Back to Blog
Choosing and Installing the Right Chatbot for Your Site

Choosing and Installing the Right Chatbot for Your Site

Choosing and Installing the Right Chatbot for Your Site

Hands plugging network cable for chatbot setup

For most website owners, a plugin or widget delivers the fastest payoff and takes under an hour to configure. E-commerce brands with a real catalog or WhatsApp ambitions need a more capable, API-based setup. If you run WordPress or a simple business site, install a widget today. If you sell products at volume or want WhatsApp automation, plan a short discovery call with an integration partner before you commit to a tool.

TL;DR:

  • Basic widget setups are quick and inexpensive for simple websites, but larger e-commerce sites need API integration for catalog syncing and automation.
  • Rule-based bots handle FAQs well but struggle with open-ended questions, while AI/LLM options offer better nuance at a higher setup cost.
  • Proper platform-specific installation, including testing mobile responsiveness and privacy compliance, is crucial before launching.
  • Integrating with WhatsApp Business API, CRMs, and analytics significantly enhances chatbot functionality and operational value.
  • Custom AI solutions become justified only when product complexity or omnichannel needs exceed the capabilities of off-the-shelf widgets.

Table of Contents

What Kind of Chatbot Site Web Setup Fits Your Business?

Every chatbot on a website today runs on one of three engines: rule-based, AI/LLM, or a hybrid of both. Picking wrong wastes months of tweaking, so it pays to understand the trade-offs before you touch a plugin marketplace.

Rule-based chatbots follow decision trees. A visitor clicks a button or types a keyword, and the bot matches it to a pre-written response. There’s no ambiguity and no risk of the bot inventing an answer, which makes this approach solid for FAQs, store hours, shipping policies, and simple lead capture forms. The catch: anything outside the scripted paths dead-ends fast, and every new question means manual rebuilding.

AI/LLM chatbots use language models to understand free-form questions and generate natural responses, even ones nobody scripted. They handle nuance, follow-up questions, and multi-turn conversations far better. The trade-off is control. A generative model can occasionally drift off-brand or give a slightly wrong answer if it isn’t grounded in your actual product data.

Hybrid chatbots combine both: rules handle the predictable, high-stakes paths (returns, order status, pricing), while an AI layer takes over open-ended questions and smooths the conversation between rule branches. Most serious e-commerce and support deployments now land here, because it balances accuracy with flexibility.

Match the approach to the job:

  • Customer support FAQs → rule-based is often enough, especially for a low-traffic site with repetitive questions.
  • Sales and product discovery → AI/LLM or hybrid, since shoppers ask open-ended questions (“what’s good for oily skin?”) that scripts can’t anticipate.
  • Lead capture and qualification → rule-based works fine if your funnel is simple; hybrid helps when you want the bot to handle objections too.
  • Order tracking and account questions → hybrid, since it needs both a strict data lookup and natural conversation around it.

Cost and speed to launch follow a predictable curve. Rule-based bots ship in days and cost little to maintain. AI/LLM and hybrid systems take longer to configure properly (writing guardrails, testing edge cases) but scale much better as your traffic and question variety grow.

How Do You Add a Chatbot to a WordPress, Shopify, or Custom Site?

The mechanics differ by platform, but the sequence is the same everywhere: pick a tool, drop in the integration, configure the basics, then test before you go live.

Installing on WordPress

  1. Pick a plugin from the WordPress repository. Search “chatbot” or “live chat” in your dashboard, or browse listings like the Tidio live chat plugin, which is one of the more widely installed options for WordPress sites.
  2. Install and activate it like any other plugin, then open its settings panel, usually added as a new menu item in your WordPress sidebar.
  3. Place the widget. Most plugins let you choose bottom-right or bottom-left placement, and some let you exclude specific pages (checkout pages, for instance, where a popup can be distracting).
  4. Configure language and greeting. Set the default language to match your audience, and write a greeting that sounds like your brand, not the vendor’s default “Hi there! How can I help?”
  5. Add GDPR cookie consent. If the chatbot sets cookies or logs conversation data, your consent banner needs to cover it explicitly. Most European visitors expect this, and skipping it creates compliance risk.
  6. Connect your knowledge base or FAQ page if the plugin supports importing content, so answers stay grounded in what you’ve actually published.

For a WordPress or WooCommerce store specifically, a dedicated WordPress chatbot integration handles catalog syncing and order lookups that generic plugins often can’t.

Installing on a custom HTML or JS site

Sites without a CMS need a manual snippet install, but it’s still a five-minute job for anyone comfortable editing HTML:

  1. Sign up with your chosen chatbot vendor and generate your unique JavaScript snippet from their dashboard.
  2. Paste the snippet just before the closing </body> tag on every page where you want the widget to appear.
  3. Load the script asynchronously (most vendor snippets already use the async or defer attribute) so the chatbot doesn’t block your page’s initial render and hurt load times.
  4. Check keyboard accessibility. A chatbot widget should be reachable via Tab key navigation and announce itself properly to screen readers. Many default vendor widgets fail this out of the box, so test it manually.
  5. Confirm the widget renders correctly on mobile, since a poorly sized chat window can cover your entire mobile screen.

