Industry Research · Customer Community & AI

The support forum is becoming AI infrastructure.

By Albert Richer, Founder & Lead Editor Updated August 17, 2026 / 7 min read

For a decade, the pitch for a customer community was simple: give people a place to answer each other’s questions and you deflect support tickets. That was the whole business case, and it badly undersold what a community was quietly becoming.

In 2026 the community has a second audience that never files a ticket: machines. Google now treats forum threads as a distinct kind of source. Reddit answers questions with AI built on top of its own discussions. And every AI support agent a company buys is only as good as the solved problems it can pull from. The archive of questions and answers piling up inside a community turned into infrastructure.

The forum became a knowledge layer, and machines are the readers

Community content used to sit behind a support portal. Now it’s structured input for the systems that answer questions about you.

Start with Google. It now publishes a dedicated schema vocabulary for forum content, so its systems can read a thread’s posts, authors, and replies as a discussion and surface them in Search features built for exactly that. Google expanded the supported forum and Q&A properties in March 2026 to understand comment structure better.Google Community discussion is no longer just content behind a portal; it is a recognized kind of source.

The clearest sign that the conversation itself is the asset is Google’s expanded Reddit partnership. Google didn’t just want links to Reddit pages; it obtained structured, fresher access to Reddit’s continually changing posts and comments to understand, display, and train on them.Google And it shows up in the citations: one analysis of more than 150,000 AI answers found Reddit cited in roughly 40% of them across ChatGPT, Perplexity, Gemini, and Claude, making it the single most-cited source in AI search.Citation analysis

Source: published 2025-2026 AI-citation analyses. Measures differ by study (share of all citations vs. share of top sources), so treat these as directional, not identical metrics.

Reddit is also showing what community search becomes next. Its AI Answers feature summarizes relevant discussions and links back to the underlying posts, and in May 2026 Reddit merged Answers into its main search.Reddit Users ask a question, AI synthesizes the discussion, and the original human conversation stays underneath as the evidence layer. For a customer community, that is the whole opportunity: five years of solved implementation problems and expert replies can become retrieval material for an AI agent instead of an archive nobody searches.

That matches what I keep running into. I stood up a modern community on Circle to test it, and the thing that surprised me wasn’t the posting, it was how cleanly every answer became a searchable, linkable little document, the kind of structured record an AI can actually retrieve. The old forums I ran a decade ago buried that same knowledge in a wall of replies no machine could parse.

An AI agent is only as good as the knowledge behind it

Buying an agent is easy. Building the knowledge system that makes it useful is the hard part, and that system is your community.

The economics are why everyone is buying. McKinsey’s 2026 service data puts an AI-resolved ticket at roughly $0.62 versus $7.40 for a human agent, about a 12x gap.McKinsey The best tools deliver on it: top-quartile AI-native platforms now resolve up to 58.7% of tickets end-to-end, while legacy systems with bolt-on AI stall at 10% to 25% true resolution.Lorikeet Adoption has followed the math. Salesforce’s 2026 State of Service survey of 3,075 professionals reports AI-agent adoption jumped from 39% to 66% in a single year, with 70% seeing value within 60 days, though Salesforce sells these agents, so read it as sponsored data.Salesforce

Here is the catch. Intercom surveyed 2,470 support professionals and found that while 82% invested in customer-service AI last year, only 10% call their deployment mature. Among the mature teams, 87% reported improved metrics versus 62% overall, and 40% of teams say their agents now spend more time training and optimizing the AI.Intercom Gartner is blunter: it predicts more than 40% of agentic AI projects will be canceled by the end of 2027 over cost, unclear value, and weak controls, warns that a third of companies will hurt customer experience by deploying AI too early, and estimates only about 130 of the thousands of “agentic” vendors are the real thing, the rest is “agent washing.”Gartner

What separates the 10% from everyone else isn’t the model. It’s the knowledge system feeding it: community answers, knowledge base, docs, and ticket history, turned into something an agent can retrieve, with a clean handoff to a human when it can’t. The community sits at the front of that pipeline, which is why the market is consolidating around whoever owns it. In June 2026 Salesforce agreed to buy Fin, formerly Intercom, for $3.6 billion, folding a flat-fee AI support agent into Agentforce, its per-action platform that just hit $1.2 billion in ARR.Salesforce/Fin The per-action vendor bought the flat-fee vendor.

When AI reads the community, moderation becomes data quality

A bad forum used to waste a customer’s time. An AI-connected bad forum can teach the support system the wrong answer.

Once an AI retrieves from your community, the stakes on what lives there invert. Spam, fake engagement, and confidently wrong replies stop being an annoyance and start becoming training data, amplified into the answers your agent gives and the answers AI search gives about you. The more valuable the corpus, the more expensive it is to let it rot.

Reddit’s own numbers show the scale of defense that now takes. It says newer automated systems cut users’ exposure to spam by 20%, revoke nearly 2 million fake votes a day, and dropped enforcement time on hateful or violent content from hours to under five seconds.Reddit That is the price of keeping a community trustworthy enough for machines to learn from.

And moderation is now enforced law, not a threat. In December 2025 the European Commission issued its first non-compliance fine under the Digital Services Act, €120 million against X, for deceptive design, an opaque ad repository, and blocking researcher data access.European Commission The EU’s Appeals Centre received more than 24,000 disputes in its first year, about one every 22 minutes, and disagreed with platforms’ own moderation decisions 59% of the time; in hate-speech leave-up cases it overturned platforms 70% of the time, and TikTok specifically 83%.Appeals Centre Those transparency reports are now standardized and machine-readable, so moderation itself is becoming auditable data.European Commission

All of this while the teams doing the work got thinner. Community professionals now manage roughly twice as many members per admin on flat headcount, member participation rose 40% in 2025, and 63.3% name sustaining participation as their biggest challenge, with resourcing close behind at 34.7%.Hivebrite That squeeze is exactly why platforms are baking AI moderation, reputation systems, and compliance dashboards straight into the queue.

Which is where our own evaluations get concrete. Khoros is the one I keep coming back to for this: it scores dozens of member activities into reputation and runs the moderation and audit tooling a Fortune 100 community actually needs, SOC 2 audits and all. The catch, and I hit it firsthand chasing a quote, is that there is no public pricing and it lands somewhere in the five figures a year, so the governance is enterprise-grade and so is the bill.Khoros