Social and community came to 1.9%, and a single code host holds the top of it
The whole social and community class took 1.9% of 20,143 citations while one repository appeared in 243 of 1,260 readings, which narrows a community plan considerably.
Published by AI Knows Us · Updated 19 October 2026
Across the 1,260 readings of ChatGPT and Gemini we took on 16 October 2026, the entire social and community class accounted for 1.9% of the 20,143 citations recorded. Inside that class, one host, github.com, appeared in 243 of the 1,260 readings and is the only social or community entry anywhere in the top ten. Put those two facts together and a community strategy for this category gets much narrower and much more specific than the advice usually given.
The earlier bank of questions, read on 15 October 2026 across 597 usable readings, put the same class at 1.5% of 10,652 citations. So the class is small on two independently designed banks, and the top of it is a place that holds files rather than conversation.
The arithmetic, stated plainly
The class share is counted in citations. The host figure is counted in readings that cited that host. They are different units and they cannot be divided into one another, which is exactly why a page like this has to say so rather than produce a tidy percentage that means nothing.
| Figure | Unit | Value | Date |
|---|---|---|---|
| Social and community class | Share of 20,143 citations | 1.9% | 16 October 2026 |
| Same class, earlier bank | Share of 10,652 citations | 1.5% | 15 October 2026 |
| github.com | Readings citing it, of 1,260 | 243 | 16 October 2026 |
| Forums, question sites, discussion platforms | Appearances in the top ten | None | Both dates |
What the table supports is a direction and a size, not a ranking inside the class. The direction is towards places that hold an object. The size is small.
The narrow version of a community plan
The broad version is to be active everywhere people in your category talk. The narrow version does three things and stops.
- Put the artefact somewhere with a stable address and a change history. A template, a specification, a dataset, a calculation. The thing itself, not a description of it.
- Correct the facts about you where they already exist. A thread that names you with a stale price is a liability you can fix in an afternoon, and fixing it is worth more than a new post.
- Be present where your buyers genuinely are, for their sake. Judge that on replies and enquiries, not on any source list, because the class is 1.9%.
Everything else that gets sold as community work is being sold against a class share that two separate banks of questions measured at under two points.
Why the code host is the exception that proves the rule
A repository is social in the sense that people collaborate on it, and it behaves like documentation in every way that matters to a source list. It carries a canonical object rather than an opinion about one, a version and a date that a machine can read, a stable path that keeps pointing at the same thing, and attribution written into the record rather than claimed in prose.
None of those four is available to a thread, which gets edited, locked, deleted or buried, and none of them requires the author to be persuasive. That is the practical difference between the one entry in our top ten and the rest of its class.
What this reading does not establish
It is one category, ours, on two dates, read on two apps: ChatGPT and Gemini. A bank of questions about machinery, conveyancing or veterinary practice would surface a different community layer, and nothing here forecasts it. Our product reads five surfaces and three of them were asked nothing in either run. There is no per industry figure in either bank. City level numbers come from a separate scan of ten named places. And both class shares count a domain appearing in a source list, which is a different event from a business being named in the answer text.
We sell AI visibility measurement and the publishing work that follows it, so the figure that argues against us is printed with the rest: our own domain was in the source list of 6 of those 1,260 readings. The Google Search Console side of our own record, with both windows and the export date, is our proof page.
Three ways a community plan gets measured wrongly
The class is small, which makes the measurement easy to get wrong in ways that flatter the activity.
- Counting your own posts as the result. Volume posted is an input. The output is whether a page carrying your facts turns up in an answer, and those are different columns.
- Counting a mention in a thread as a citation. A name appearing in text and a domain appearing in a source list are two separate events with two separate denominators, and a report that slides between them has stopped being a report.
- Averaging the class with the rest of the table. A blended visibility score built across own site, documentation, news, community and directories hides the only thing you can act on, which is which class your absence is in.
The fix for all three is the same and it is boring: one column per class, one denominator per column, and a date at the top. A reading built that way is comparable to your own next reading, which is more than any blended score can offer.
Questions about social and community sources in AI answers
What share of citations is social or community?
1.9% of the 20,143 citations recorded on 16 October 2026, and 1.5% of the 10,652 recorded on 15 October 2026 from a differently written bank.
Does that include forums and question sites?
Yes, the class covers them, and none of them reached the top ten on either date. The single social or community entry in the 16 October top ten is a code host at 243 readings.
Should I stop posting in community spaces?
Post where your buyers are, because people reading you is a sufficient reason. What these two class shares settle is that it is not an efficient route into a source list in this category.
Why does a repository behave differently from a forum?
It holds the thing being discussed rather than a view about it, with a version, a date and an address that stay put. Those are the fields an answer about the present actually needs.
Which engines and dates produced these shares?
ChatGPT and Gemini on 15 and 16 October 2026, at 102 then 210 questions per market across three markets. The first run's method and complete source mix are published as 597 readings across ChatGPT and Gemini in three markets.
Publish the object, then go and be useful
The honest community plan in this category is one artefact published properly and a habit of correcting stale facts about yourself wherever they already sit. Both are finite pieces of work with a visible finish line, which is more than can be said for being active everywhere. The class is 1.9% and the top of it is a filing cabinet, and that is a cheaper thing to know before the budget than after it.