Vendor documentation for tools that support multiple site builders, including Elfsight’s setup guides for WordPress, Shopify, Wix, and Webflow, walks through platform-specific quirks worth checking before you paste anything live. General install guidance, including choosing a platform and configuring the conversation flow in the vendor’s dashboard before touching code, applies whether you’re on WordPress or a hand-coded site.

Installing on Shopify or WooCommerce

E-commerce platforms need one extra step beyond a basic install: connecting the product catalog so the bot can answer “is this in stock?” or “what sizes do you have?” without a human.

  1. Install the chatbot app from the Shopify App Store, or the plugin from the WooCommerce marketplace.
  2. Authorize the app to read your product catalog, orders, and customer data (most apps request these permissions during setup).
  3. Map key intents to catalog fields: price, stock status, variants, and shipping estimates so the bot pulls live data instead of guessing.
  4. Test a handful of real product queries before launch. Ask about a discontinued item, a backordered SKU, and a bundled product, since these edge cases expose gaps fast.

Pro Tip: Run your first ten test conversations pretending to be your most confused customer. Ask vague questions, misspell product names, and switch topics mid-sentence. If the bot handles that gracefully, real visitors will have an easier time than your test run.

Final testing checklist before going live

Before you announce the chatbot to real visitors, run through this list once:

  1. Send five sample questions covering your top support topics and confirm accurate answers.
  2. Test the escalation path: does an unanswerable question correctly hand off to a human or capture contact details?
  3. Check mobile rendering on at least two screen sizes.
  4. Verify GDPR consent language appears before any conversation data gets logged.
  5. Confirm the widget doesn’t overlap your cookie banner, footer links, or checkout button.

How Do You Design Chatbot Conversations That Actually Work?

A chatbot’s technical setup matters less than what it says once someone starts typing. Good conversation design starts with data, not guesswork.

Pull your top twenty support tickets, live chat transcripts, or search queries from the last few months. Group similar questions together. Each group becomes an “intent,” and the specific details inside each question (“order #4521,” “size medium,” “next Tuesday”) become “slots” the bot needs to capture to give a useful answer. A returns question, for example, might need the order number and the reason for return as its two slots before the bot can act.

Hand sorting cards for chatbot intent grouping

Once intents are mapped, write answer templates that sound like an actual employee, not a legal disclaimer. Keep each answer under three sentences where possible, and build in guardrails for generative responses: a explicit instruction to the AI layer never to quote a price, policy, or promise the business hasn’t actually made. This matters more than it sounds. An AI chatbot that improvises a refund policy on the spot can create a real liability.

No conversation design survives contact with real users without a fallback plan. Build these into every flow:

  • A clear “I don’t understand that” response that doesn’t loop the visitor back to the same unhelpful message twice.
  • An escalation trigger after two failed attempts, routing to a human agent with the full conversation history attached so the visitor doesn’t repeat themselves.
  • A visible way to reach a human at any point, not just after the bot fails.
  • Logging of every unresolved question so you can add it to your intent list later.

Pro Tip: Review your “unresolved” log weekly for the first month, then monthly after that. The questions that keep showing up unanswered are your highest-value additions, often more useful than any generic chatbot template.

Treat your answer templates the way you’d treat ad copy. Run two versions of a key response (a shorter, more direct answer versus a warmer, more conversational one) and track which one leads to more completed conversations or fewer escalations. Small wording changes routinely move the needle more than people expect.

Which Integrations Actually Move the Needle?

A chatbot that only lives on your website leaves value on the table. The integrations you choose determine whether it’s a nice-to-have widget or a real operational tool.

WhatsApp Business API is the single most requested upgrade among e-commerce merchants, and for good reason: it enables automated order confirmations, shipping updates, and two-way support inside the app most of your customers already have open. It’s a different thing entirely from a simple click-to-chat button, though. Shopify’s documentation on WhatsApp Business for e-commerce makes clear that full automation requires the API tier, not the free consumer app, and that means business verification through Meta and, usually, a third-party provider to manage the technical connection.

  • CRM and helpdesk integration pushes every chatbot conversation into your existing sales or support pipeline, so a lead captured at 11 p.m. shows up in your CRM by morning instead of sitting in a chat log nobody checks.
  • Catalog and order sync lets the bot answer “where’s my order?” or “do you have this in blue?” using live data instead of a static script, which is often the single biggest driver of chatbot usefulness for online stores.
  • Analytics integration tracks which questions get asked most, where conversations drop off, and how many resolve without human help.

The category of tools handling these connections is genuinely broad. G2’s chatbot category page lists dozens of vendors with wildly different integration depth, from basic widget-only tools to platforms that connect deeply into commerce and messaging stacks.

