Key takeaways
- Government sites are overrepresented in AI Overviews specifically: Pew Research found .gov domains make up 6% of sources linked in AI summaries versus 2% in standard search results, a 3x boost in the AI layer.
- Wikipedia's dominance is real but engine-specific. ChatGPT treats it as a top reference source; Google AI Overviews cite it less than you'd think and surround it with YouTube, Reddit, and Quora.
- Ranking #1 organically does not guarantee citation. BrightEdge found Wikipedia makes it into AI Overviews on fewer than half the queries where it holds an organic ranking, and roughly a third of exclusions still had Wikipedia at position #1.
- Smaller brands cannot out-Wikipedia Wikipedia. The realistic plays are Wikidata, narrow niche queries, product pages (now the most-cited content format), and structured, citable content.
- Visibility does not transfer between engines. Track each one separately, because a win in ChatGPT tells you almost nothing about Google AI Overviews.
The pattern, in numbers
If you've spent any time clicking the links inside a Google AI Overview, you've probably noticed the same handful of domains over and over. Wikipedia. YouTube. Reddit. And, depending on the query, a suspicious number of .gov sites.
The numbers back this up, though some of them surprised me.
Pew Research Center analyzed 12,593 Google searches with AI summaries from US adults in March 2025 and found that .gov websites made up 6% of sources linked in AI Overviews, versus just 2% in standard search results. Government sites get a 3x boost in the AI layer specifically. Wikipedia, YouTube, and Reddit together accounted for 15% of AI-summary sources. News sites, by contrast, got no AI-layer boost at all: 5% of sources in both AI summaries and standard results.
On the ChatGPT side, the institutional skew is even stronger. Profound's analysis of 10 million citations found Wikipedia is ChatGPT's single most-cited source at 7.8% of total citations. And in healthcare queries, BrightEdge found that every one of ChatGPT's top five most-cited domains is a government agency or nonprofit hospital system: nih.gov (43% citation presence), medlineplus.gov (33%), mayoclinic.org (27%), cdc.gov (18%), and clevelandclinic.org (17%). YouTube does not appear in ChatGPT's healthcare citations at all.
Here's the part that surprised me, though. Promptwatch's own citation share data for Google AI Overviews in June 2026 shows Wikipedia's raw share is small, at 0.15% of all citations, up 50% month-over-month but from a tiny base. The real concentration in AI Overviews is YouTube (4.16%) and google.com itself (2.30%), which grew sixfold in five months. Third-party studies like Surfer's AI citation report, which analyzed 36 million AI Overviews, report much larger Wikipedia shares (~18.4%), but that reflects a different methodology and a broader definition of what counts as a citation. The honest read: Wikipedia is structurally trusted everywhere, but in Google's AI layer it's one voice in a noisy, platform-heavy room, while in ChatGPT it's a cornerstone.
Why Wikipedia dominates
Wikipedia's position in AI search isn't an accident, and it isn't just about domain authority either. Three structural factors do the work:
Training-data weight. Large language models were trained on Wikipedia and had it weighted as high-quality factual data during pretraining. The models literally absorbed its phrasing, its structure, and its factual claims. When an AI system retrieves and summarizes, Wikipedia content is the path of least resistance.
Retrieval friendliness. Wikipedia articles rank near the top of retrieval pools because of extreme domain authority, but also because of clean structural formatting: descriptive headings, cited claims, enumerated facts, stable URLs, encyclopedic tone. A retrieval system can parse a Wikipedia article cheaply and cite it confidently. Citing Wikipedia is low-cost for the model and high-credibility for the user.
Neutrality. Wikipedia's editorial voice is deliberately neutral and sourced. For a system that needs to summarize contested topics without taking sides, that's exactly the kind of source it wants.
There's a nuance worth dwelling on, because it changes how you should read all this. BrightEdge analyzed tens of thousands of keywords where Wikipedia holds an organic ranking and an AI Overview is present, and Wikipedia makes it into the AI Overview on fewer than half of those queries. When it is cited, 75% of the time it holds a top-3 organic ranking. But roughly a third of the exclusion cases still had Wikipedia sitting at position #1.
