AI engines quote the thread. Your service page is not in the conversation.

What the measurements actually say

Independent studies of AI citation sources disagree sharply on the size of the effect. Published 2026 figures for Reddit range from a couple of percent of Google AI Overview sources in one dataset to close to half of Perplexity’s sources in another. The spread is the honest finding: the number moves with the query set, the vertical, the language and the month it was measured.

What the studies agree on is direction. Across ChatGPT, Perplexity and AI Overviews, a small group of community and reference domains — Reddit, Wikipedia, YouTube and LinkedIn — account for a large share of everything cited. If a buyer asks an assistant for a recommendation in your category, the answer is more likely to be assembled from those than from your service page.

We report the range and the source rather than one convenient percentage, because a single number quoted without its method is how this field filled up with statistics nobody can defend.

Why the shortcut does not work

The obvious idea — spin up a few accounts, post glowing recommendations, get quoted — fails on mechanics before it fails on ethics. Reddit’s content policy prohibits content manipulation, including sockpuppeting and posting mainly to promote a business. Communities have moderators who do this all day and are good at it. A removed comment cites nothing, and an account tied to your brand that gets banned takes its mentions with it.

There is a slower failure too: it works for a while, someone notices, and the thread that ranks for your category becomes a thread about a company that astroturfs. That is a worse outcome than not being mentioned at all.

How we do it instead

1. Find the threads that already matter. Not every thread — the ones already ranking for your commercial queries and already appearing as citations in AI answers. That list is usually shorter than clients expect.
2. Check what each community allows. Every subreddit sets its own rules. Some ban promotion outright, some route it to a designated weekly thread, some only ask for disclosure. We read them before writing anything.
3. Answer the question properly, with the affiliation stated up front. In the first sentence, not a footer. If a comment would not be worth posting without the mention, we do not post it.
4. Contribute far more than we promote. Most of what gets posted will not mention you at all. Roughly nine useful contributions for every one that does is both what the platform expects and what keeps the account credible.
5. Never the same text twice. One comment, one thread. Repeating a paragraph across communities is the fastest way to be classified as spam.
6. Measure against AI answers. The metric is whether the assistant’s answer changed, not how many comments were posted.

Where this fits, and where it does not

This does not replace your own pages. Community citation moves the recommendation; your site still has to convert the person once the assistant sends them. It also does not work in every category. Where buyers do not discuss options publicly there is no thread to be useful in, and we will say so rather than manufacture one.

It works best where a real discussion already exists, where you have genuine expertise to contribute, and where somebody on your side is willing to be named. That last condition is usually what decides whether we recommend it.