What does automatic help centre import actually mean?
Three different mechanisms get sold under the same phrase, and they lead to very different weeks after signup.
The first is a crawl. You give the tool an address, it fetches the public pages, breaks them into passages and uses them as the source for AI answers. Nothing becomes an editable article inside the tool, and your existing site stays the single source of truth.
The second is a migration. The tool reads your articles out of your current platform through an API or an export file, then recreates them as native articles with categories, authors and sometimes translations attached. You now own two copies of the same content and have to decide which one people edit.
The third is a bulk upload. You hand over a CSV, a folder of HTML or Markdown files, or a stack of PDFs, and the tool ingests them once.
Most vendors support more than one mechanism, often at different plan levels. When a page promises automatic import, read it as a question: which of the three, how often does it repeat, and who is responsible for the content afterwards?
Which tools read a help centre by crawling a URL?
Crawling asks the least of you, because the only input is an address you already have. It is the normal approach for AI-first assistants, which treat your published pages as a retrieval source rather than as records to own, and it is increasingly offered as a layer on top of traditional ticketing suites as well.
Corebee works this way. You point it at your website and your knowledge base, it reads what is published there, and it answers from that material across web chat, email, voice, WhatsApp, Slack, Facebook, Instagram, SMS and Telegram without anyone rebuilding the content once per channel.
Zendesk, Intercom, Freshdesk and Tidio have all added AI answering that can draw on web and help centre content alongside their article editors. The differences worth checking are practical rather than philosophical: which plan tier the AI sits on, whether AI replies are billed apart from agent seats, and how often the crawl refreshes.
The limitation of crawling is that it sees only what a visitor sees. Pages blocked by your robots file, content behind a login, text baked into images, and articles that render only after a JavaScript call can all be skipped quietly. A crawl that reports success is not the same as a crawl that found everything.
Which tools expect a migration or a file upload instead?
If a platform's model is that articles live inside it, importing is a migration project rather than a crawl.
Zendesk has a mature help centre product and a well documented API, and there is a sizeable ecosystem of migration tools and consultants built around moving content into and out of it. Freshdesk takes a comparable approach with its solutions articles. Help Scout offers a tidy docs product that suits small teams who want a clean, hand-curated library rather than a sprawling archive. Chatwoot, being open source, gives you full API access and the option to self-host, which means the import is entirely scriptable and entirely your responsibility. Gorgias is shaped around ecommerce, so its content and macros tend to be organised around order status, returns and shipping questions.
None of these approaches is wrong. A migration gives you versioning, permissions and editorial control inside one system. It also costs more up front, and it creates the duplicate-content problem: once the articles exist in two places, someone has to decide where edits happen, and in practice edits happen in both.
How do these options compare for a small team?
Start from how your content is maintained today rather than from the feature grid.
If your help centre is already published on your website or a docs platform, and the marketing or product team keeps it current, a crawl-based tool fits the way you already work. You keep editing where you always edited, and the assistant follows along. If your help articles only exist inside your current support tool, you are migrating whatever you choose, so the quality of the vendor's importer and the availability of migration help matter a great deal.
Budget shape matters as much as the import itself. Per-seat pricing punishes you for giving part-time staff a login, which is exactly what a small business wants to do during busy periods. Per-resolution pricing on AI answers means a good import, one that lets the assistant handle more questions, raises your bill. Flat pricing removes both of those tensions. If you are weighing a full suite against something lighter, a side-by-side look at Zendesk alternatives is a reasonable place to see the trade-offs laid out.
Finally, think about who will do the work. A migration with field mapping, redirects and category rebuilding is a developer-shaped task. A crawl is not.
What should you check after the import and before customers see it?
An import that technically succeeded can still produce bad answers, so review before you go live.
- Stale articles. Old policies, retired products and superseded pricing pages are the most common source of confidently wrong answers.
- Internal documents. Make sure nothing meant for staff only, such as refund approval thresholds or escalation scripts, was picked up along with the public pages.
- Duplicates. Two versions of the same article, one updated and one not, give the assistant no way to pick correctly.
- Formats the crawler may skip. PDFs, tables rendered as images, and embedded video transcripts often carry the detail customers actually want.
- Other languages. If you publish in more than one language, confirm each version was read, not just the default one.
- Handoff behaviour. Test what happens when the assistant does not know. A clean escalation to a person beats a guess every time.
Ask a few colleagues to put the questions they get most often to the assistant, and read the replies as a customer would rather than as someone who already knows the right answer.
How do you keep the imported content current afterwards?
Import is a moment; accuracy is a habit. Decide early where the canonical version of each answer lives, and make everything else a reflection of it.
Crawl-based tools make that easier, because updating the page updates the answer at the next refresh. Check how often that refresh runs and whether you can trigger it by hand after a policy change. Migration-based tools need a working agreement instead: the library inside the platform becomes canonical, and your website links to it rather than repeating it.
Either way, a short monthly review pays for itself. Look at the conversations where the assistant handed off to a human and ask whether a missing or unclear article caused the handoff. Those transcripts are the best editorial brief your knowledge base will ever get, and a shared inbox makes them easy to read in one place.
What does a genuinely short setup look like?
A small business evaluation should be measured in an afternoon, not a quarter.
The quick version looks like this: connect your site so the content can be read, connect one channel, put a batch of real customer questions to the assistant, fix the gaps you find, then switch on the remaining channels. Setup with Corebee runs about ten minutes and needs no developer, it costs a flat $99 per month on the pricing page with unlimited seats and unlimited AI conversations and no per-seat or per-resolution fees, and it hands off to a human whenever someone asks or the answer is not clear. It also books meetings through Cal.com or Calendly and syncs to HubSpot, Salesforce, Pipedrive, monday and Zoho, and if most of your customers message rather than email, WhatsApp can be live in the same session.
Whichever vendor you pick, judge the import by the answers it produces on your hardest questions, not by the number of articles the dashboard says it found.