How to Pick the Right AI Support Agent for SaaS


TL;DR
When you delete or hide docs for deprecated features, your AI agent hallucinates answers and customers file tickets citing workflows that no longer exist. A clearly labeled deprecated section keeps old content findable but contextually honest, so both humans and AI know exactly what era they are reading from.
You sunset a feature. You pull the help article. Six months later your support queue fills with tickets from customers who found a cached version on Google, a community post quoting your old docs, or — most frustratingly — an AI answer that confidently described a workflow that stopped existing in Q2.
Deleting documentation doesn't erase it from the internet. It just means your authoritative version is gone, so every other version wins by default.
Three things break at once.
First, organic search still surfaces the Google cache or a third-party screenshot for weeks or months after removal. Customers arrive expecting the feature to exist, find nothing in your help center to confirm or deny it, and open a ticket.
Second, if you run an AI support agent — ours included — the model's training or your knowledge-base sync may have ingested the old article before you deleted it. Without a clear deprecation signal in the content itself, the agent has no way to know the feature is gone. It either repeats the old instructions confidently or gives a hedged non-answer that frustrates the customer either way.
Third, your own team loses institutional memory. A new support hire who joined after the deprecation has no reference point when a customer insists the old workflow existed. The ticket escalates unnecessarily.
A deprecated section is not a dumping ground. It is a deliberately maintained archive with exactly three elements per article:
The section lives in your help center under a clearly named category — "Deprecated & Discontinued" works — with a robots meta tag or sitemap exclusion if you want to suppress new indexing, but without a 404 or redirect that wipes the content entirely.
Chattering indexes your help center to answer customer questions. So does every comparable AI support tool. When we sync your content, article context matters enormously. An article titled "How to use the Legacy CSV Importer" that opens with "This feature was removed on [date]. Use the new Data Import flow instead — see this article" gives the AI agent everything it needs to respond accurately: the feature is gone, here is the date, here is the redirect.
An article that simply 404s — or worse, an article that still says "follow these five steps" with no deprecation signal — gives the agent nothing useful, or gives it actively wrong instructions.
The deprecation banner is not just for humans. It is a structured fact that your AI agent can cite when a customer asks about the old workflow.
Say a deprecated feature generated 40 help articles over its lifetime. If even a fraction of your monthly ticket volume involves customers confused about what happened to it, a graveyard section eliminates the ambiguity in a single pass. The customer finds the article, reads the banner, clicks the migration link, and never opens a ticket. Your agent, when asked, surfaces the deprecation context and the redirect rather than stalling or guessing.
You do this once per deprecated feature. The payoff compounds as your product ages and your feature history grows.
Start with your last two major version changes. Pull the articles that now describe dead workflows, add the standard banner, move them to the deprecated category, and update any internal links in your current docs to point to the archived version with an explanatory note. The whole pass for two version cycles usually takes a support lead an afternoon.
For new deprecations, make the graveyard move part of your feature-removal checklist. Engineering announces the sunset → support lead moves the articles and adds the banner → knowledge base sync picks up the new context → AI agent answers accurately from day one of the removal.
Help centers degrade in exactly the same way codebases do: not through dramatic failures but through quiet accumulation of things that used to be true. A graveyard section is the documentation equivalent of clear deprecation warnings in code — it doesn't pretend the past didn't happen, it just makes clear what era you're reading from.
Customers who hit that section don't feel lost. They feel informed. And an AI agent with access to well-labeled deprecated content doesn't hallucinate. It tells the customer exactly what happened and where to go next.
Only if the new article is a true one-to-one replacement. If the feature was discontinued with no equivalent, a redirect misleads the customer — keep the archived article with a deprecation banner instead.
Add a noindex meta tag to the deprecated category if you don't want the pages to rank, but keep them accessible by direct URL so existing links and your AI agent can still resolve them correctly.
When a Chattering-indexed article contains a clear deprecation signal — typically a banner stating the removal date and the replacement — the agent surfaces that context directly in its answer rather than repeating outdated instructions.