For the past month we have been pointing a measurement tool at a question most local business owners have started asking out loud: when someone asks ChatGPT for a recommendation, where does the answer come from?
We had a selfish reason to look. Our own site appears in none of those answers, and we wanted to know what would actually change that. What we found reorganized our plans, so we are publishing the numbers rather than keeping them.
The short version: YouTube and Reddit supplied more than half of every source these assistants drew on. Company websites — the thing most businesses spend their marketing budget on — barely registered.
What We Measured, Exactly
We tracked six buying questions a local business owner might realistically type:
- AI receptionist for small business
- AI answering service
- virtual receptionist service
- AI appointment scheduling
- missed call text back service
- AI chat agent for website
Each one ran weekly against two platforms — ChatGPT and Google's AI Overviews — for four consecutive weeks, from 9 to 27 July 2026. Six questions, two platforms, four captures: 48 observations. For each one we recorded which brands the answer named and which source domains it drew from.
Two things worth stating plainly before the numbers. First, this is a small, single-category sample — six questions in one industry over one month. It is enough to show a strong pattern and not enough to be a law. Second, the source data comes through a commercial data broker that samples these platforms on a schedule, not from scraping the assistants directly. The long tail carries some noise: a cluster of medical reference sites turns up against a single keyword and looks like topical bleed rather than a real signal. We have left it in the totals rather than quietly trimming the inconvenient rows, and we would rather you know that than not.
The Headline: Two Platforms, Two Diets
Across all 48 observations, these were the domains the assistants pulled from most:
| Source | Share of all citations | Appeared for |
|---|---|---|
| youtube.com | 33.1% | All 6 questions |
| reddit.com | 18.8% | All 6 questions |
| en.wikipedia.org | 4.4% | 5 of 6 questions |
| quora.com | 2.1% | 2 of 6 questions |
| linkedin.com | 1.7% | 1 of 6 questions |
YouTube and Reddit together account for roughly 52% of everything cited. Neither is a marketing site. Both are places where people talk to each other.
But the aggregate hides the more useful finding. The two platforms do not read the same internet at all.
ChatGPT reads text
ChatGPT's most-cited sources, in order: Reddit, Wikipedia, Reuters, TechRadar, Forbes, Business Insider, arXiv, Trustpilot.
That is a diet of community discussion, reference material, established press and review platforms. It is what you would expect from a system trained to prefer sources that look authoritative and are heavy on text.
Google's AI Overviews watch video
Google's list is dramatically different: YouTube first by an enormous margin, then Reddit, then topical authority sites, then Wikipedia.
This makes sense in hindsight — Google owns YouTube, has a decade of investment in understanding what is inside a video, and surfaces it accordingly. But it is not what most businesses assume, and it changes what "doing content" should mean.
What This Actually Means If You Run a Local Business
The instinct when someone says "we need to show up in AI" is to write more pages. Our data says that instinct is mostly wrong, or at least badly incomplete.
Here is the uncomfortable part. We are a company that sells AI automation. We have a technically clean website — structured data, fast pages, every AI crawler explicitly welcomed, a real blog. Across all 48 observations, we were cited zero times and mentioned zero times. Not a low number. Zero.
A well-built website is table stakes. It is what lets a search engine understand what you do. It is not what gets you recommended by an assistant, because the assistant is largely reading what other people published about you.
That reframes the work into three practical shifts:
1. Presence beats publishing
If Reddit is a fifth of what ChatGPT reads in your category, being genuinely present in the communities where your customers ask questions is worth more than another article on your own domain. Genuine is the operative word — most communities will remove you quickly for showing up to advertise, and they are right to.
2. Video is not optional for Google
If a third of Google's cited sources are YouTube videos, then a business with no video is invisible to a large slice of AI-mediated search. For a local service business this does not mean production budgets. It means answering the questions customers actually ask, on camera, findable.
3. Third-party listings do double duty
Review platforms and comparison sites appeared repeatedly in the results. A profile on one of those is simultaneously a place an assistant can cite, a link back to your site, and a signal that helps AI tell your business apart from others with similar names. Few marketing actions pay three ways at once.
A Note on Which Brands Got Named
We also recorded which companies the answers mentioned. The most-named entities were not specialist vendors at all — they were Microsoft, Amazon, Google and OpenAI, with category-specific tools appearing well below them.
Part of that is our extraction counting any recognized company, including the platform names an assistant mentions in passing. But part of it is real and worth sitting with: when a general-purpose assistant is asked about business software, it reaches for the largest, most-written-about names first. If you are small, you are not competing for attention against the other small vendors. You are competing against the default gravity of the biggest brands in technology.
The way out of that is specificity. An assistant asked a generic question returns generic giants. Asked "who does AI phone answering for dental practices in Las Vegas," it needs sources that talk about exactly that — which is a much shorter list, and a list a small business can realistically get onto.
What We're Doing About It
We are following our own findings, and we will report back on whether they work:
- Getting listed on the review and comparison platforms that appeared in the results.
- Publishing video, because a third of Google's citations are YouTube and we have none.
- Participating in the communities where these questions get asked, as operators rather than advertisers.
- Continuing to measure weekly, so we can tell you whether any of it moved the number.
We publish our AI-visibility numbers because measuring this is the service we sell, and a scoreboard nobody can check is not worth much. Ours currently reads zero. We would rather show you that honestly than describe a result we have not earned.
Frequently Asked Questions
Where do AI assistants get their answers about business software?
In our measurement, YouTube supplied 33% of cited sources and Reddit 19% — more than half between them. Vendor websites were a small minority. The platforms differ sharply: ChatGPT leaned on Reddit, Wikipedia and established press; Google's AI Overviews leaned heavily on YouTube.
Does publishing blog posts get you cited by AI?
Not on its own. Across 48 observations the cited sources were overwhelmingly third-party — community discussion, video, reference sites and press — rather than the marketing sites of the companies being discussed. Owned content helps a search engine understand you; being recommended by an assistant depends far more on what others have published about you.
What is the difference between SEO and answer engine optimization?
SEO is largely a first-party game — your pages, your structure, your keywords. Answer engine optimization is mostly a third-party game: whether the places an assistant reads already mention you. A business can have an excellent website and appear in zero AI answers. That is precisely the position we measured for ourselves.
How do you measure AI citations?
We run a fixed set of buyer questions against AI platforms weekly through a data broker, then record which brands are named and which source domains the answers draw from. This article covers six keywords, two platforms and four weekly captures between 9 and 27 July 2026 — 48 observations.
Is this sample big enough to draw conclusions from?
It is big enough to show a strong pattern in one category and not big enough to generalize confidently to others. Six questions, one industry, one month. We are stating the size plainly so you can weigh it yourself, and we will keep publishing as the sample grows.
The Bottom Line
If you want an AI assistant to recommend your business, the highest-leverage work is mostly not on your website. It is on the platforms the assistant already reads — community discussion, video, review sites, and coverage by people who are not you.
That is slower and less comfortable than publishing another page. It is also what the data says.