That's the most important sentence in this guide for smaller brands. Ranking reflects topical authority. AI Overview citation reflects whether the content format can directly serve what the query needs right now. Even the most authoritative site on the planet gets skipped when the format doesn't fit the question.
Why government sites dominate
The .gov story is simpler in some ways and more interesting in others.
Google's quality frameworks explicitly hold "Your Money or Your Life" (YMYL) topics, health and finance chief among them, to the highest standard, tied directly to E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) in its quality rater guidelines. When an AI Overview needs to answer "what are the symptoms of iron deficiency," a random health blog and nih.gov are not interchangeable. The system reaches for the source with the least reputational risk.
Government sites also share Wikipedia's structural advantages: stable URLs, dense factual content, clean formatting, no commercial agenda. A CDC page is cheap to cite and hard to blame.
What's genuinely interesting is how differently the two engines handle the same category. In healthcare, BrightEdge found Google AI Overviews moving toward YouTube (citation presence up to roughly 1.58x its June 2026 baseline by late July) while nih.gov drifted below its baseline. ChatGPT went the opposite direction, all institutions, no video. BrightEdge's conclusion was blunt: the two engines are building different definitions of what counts as an authoritative source.
The dominance is not uniform, and that matters
It's tempting to read all this as "AI search loves big, old, institutional websites, game over." That reading is wrong in two important ways.
First, the citation mix is volatile in ways that should make everyone cautious. Reddit's share of ChatGPT citations collapsed roughly 90% almost overnight in September 2025 after Reddit changed platform access, and Promptwatch's data shows reddit.com's share of ChatGPT Search citations fell from roughly 4% to 0.5% on August 14, 2026. Platform-dependent citation share can evaporate. Wikipedia and .gov domains, by contrast, are structurally stable, which is precisely why the engines lean on them. But it also means the "room" you're competing in can change under your feet.
Second, and more hopeful: the content formats AI Overviews cite are shifting toward things smaller brands actually produce. Promptwatch's citation-type data for July 2026 shows product pages overtook listicles as the most-cited format in Google AI Overviews from July 28 onward, ending the month at 17.9% versus 16.2% for listicles. The same pattern shows up in ChatGPT. AI search as a whole is warming to brand-owned commercial pages.
There's also a pure arithmetic argument for smaller brands. Google AI Overviews cite roughly 10 sources per answer, about double ChatGPT's ~5, and that count is steady over time, per Promptwatch's sources-per-response analysis. Ten slots is a lot of room. It makes AI Overviews the more forgiving engine to break into for mid-authority domains.
What this means for smaller brands
Let me be honest about the hard part first. You cannot compete with Wikipedia on Wikipedia's terms, and you shouldn't try.
Wikipedia's notability bar requires significant coverage in three to five reliable, independent sources with no financial or promotional relationship to the subject. Most brands don't meet it. That's the bar working as designed, not a failure on your part. And attempting to self-edit your way in tends to backfire: accounts that create articles about only one company or executive are an obvious conflict-of-interest signal to Wikipedia's editors, and paid editing usually gets flagged and deleted. Some mid-market B2B brands spend $50,000 to $200,000 over 6 to 12 months on earned media just to build the coverage base a sustainable Wikipedia article requires.
So what can you actually do? Several things, in rough order of effort.
1. Start with Wikidata, not Wikipedia
Wikidata has a substantially lower notability threshold, can be edited directly by brand representatives with fewer conflict-of-interest complications, and can realistically be done in an afternoon. It also feeds Google's Knowledge Graph, which powers AI Overview structured-fact retrieval. If your brand has no Wikidata entry, that's the cheapest structural-authority win available.
2. Win the queries where format fit beats authority
Remember the BrightEdge finding: even Wikipedia at #1 gets excluded when the format doesn't serve the query. The flip side is that on narrow, specific questions, a page that directly and completely answers the question can beat a higher-authority domain that answers it obliquely. Practitioner consensus is forming around this: AI Overviews appear to weight complete meaning and answer structure over raw domain authority for narrow queries. It's vendor-sourced reasoning rather than hard data, but it matches what the Wikipedia exclusion data shows.
