In what felt like the blink of an eye, GPT flooded SaaS applications. From Notion to Retool, GPT announcements have taken over Product Hunt, Hacker News, and online developer chatter. And while certain GPT integrations might be a bit extraneous, many applications will thrive because of GPT.
At CommandBar, we’re invested in one big topic regarding GPT—while GPT will certainly impact app help centers, we are curious whether GPT will replace them altogether. If you aren’t familiar, app help centers are wiki-like websites that host a collection of articles to assist users with content.
First, why are we connected to this problem? Since around 2021, CommandBar has been focused on improving application onboarding and UX. One of our main solutions is making it easy to surface help center content in-app to the user. We’ve long believed that forcing users to context-switch to an external help center isn’t a great user experience; CommandBar’s HelpHub reduces friction by encouraging users to solve their problems without leaving the app.
We have operated under the assumption that an Intercom-style help center would always exist. However, GPT has us asking a new question: Will GPT make help center websites obsolete?
How GPT + help content works
A basic GPT integration with help center content is rather simple. Help centers are full of documents detailing how an application works. Today, developers can ingest that help center content into GPT as an initial prompt. Then, they can ask GPT to answer a real user question based on the ingested information.
Of all the GPT integrations out there, this is a rather straightforward one. Of course, there are sometimes hiccups. GPT might occasionally hallucinate responses that have nothing to do with the provided corpus of help content. There are precautions a developer could take—they can and should instruct GPT to only answer questions strictly based on the provided material. But any developer who’s worked with GPT knows that is hardly a guarantee.
We’ve thought about this for our product
CommandBar isn’t a help center solution, but we have always integrated with them. Even before GPT-3 launched, we built a product that’s related to this discussion. Originally, users could use CommandBar’s HelpHub to search Intercom, Gorgias, or another help center from within an app.
CommandBar even included features for developers to provide synonyms of common search terms to better surface content. But GPT completely expands this idea; using GPT, we can generate a personalized article specific to the user’s unique query. And, to our surprise, it’s been remarkably fruitful.
But will they disappear?
The bigger question for us and the support application community is whether help center websites will totally disappear in favor of GPT-powered answers. While support applications will always exist, it’s unclear if the help center will remain a separate, Wikipedia-esque website. Imagine if that was totally replaced by a GPT-powered chat to answer any user query.
To us at CommandBar, there are some compelling arguments for why this might happen. And there are some clear roadblocks making an opposite case. Let’s run through those.
Why help centers might disappear
We can think of four excellent arguments for why help centers might actually disappear. Like poof—no more help.SaaSApp.com apps, just a single box to request AI-powered support.
The first three reasons detail why help centers are often ineffective support solutions. The final reason discusses how help centers are internally managed.
1. Help centers are often ineffective
Most of our clients have help centers, and their most common complaint is that their help centers confuse users. This is because help center articles often only partially answer questions; users then have to jump between several different articles to actually answer their questions.
For instance, “How to search Lead Conversions in Prospective Organizations” may require a user to discover separate articles detailing “Conversions” and “Prospective Organizations”. That isn’t a very efficient and friendly user experience.
GPT can fix this. GPT can answer any question with a response that pulls the right amount of information from each article, synthesizing it into a single answer for the user.
2. Help centers are often overwhelming
Oftentimes, help centers have articles that detail features in full. This is great for learning about how an application works but not effective for answering specific questions. If you’ve ever jumped through a help center and found yourself Cmd-F’ing a lot, you’ve likely run into this problem.
GPT is effective for only providing information that’s actually relevant to the user.
3. Users don’t know what to search for
We’ve dealt with this one a lot at CommandBar: Users often don’t know how to search for their problems. It’s not their fault. Different apps use different terms, and sometimes it’s easier to describe a feature than to ask about it. Users know the gist of what they need to ask, but cannot nail the right verbiage.
This results in users jumping around a help center trying to answer their questions. But GPT can actually infer a user’s intent by digesting their language, and if GPT’s response is not up to par, users can ask follow-up questions that GPT will be able to build off of.
