The Most-Cited Sources in Google's AI Overviews (We Checked 24 Searches)

By Minel Gunesoglu, founder of Is My Brand in AI · Original research, June 10, 2026 · Updated August 8, 2026

TL;DR: We ran 24 'best [category]' searches through Google's AI Overview and logged every cited source. Reddit was most-cited — 13 of the 21 Overviews that triggered (62%), often the top source. Brand-owned pages were rarely cited; AI answers need third-party validation, not just your own page.

Everyone in marketing now repeats the same advice: "get cited by AI." Almost nobody shows you who actually gets cited, and when a page does put a number on it, the number usually arrives with no way to check it: across thirteen pages ranking for AI-visibility queries, the median one linked a source for about a third of its own figures, which is what we counted in The AI Search Evidence Index. So instead of guessing, we measured it. We took 24 ordinary "best [category]" searches, ran each one through Google, opened the AI Overview, and wrote down every website it listed as a source. This page is the raw result and what it means for your brand.

One number frames the whole thing: of the 21 searches that produced an AI Overview, Reddit was cited in 13 of them. No brand's own website came close. If you have been pouring effort into your product pages and wondering why AI never mentions you, this is why. The citation almost never points at the seller. It points at the places people talk about the seller.

How we ran the test

We kept the method deliberately simple and reproducible, so you can repeat it yourself.

• Engine: Google AI Overview, the source panel it labels "N sites."
• Queries: 24 buyer-intent "best [category]" searches, spread on purpose across four very different areas so the finding would not be specific to one niche: business software (project management, CRM, email marketing, accounting, VPN, web hosting, password manager, AI writing, help desk, HR), consumer products (running shoes, office chair, robot vacuum, mattress, electric toothbrush, coffee maker), finance (high-yield savings, travel credit card, budgeting app), and lifestyle (protein powder, meal kits, standing desk, note-taking app, web conferencing).
• What we logged: for each search, every distinct domain in the AI Overview source panel, the total source count, and the type of site.
• When: a single snapshot on June 10, 2026, US English results.
• The data: published in full as CSV and JSON — one row per query, with the cited domains, so every count in the table below can be recounted from the rows.

The point of mixing protein powder with CRM software is coverage. If we only tested software, "Reddit dominates" might be a quirk of one industry. Seeing the same pattern hold from supplements to enterprise tools is what makes it a pattern.

Finding 1: Most "best" searches get an AI Overview, but not all

On the day we ran them, most of the 24 searches returned an AI Overview, and a few pure physical-retail product searches returned a Shopping grid with price and size filters instead of an AI answer. Which queries trigger an AI Overview is not fixed: when we re-checked the next day, a product search that had shown only Shopping was already showing an AI Overview. So treat this split as a snapshot, not a rule.

The practical read still holds. If you sell software, services, or finance products, an AI Overview is usually the first thing a buyer sees. If you sell a physical retail product, it is less predictable, and part of the battle is still in Shopping and traditional reviews. When an AI Overview did appear, it drew on about 7 sources (ranging from 5 to 12).

Finding 2: The most-cited sources

Counting how many of the 21 AI Overviews cited each site:

Below the top six sat a long tail of two-time and one-time sources: Bankrate, CNET, Amazon, Salesforce, Wirecutter, Consumer Reports, Healthline, Serious Eats, HubSpot, Hostinger, Experian, and many niche blogs. The shape is clear. A small set of forums and big review brands appear again and again, and then a wide scatter of everyone else.

What is striking is who is missing from the top: the brands being searched for. When someone searches "best CRM," Google's AI did not build its answer from Salesforce's or HubSpot's homepage. It built it from Reddit threads, Forbes, and YouTube reviews, and only then name-checked the products.

Finding 3: Four types of source, and forums lead

Sort the citations by type and a structure appears:

• Forums and community posts (Reddit, plus a Clark Howard community thread): present in 14 of 21 AI Overviews. This is the single most common ingredient.
• Independent review media (Forbes, PCMag, CNET, NerdWallet, Bankrate, Wirecutter, Consumer Reports, Serious Eats): the backbone. Nearly every AI Overview included at least one.
• Vendor and brand sites (Salesforce, Zendesk, HubSpot, Jasper, Slack, Hostinger): they do appear, but mostly in business-software answers, and almost never as the only source.
• Video (YouTube): just under a third of answers (6 of 21).

The headline for a brand owner is uncomfortable but useful: the two source types that dominate are the two you do not own. You cannot publish your way to a forum consensus, and you cannot self-publish your way into a Forbes roundup. Both have to be earned.

Finding 4: The mix changes by category

Reddit is not everywhere. Where it shows up, and where it does not, follows a logic worth understanding before you copy anyone's "AI playbook."

