Should a Blog Have an AI Chatbot? When It Helps a Content Site (and When It Doesn't)
On a blog with hundreds of evergreen posts, a chatbot can point readers to the right guide. On a small or news-only blog, it mostly gets in the way.
Published Updated 8 min read
A blog should have an AI chatbot when it has a large archive of evergreen posts and readers keep asking which one they need: "which tool works for X", "where's the guide for Y", "do you have a tutorial for beginners". The bot answers in a sentence or two and links the right post. A blog with a handful of posts, or one that's mostly news, usually doesn't need one: the menu and a search box already do the job.
I build ChatBento, and content sites are a case where I'd be more careful than with a shop or a service business. A store chatbot answers questions about one business. A blog chatbot is really a guide to your archive, and it's only as good as the posts you let it read.
What can a chatbot do on a blog that site search can't?
Site search matches words. A reader who types "cheap hosting for a portfolio" gets every post containing "hosting", sorted by date or by a ranking they can't see, and has to open five tabs to find the one that answers the question.
A chatbot matches the question. It finds the passages in your posts that answer it, replies in plain words, and points to the post. That works because of retrieval: the bot looks up the relevant pieces of your content for each question instead of relying on what the AI model happens to know.
Here's how I'd compare them honestly:
| Site search | Chatbot | |
|---|---|---|
| Best at | Finding a post when you know a word in its title | Finding a post when you know your problem, not the word |
| Reader effort | Scan a results list, open several posts | Read one short answer with a link |
| Handles vague questions | Poorly | Well |
| Shows everything you've written on a topic | Yes | No, it picks the most relevant few |
| Cost | Often included with your platform | A monthly plan, with a message limit |
| Risk of a wrong answer | None, it only lists posts | Low if it sticks to your posts, but not zero |
They aren't rivals. Keep the search box for readers who want to browse everything you've written about a topic. Add a chatbot for readers who want an answer and the one post that has it.
Which blogs does a chatbot actually help?
The content sites that suit a bot share one trait: readers arrive with a problem, not to read the latest post.
- Tutorial and how-to blogs. "How do I export my layers?" has an answer somewhere in post 214, and the reader doesn't know that.
- Tool and software review blogs. "Which of these works for a team of three?" is a question your comparison posts answer, but not in one place.
- Template, plugin and download sites. "Is there a free version for Figma?" "Does this one work with WordPress 6?" Short factual answers, with a link to the download page.
- Niche publishers with deep archives. A gardening blog with 800 posts is a reference library. Readers want the one about their plant and their climate.
In each case the questions repeat and the answers are already written. The bot's job is to connect the two.
When does a chatbot not help a blog?
I'd skip it, or at least wait, in these cases:
- You have a few dozen posts or fewer. A clear menu and category list will get readers where they need to go faster than a conversation.
- The site is mostly news. News posts go stale quickly, readers usually arrive for one story from search or social, and the questions they'd ask ("what happened next?") are about events you haven't written up yet. A bot trained on last month's news will answer from last month.
- It's a personal or opinion blog. People read essays for the essay. A bot summarising your argument in three sentences isn't what they came for.
- Your posts contradict each other and you won't fix them. If the 2022 post recommends one tool and the 2026 post recommends another, the bot may quote either. Outdated posts are the usual source of outdated answers.
There's also a cost question. A popular blog can get a lot of curious chats, and every reply uses part of the monthly message limit. On my tool the limit is hard: when it runs out, the widget shows a contact form instead of AI answers until the next month, so the bill stays flat. That protects your wallet, but a busy blog needs a plan sized for its traffic.
Which posts does the bot read first when you hit the page limit?
This matters most for a big blog. Every chatbot plan caps how many pages it will train on. ChatBento's caps are 50 pages on Free, 200 on Hobby ($8 a month), 2,000 on Standard ($30) and 10,000 on Pro ($100), per chatbot.
If your blog has 600 posts and your plan allows 200, which 200 does the bot get? In ChatBento's case, the crawler orders pages like this:
- The page you start it from (usually your homepage).
- Pages linked from that page, shallow addresses first.
- The rest of your sitemap, from shallow addresses to deep ones.
- Listing pages (tags, categories, authors, archives, search, page 2 and on) last.
It also skips pages marked noindex, and pages that need JavaScript to show their text may come through empty.
What that means in practice:
- Your homepage and main menu decide a lot. Posts you link from the homepage, like a "start here" list or your best guides, get read first. Posts buried five clicks deep in the archive may not make the cut.
- Choose what the bot reads, don't leave it to the order. After the first import, look at the list of trained pages and delete the ones that don't help: dated news, roundups of things that no longer exist, giveaway announcements, "we're back from holiday" posts. That frees slots.
- Add your evergreen guides one by one if they were missed. You can add a single page by its address, which is the way to make sure your ten most useful tutorials are always in.
- Add new evergreen posts when you publish them. A guide the bot hasn't read is a guide it can't recommend.
I'd rather have 200 well-chosen evergreen guides than 2,000 posts including every outdated one. Fewer, better pages give better answers.
Why tag, category and archive pages are poor training material
Tag, category and archive pages are lists of other posts: a title, an excerpt, a "read more" link, repeated twenty times. For a chatbot they're weak material for three reasons:
- They hold no answers. An excerpt is the first two sentences of a post, not the part that answers the question.
- They duplicate content. The same excerpt can appear on a tag page, a category page and three archive pages, which crowds out the real post when the bot looks for relevant text.
- They make bad link targets. If the bot answers from a tag page, the link it gives is to a list, and the reader has to search again.
My crawler puts listing pages at the end of the queue for this reason, and many blogs hide them from search engines anyway. If any slip into your trained pages, delete them.
How do you keep answers linking to the source post?
The point of a blog chatbot is to send people to your posts, not to replace them. A few things keep it that way:
- Each piece of text the bot retrieves comes with its page title and address, so it can link to the post it answered from. Clear, specific post titles help it pick the right one.
- Write one topic per post where you can. A post that answers one question well is easier to find and link than a 6,000-word "everything about X" page.
- Add an instruction for links. Something like: "When you answer from a blog post, end with a link to that post so the reader can see the full guide." Built-in replies are short (one to four sentences, or a short list) and can include links that appear in your content, so an instruction like this fits naturally. I wrote more about this in How to Write Instructions for Your Support Chatbot.
- Update or remove posts that disagree. If your old and new recommendations conflict, fix the old post or take it out of the bot's pages. More on choosing what the bot reads is in How to Train an AI Chatbot on Your Website.
What should you check after the first month?
Read the conversations. On a blog they're unusually useful:
- Questions the bot couldn't answer are posts you haven't written. That's a content plan written by your readers, which I covered in Your Chatbot's Conversations Are a Content Plan.
- Questions it answered with the wrong post mean two posts overlap or a title is unclear.
- Your message usage tells you whether the plan fits. If you hit the limit in the third week, either the bot is popular (good) or it's answering a lot of off-topic chat. The bot already declines questions unrelated to your site, so check which it is before upgrading.
The short checklist
- At least a few hundred evergreen posts, or a clear reason readers need help finding them
- Your best guides linked from the homepage or a "start here" page
- Trained pages reviewed: dated news, giveaways and roundups removed
- Tag, category and archive pages removed if any were trained
- Missing evergreen guides added one page at a time
- An instruction to link the source post in every answer
- Old posts that contradict newer ones updated or removed
- Conversations read after a month, unanswered questions turned into posts
If your blog works more like a knowledge base, the help center search page shows how the same setup answers from docs and guides, and the pricing page lists the page and message limits for each plan.