Support tickets are down. That should be good news: deflection is working, support costs should be coming down. Increasingly, it is not.
Customers are resolving product issues, configuration questions, and troubleshooting problems in AI conversations that never touch your support infrastructure. They are asking ChatGPT how to configure an integration, asking Perplexity why an error is occurring, asking Claude to explain a workflow your documentation covers poorly.
Some of these conversations end with a correct answer and a satisfied customer. Others end with a customer acting on confidently wrong information, making the problem worse, and eventually opening a ticket frustrated about something that should have been simple. Or silently frustrated about your product or service.
Either way, none of it shows up in your support dashboard.
What Is DIY AI Support?
DIY AI support describes the growing behavior pattern of customers using public LLMs like Gemini or ChatGPT to resolve product questions instead of using a vendor’s official self-service channels. The customer gets an answer, or attempts to, entirely outside any system the vendor controls or measures.
This is not the same phenomenon as traditional self-service, and the distinction matters. Traditional self-service assumes the customer engages a channel you built: your knowledge base, your chatbot, your community forum. You control the content, you can measure engagement, and you can improve the experience based on what you observe.
DIY AI support removes the vendor from that loop entirely. The customer’s first stop is a general-purpose AI system that has ingested your public content, synthesized it with what everyone else is publishing about your product category, and produced an answer you never see and cannot correct in the moment. Even your competitors can shape this channel, by publishing content that frames your product as difficult and positions theirs as the easier alternative.
The Post-Sales Dark Funnel
There is a direct parallel between the DIY AI support channel and the pre-sales dark funnel. The dark funnel originally described buyer research happening in channels vendors could not track: peer conversations, analyst calls, and increasingly, AI-mediated research where buyers built complete vendor shortlists without visiting a company website. Marketing lost visibility into a large and growing share of the buying journey.
The same structural blindness is now happening on the other side of the sale. Customers are resolving product friction in AI conversations that generate no ticket, no chat log, no CSAT survey, and no signal your support or CS team can act on. The dark funnel did not stay confined to pre-sales. It has a post-sales mirror, and most CX organizations have not noticed it forming.
There is an important overlap worth understanding, because it changes how you should think about content strategy. A meaningful share of the questions customers ask AI systems are not cleanly pre-sales or post-sales. “What is the best way to structure a multi-region deployment for this type of platform?” could be asked by a prospect evaluating vendors or a customer three months into implementation. “How do I synchronize customer data from my CRM with this product?” is the same question whether the audience is qualifying you or already running you.
The content that answers these questions well serves both audiences, which means your AEO strategy and your knowledge base are no longer separate workstreams targeting separate funnel stages. They are increasingly the same content, consumed by AI systems that make no distinction between a buyer and a customer asking a “how do I” or “what’s the fastest way” question.
Is this a Support Problem or a CX Problem?
DIY support does not fit neatly into a CLG conversation about retention and expansion signals. Its impact is more foundational: it changes what “customer experience” actually means, because a meaningful share of the experience customers have with your product is now mediated by content and systems the CX function has never audited.
The quality of your public documentation is a direct determinant of what AI systems tell your customers when they hit a problem. Sparse help articles, gated troubleshooting guides, and outdated FAQs do not just weaken your search rankings. They shape the accuracy of the answer a customer receives at 11pm on a Tuesday when nobody on your team is available and the customer has already decided not to open a ticket.
This connects directly to Experience Yield, the revenue return on customer experience investment. Most CX programs measure the interactions they control: CSAT after a support ticket, satisfaction after onboarding, NPS after a QBR. DIY support happens entirely outside that measurement boundary.
A customer who got a wrong answer from a public LLM, struggled silently, and grew quieter about renewal never appears in a CSAT report. The experience happened. Your measurement system simply was not built to see it.
What DIY AI Support Means for CX Strategy
Public documentation is now support infrastructure, whether you designed it that way or not. Content that used to be secondary — sparse help articles, internal-only troubleshooting notes, documentation gated behind login walls — is now a primary channel, because AI systems can only synthesize what they can actually access.
If your best troubleshooting content sits behind a support portal login, an AI system answering a customer’s question at 11pm has nothing accurate to draw from and will either give a generic answer or an outdated one. And sometimes, LLMs will answer technical questions without using any of your content, relying purely on blog posts, forums, communities, and potentially stale third-party sources.
Knowledge lifecycle management becomes a retention lever, not a documentation chore. Content that is accurate but stale is almost as dangerous as content that was never written. A help article describing a workflow your product changed six months ago will confidently mislead both AI systems and the customers relying on them. Keeping documentation current is no longer a “nice to have” content operations task. It directly determines whether the answer a customer gets is correct.
Your existing self-service metrics cannot see this behavior, so you need a different measurement approach. Ticket deflection rate, help center search volume, and chatbot resolution rate all measure engagement with channels you control. None of them capture what happens when a customer skips those channels entirely.
Understanding DIY AI support requires testing what AI systems actually say about your product when asked common support questions, not just monitoring the channels you built.
For example, we at A6 Group track a “health score” for each support scenario, based on public LLMs and your own self-service channels. This is part of our isalo AI-native managed service: we continuously test your self-service and support experience across digital channels, and surface issues and recommendations to your team.
The overlap between pre-sales and post-sales content means AEO and knowledge base strategy should not be run by separate teams working from separate content libraries. A prospect asking an AI system how your category of product handles a specific technical requirement, and a customer asking the identical question three months post-implementation, are being served by the same underlying content.
Treating AEO as a marketing initiative and the knowledge base as a support initiative creates duplicated effort and inconsistent answers to the same question depending on which team wrote which piece of content.
What Good Looks Like
A mature response to DIY AI support does not try to eliminate it. Customers are not going back to searching help centers first once they have discovered that a conversational AI system can often solve the problem faster. The realistic goal is ensuring the answer AI systems give is accurate, complete, and reflects your product as it actually works today.
Sometimes, it also means not giving a “how to” answer, and guiding the customer to contact support or their CSM instead. LLMs have a tendency to answer questions at all costs, with pages of step-by-step instructions for complicated tasks. Support might have a simple, direct answer to the customer’s need that the LLM completely missed.
That requires treating your public content (documentation, community discussions, third-party reviews, structured help articles) as a system that needs continuous testing, not a static library that gets updated when someone notices it is wrong.
The questions customers ask change as your product changes. The answers AI systems give change as models update and as your competitors publish content that reshapes how AI systems describe the category. A documentation audit run once a year will not catch what is happening in the interim, and DIY support has no quarterly cycle. It happens every day, in real time, whenever a customer has a problem and decides asking an AI is faster than opening a ticket.
DIY AI support is not a reason to abandon your self-service and support infrastructure. It is a reason to recognize that a growing share of the customer experience is now happening in a channel you do not operate and cannot directly moderate. The CX function’s job has expanded to include making sure the AI systems your customers are already using say the right thing when they get there.
Wondering what AI systems are actually telling your customers right now? A6 Group operates isalo, an AI-native service that continuously tests your self-service and support experience across digital channels, surfacing where the answers customers get are breaking down before it costs you a renewal. Contact us to learn more.