. A modest video that directly answers a question beats a viral one that doesn't.
Reddit is fading, gradually
Reddit had a monster run in AI search through 2025, and plenty of GEO advice still treats it as the cheat code. The 2026 data says that's fading.
In Google AI Mode specifically, Promptwatch's Reddit tracking shows Reddit's citation share falling from a 2.22% average in the first week of the July 7 to August 17, 2026 window to 1.54% in the final week, a 30.5% relative decline. AI Overviews declined more gently, from 2.37% to 2.10%, an 11.3% relative drop.
The contrast with ChatGPT is stark: ChatGPT's Reddit share collapsed from roughly 4% to 0.5% overnight on August 14, 2026, an 86% drop in a single day. Google's decline is a slope, not a cliff. Reddit still matters, especially for community-driven queries, but building your entire AI visibility strategy on Reddit mentions looks increasingly like chasing last year's trend.
How-tos, news, and the formats that held steady
How-to content is the most boring and most reliable performer in the dataset. It sat at 12.1% in Q1 and roughly 15% in July, barely wavering. If you produce instructional content, AI Overviews and AI Mode will keep citing it as long as it genuinely answers the question in the title. Step-by-step structure with clear headings is the whole game.
News articles slipped from 15.3% to around 13.5%, which makes sense as AI Mode leans more on evergreen and commercial content for the conversational queries it handles. News still wins for anything time-sensitive, and freshness signals like current-year references help, but you can't build an AI citation strategy on news alone.
Comparisons held flat at 3-4%. Small share, but comparison pages are high-intent, and they're one of the few formats where a mid-sized brand can realistically beat a publisher to the citation slot.
A myth worth killing: markdown files
Every few weeks someone tells me you should publish .md files so AI search engines can cite you. Promptwatch's format research puts that to bed: markdown files account for 0.05% of all AI search citations across ChatGPT, Claude, Perplexity, and Google AI Overviews. HTML takes 99.94%.
Markdown and llms.txt have a real purpose, but it's for AI coding agents like Claude Code, not consumer AI search. Don't restructure your site around it.
How many citation slots are you competing for?
Worth knowing before you set expectations. Promptwatch's sources-per-response data shows Google AI Overviews citing roughly 10 sources per answer, about double ChatGPT's ~5, and that count has stayed steady. Ten slots per answer makes Google's AI surfaces the more forgiving place for mid-authority domains to break in. AI Mode runs on query fan-out, decomposing one query into 12-15 sub-queries on average (and 50+ for complex ones), which multiplies the retrieval opportunities but also means your page needs to answer a cluster of related sub-questions, not one narrow query.
What to actually do with this
Here's how I'd translate the data into a content plan for the rest of 2026:
- Rebuild product pages as answer pages. Product pages are the fastest-growing cited format on both Google and ChatGPT. That means specs, pricing context, use cases, and FAQ sections directly on the product page, not gated behind a separate "learn more" hub.
- Keep making listicles, but make them honest ones. The share fell from 26% to 18%, which is a decline, not a disappearance. Listicles are still a top-two format. The ones that get cited tend to have real comparison criteria rather than affiliate padding.
- Invest in video with retrieval-friendly metadata. Doubled citation share, mid-tier videos winning, and ChatGPT's near-total ignore of video means this is a Google-specific advantage most competitors haven't noticed yet.
- Write long-form LinkedIn articles for B2B queries. AI Mode cites Pulse articles at 4x the rate of company pages. Question-style headline, direct answer in the first paragraph.
- Don't abandon Reddit, but stop over-indexing on it. Gradual decline across Google surfaces, cliff-edge decline in ChatGPT.
- Structure everything for chunk-level retrieval. Short paragraphs, clear heading hierarchy, FAQ and Article schema, and pages that answer sub-question clusters together. Fan-out retrieval feeds passages into answers, not whole pages.
Tracking your own citations
All of this is directional until you see what AI Mode actually cites for your prompts. A platform like Promptwatch tracks citation types, domain share, and page-level citation data across AI Mode, AI Overviews, ChatGPT, and the other major engines, so you can watch these format shifts play out for your own queries rather than relying on industry averages.
If you want to compare the broader field of tracking and optimization tools, the AI rank tracking tools directory at ai-rank-tools.com covers most of the options, and the GEO software directory at bestgeosoftware.com lists full optimization platforms.
The bottom line
The 2026 data points one direction: Google's AI surfaces are warming to brand-owned, commercially direct content. Product pages, how-tos, and video are the formats gaining share; listicles, landing pages, and Reddit are the ones giving it up. If your AI visibility strategy was built on 2025 assumptions, the product page section of your site is where I'd start rewriting.