3. Build product pages that answer questions
Product pages are now the most-cited format in AI Overviews. If your product pages are thin marketing brochures, you're leaving the fastest-growing citation category on the table. Add specifications, use cases, comparisons, pricing logic, and the questions your customers actually ask. Make the page the complete answer, not a teaser.
4. Structure your content like it will be parsed by a machine, because it will
Descriptive headings. Claims backed by sources. Stable URLs. Clean formatting. Defined entities. This is the unglamorous part of GEO, and it's the part that most directly imitates what Wikipedia and .gov sites do structurally. Schema markup helps here too; some vendor analyses report large selection boosts for AI Overview inclusion, though treat those specific numbers as directional rather than gospel.
5. Invest in earned media, and be patient
Aggregated practitioner data (reported by Ziptie, drawing on multiple vendor studies) suggests the large majority of AI citations trace back to earned, third-party coverage rather than owned content, and that each placement's citation effect compounds for 18 to 24 months. That's a PR investment with an unusually long tail. It's also the only realistic path to ever meeting Wikipedia's notability bar, so the two goals align.
6. Track each engine separately
This is the one that ties everything together. Source-level wins do not transfer between engines. ChatGPT's Wikipedia neighborhood is institutional (Britannica appears in 43% of ChatGPT responses that also cite Wikipedia, per BrightEdge). Google AI Overviews' Wikipedia neighborhood is social (YouTube at 13% co-citation, Reddit at 9%, Quora at 6%). A strategy that wins in one room can be irrelevant in the other.
A platform like Promptwatch handles this well: it tracks your citations, citation trends, and share of voice across ChatGPT, Google AI Overviews, AI Mode, Perplexity, Claude, and the rest, per engine, so you can see which room you're actually competing in and where the gaps are.
How the engines differ, at a glance
| Dimension | Google AI Overviews | ChatGPT |
|---|---|---|
| Wikipedia's role | Present but small share; surrounded by social platforms | Top-cited source overall (~7.8% of citations per Profound) |
| Wikipedia's co-citation neighborhood | YouTube, Reddit, Quora, IMDb | Britannica, Merriam-Webster, institutional references |
| Healthcare citations | Skewing toward YouTube; nih.gov drifting down | All top 5 are government agencies or nonprofit hospitals |
| Sources per answer | ~10, steady | ~5, about half of Google's |
| Most-cited content format (July 2026) | Product pages (17.9%), overtaking listicles | Product pages lead here too |
| .gov representation | 6% of AI-summary sources vs 2% in standard results (Pew) | Heavy in YMYL categories, especially health |
Tools worth using for this work
You'll want three categories of tooling: visibility tracking, content optimization, and competitive citation research.
| Category | What it does | Examples |
|---|---|---|
| AI visibility tracking | Shows which prompts cite you, which pages get cited, and where gaps are, per engine | Promptwatch, Profound, ScrunchAI |
| Content optimization | Helps structure pages for readability and completeness | Surfer SEO, Clearscope, Frase |
| Citation and competitor research | Shows who's getting cited in your niche and why | Ahrefs, Semrush, BrightEdge |
If you're evaluating tracking tools specifically, the AI rank tracking tools directory at ai-rank-tools.com has a solid categorized list, and bestgeosoftware.com covers the broader GEO platform market.
The bottom line for small brands
Wikipedia and government sites dominate AI citations for structural reasons: training-data weight, retrieval-friendly formatting, neutrality, and reputational safety. You cannot replicate those advantages, and chasing them directly (especially via Wikipedia self-editing) will waste your money and possibly damage your reputation.
But the data points somewhere more workable. AI Overviews cite ten sources per answer. Product pages are the fastest-growing citation format. Even Wikipedia gets excluded when the format doesn't fit the query. And the engines disagree with each other enough that no single source dominates everywhere, which means no single competitor dominates everywhere either.
The realistic strategy for a smaller brand: fix your Wikidata entry this week, rebuild your top product pages as complete answers this quarter, structure everything you publish for machine parsing, invest in earned media with an 18-month horizon, and measure your visibility per engine so you know which room you're actually in. That's not a consolation prize. That's how you win the slots the giants leave open.