• Consumer and general queries (project management, CRM, VPN, web hosting, password managers, coffee makers, note apps): Reddit is usually the top or near-top source, alongside big review media. These categories have huge public discussion, and the AI leans on it.
• B2B software niches (accounting, AI writing, help desk, web conferencing): Reddit was usually absent. With less open forum chatter, the AI fell back on vendor blogs plus PCMag-style roundups. Here, getting into the listicles and publishing genuinely useful vendor content matters more.
• Finance, especially credit cards: the AI cited only authoritative finance media (NerdWallet, Bankrate, Experian, Credit Karma, The Points Guy, CNN). No forum at all. The more a topic touches money or health, the more Google narrows to recognized authorities.
• Physical retail products: the least predictable. A pure-retail search may return a Shopping grid instead of an AI Overview, and whether it does shifts from day to day.

So "be on Reddit" is the right answer for a consumer app and the wrong first move for a credit-card issuer. Match the tactic to where your category's citations actually come from.

What this means for your brand

Pull the four findings together and the strategy writes itself.

1. Your own website is rarely the citation. Owned content alone will not put you in AI answers. It is necessary, not sufficient.
2. For consumer and general categories, Reddit is the highest-leverage surface. It is the most-cited domain on the open web for these answers. Genuine, non-spammy presence in the relevant subreddits is one of the most reliable ways to earn citations here, and we walk through this in how to improve brand visibility in AI search.
3. For everyone, earn third-party coverage. Getting included in the "best of" roundups that PCMag, Forbes, Zapier, and niche reviewers publish is the backbone of citations in every vertical.
4. Match the channel to your category. B2B software: vendor content plus roundups. Finance and health: build the kind of authority Google trusts. Consumer: forums and reviews.
5. You cannot manage what you do not measure. Before you act, check where your brand actually stands across engines. Our guide to tracking brand mentions shows a free way to do it by hand, and when the hand version stops fitting in your week, the best AI search visibility tools comparison covers what the paid trackers watch and what they miss.

If you want the surface-by-surface version of this, see AI Overview optimization and how to get cited by ChatGPT. The principle is the same everywhere AI answers questions: it cites the sources it trusts, and you have to earn your way onto that list.

Honest limitations

This is a snapshot, not a law of physics, and we would rather tell you the caveats than oversell the number.

• It is a single engine (Google AI Overview), a single day (June 10, 2026), and a defined 24-query sample. A different day or account can return a different mix.
• AI Overviews are personalized and volatile. Run the same search on another day and both the sources and whether an AI Overview appears at all can change; we saw exactly that on a re-check. The aggregate pattern (Reddit and review media dominate) is far more stable than any single query.
• We identified most domains from the AI Overview source titles. A minority of long-tail sources were classified by type but not resolved to an exact domain. This does not move the most-cited ranking, which is driven by the sites that recurred across many searches.
• Searches were run in English with US settings. Results in other markets and languages will differ.

We did widen this. We ran the same buyer questions against ChatGPT, Perplexity, Gemini and Google's AI Overview to see whether they cite the same sources — and they barely overlap. Each engine pulls from a different world, and Reddit was the one near-constant that bridged them. The methodology above is still the version anyone can reproduce.

Update, August 2026: this list has a half-life

There is one limitation above that we can now put a number on, because we re-ran the cross-engine study seven weeks after the first round and compared the two source lists question by question.

Of the sources cited in June, 48% were gone seven weeks later. That measurement is from Perplexity, the one engine we had full coverage for in both rounds, so treat it as indicative for AI answers generally rather than as a measurement of Google's AI Overview specifically. But the direction is not in doubt, and it changes how you should read the rankings on this page.

It cuts both ways, and the second half is the more useful one: no question lost its sources entirely. Roughly half persisted, and the survivors were consistently the same kind of site — deep, topic-specific resources and large reference destinations rather than whatever happened to rank that week.

So: the types of source on this page (Reddit, review media, forums) have held up across every study we have run. The specific domains in any single snapshot should be read as a photograph, not a league table. If you earned a citation in the spring and have not checked since, it is worth checking. The full comparison is in the cross-engine study.

Method, query list, and per-search results were logged during collection. If you are a journalist or researcher and want the underlying notes, reach us through the contact page.

Cite this study

This is original, hand-run research — you are welcome to cite or share it (CC BY 4.0). Suggested attribution:

Is My Brand in AI (2026). What sources do AI Overviews cite? A 24-query Google AI Overview study. https://ismybrandinai.com/most-cited-sources-in-ai-overviews

Google AI Overview, 24-query sample, 10 June 2026. Full method on our methodology page.

The data:

• ai-overview-cited-domains-2026-06-10.csv — 24 rows, one per query
• ai-overview-cited-domains-2026-06-10.json — same rows plus the method and the known limits

Two caveats the rows make visible: the source panel groups several links under one entry (a "Zapier +3" counts once), and a few panel entries could not be resolved to a domain and are left out of the domain list.