In other words, instead of a series of discrete questions frantically trying to address a question, GPT turns support into a conversation between the user and the help corpus of knowledge.
4. Help centers are hard to manage
There are two problems that plague help centers: they are hard to organize and hard to keep up to date.
Help centers often have redundant content, either in an attempt to point the user in the right direction or just because of a poor distribution of articles. While redundancy isn’t necessarily a bad thing for the user, it becomes difficult to manage.
GPT simplifies this problem. Instead of managing a wiki of articles, teams just need to provide GPT free-form information about a product in a master document. Granted, that document might be broken into various sections by future helpdesk software, but the general idea is that core data isn’t a bunch of articles, but rather a single cohesive corpus of knowledge. This could make managing support way easier, appealing to support teams. After all, what’s the point of a support center if it is frequently out of date?
Why help centers might not disappear
While help centers have their flaws, there are many reasons why they aren’t going anywhere in the foreseeable future.
1. Help centers help prospective users
A help center isn’t just a resource limited to existing users. Help centers are also indexed by search engines (like Google) which enable prospective users to discover the product and answer any product-related questions. Further, depending on how a help center is indexed, it might also contribute to a website’s SEO domain rank.
Personally, I cannot count the number of times I’ve searched a help center to decide if a SaaS app was compatible with my workflow.
This hybridized purpose of the help center is a pretty persuasive argument against their potential extinction. Help centers aren’t just an asset for users, but a tool for sales and marketing teams to grow user bases. It’s hard to imagine apps getting rid of that, even if GPT is technically more efficient; growth is typically a top-of-mind priority.
2. GPT isn’t flawless
GPT has its flaws. It frequently hallucinates content, and while precautions can be taken, it’s never guaranteed that GPT will always be accurate. GPT-4 might’ve improved on these issues, and future iterations of GPT will continue to improve, but the product is built on statistical models. And statistical models can fail.
This is especially bad for the users with questions that haven’t been included yet in a knowledge base. The odds of a model hallucinating content are inversely proportional to whether the content was discussed in the knowledge base.
It’s hard to imagine that companies would rely on GPT as the sole resource for assistance when its accuracy isn’t a guarantee.
3. GPT isn't designed for step-by-step image guides
When it comes to SaaS apps, questions aren’t just limited to “Do you have this feature?”. Users often need step-by-step guides detailing how to complete an action. These guides typically require images to visually cue users where to look. With images, guides reduce the mental headache of following directions.
While GPT is excellent at producing step-by-step guides, it cannot generate custom images to accompany those generated step-by-step guides. Sure, GPT might be able to return an image related to a guide, but if a guide is totally bespoke to the user’s query, there will remain some mismatch between the guide’s content and the saved images.
The likely: Help centers will rely on GPT as a backbone, but not a single surface
Help centers aren’t going to go. They are critical for helping prospective users, hosting image-driven step-by-step guides, and providing a fail-safe if GPT hallucinates content. However, GPT will become a major tool for support too, and companies will likely encourage users to use GPT-powered support when possible.
We can already see this trend happening. Many help center applications have announced features with AI-powered support. Intercom announced Fin, a GPT-powered bot that queries Intercom articles and returns personalized answers. Gorgias AI provides a similar solution, answering questions based on Gorgias content.
Third-party products will also use GPT to improve upon core help center solutions. For instance, CommandBar uses GPT to provide bespoke answers to in-app queries. This frictionlessly assists users while they attempt to use the application. And some solutions will use help center content to assist other support features like real-person chat support; solutions like Brevy are doing that for e-commerce.
Closing thoughts
We always knew that GPT would change customer support. And it already has.
However, we weren’t exactly sure how much GPT would replace support as opposed to serving as a companion. As far as we can see, GPT isn’t going to replace help center websites; apps will instead encourage users to use GPT-powered support but continue to keep app centers up as a fail-safe.
Of course, GPT is young, and potential future features could change this conversation. In six months’ time, we’ll check back in and revisit this topic. If you’re curious and want to stay updated about this, subscribe to our blog—we post articles about onboarding, UX, and AI-powered support occasionally.