On privacy, don’t treat consent as an afterthought bolted onto launch. Every chatbot that logs a name, email, order number, or phone number is handling personal data, and your consent banner, retention policy, and data export process need to account for that from day one, not after a customer asks what you’re storing.

What Should You Check Before Signing With a Chatbot Vendor?

Vendor selection is where most chatbot projects quietly go wrong. Feature lists all start to look the same after the fifth demo, so the real evaluation has to happen underneath the marketing page.

Run through these criteria with every vendor you seriously consider:

  • Language support: does it handle your actual customer base’s languages natively, or is it running everything through a generic translation layer?
  • Integration depth: can it connect to your actual CRM, catalog, and helpdesk, or only a shortlist of “supported” tools?
  • Data access and export: can you pull your conversation history and customer data out cleanly if you switch vendors later?
  • Service-level agreement (SLA): what uptime guarantee do they commit to in writing, and what happens if they miss it?
  • Customization ceiling: can you fully rewrite the bot’s tone and logic, or are you locked into their templates?
  • Pricing model: is it a flat SaaS tier, a per-message fee, or a hybrid that could spike unpredictably with traffic?
  • Compliance: do they store data in a way that satisfies your obligations under GDPR, including retention limits and the right to erasure?

Before signing anything, ask for specifics on paper, not verbal assurances:

  1. Request the exact uptime percentage guaranteed in the SLA, not just “high availability.”
  2. Ask what the average response time is for their own support team when something breaks.
  3. Confirm you can export full conversation logs and customer data in a standard format at any time.
  4. Get a written breakdown of what happens to your data and bot configuration if you cancel.

Watch for two common red flags. First, vendors who quote a low base price but charge steeply for every integration beyond the basics, turning a $50-a-month tool into a $400-a-month tool once you connect your CRM. Second, contracts with vague data-export language, which usually means “you’re not getting your conversation history back easily” if you leave.

What Does a Chatbot Actually Cost?

Pricing splits into three rough models, and each scales differently depending on how your business actually grows.

Flat SaaS tiers charge a fixed monthly fee for a set number of conversations or seats, which suits small sites with predictable, modest traffic. Per-message or per-active-user pricing scales with usage, meaning your bill grows alongside your traffic, good for a site that isn’t sure yet how much volume it will get, but riskier if you have a viral spike. Custom project plus monthly subscription is the model for deeper builds: an upfront development fee for a tailored agent, followed by an ongoing monthly cost for hosting, maintenance, and usage.

WhatsApp adds its own line item. Because the API requires a registered business account, most merchants go through a third-party provider, and message fees vary by country and message category (marketing versus transactional messages are usually priced differently). Budget for these as separate, usage-based costs rather than assuming they’re bundled into your chatbot subscription.

Realistic budget bands, based on the scope of the deployment:

  • Small business site: a WordPress plugin or basic widget typically runs from free tiers up to a modest monthly SaaS fee, often affordable enough that it’s a rounding error in the marketing budget.
  • Growing e-commerce store: catalog-aware bots with WhatsApp and CRM integration cost meaningfully more, since you’re paying for both the platform and the integration work to connect your systems.
  • Enterprise or omnichannel brand: custom-built agents with deep automation, multiple language support, and dedicated infrastructure represent a real project investment, usually structured as an upfront build fee plus ongoing subscription.

What Belongs on Your Launch Checklist?

A chatbot that launches untested finds its bugs live, in front of real customers, which is the worst possible place to discover them.

  1. Run functional tests across your top twenty expected questions, confirming each returns an accurate, on-brand answer.
  2. Test edge cases deliberately: misspelled product names, questions in a second language your site supports, and deliberately vague phrasing.
  3. Check multilingual handling if you serve more than one language market, since a bot that silently defaults to English for a French-speaking visitor creates a bad first impression fast.
  4. Run an accessibility pass: keyboard navigation, screen reader announcement, and color contrast on the widget itself.
  5. Confirm your privacy and consent language covers the chatbot explicitly, including what’s logged, how long it’s retained, and how a visitor can request deletion.

Once live, track a small set of numbers rather than drowning in dashboards. Resolution rate (conversations closed without human escalation), average handling time, and escalation rate are the three that tell you the most about whether the bot is actually working. A broader look at measurable chatbot benefits for customer service covers how these metrics typically shift after a few months of tuning.

Pro Tip: Review your unresolved-conversation log every week for the first month. That short habit alone catches more real problems than any dashboard metric, because it shows you exactly where visitors got stuck in their own words.

Review logs regularly at first, then reduce frequency once the bot’s resolution rate stabilizes above your target. Optimization isn’t a one-time setup task. It’s a small recurring habit, roughly the same amount of effort as checking your site’s analytics.

When Does a Custom AI Agent Beat an Off-the-Shelf Widget?

Most of the advice above assumes a widget or plugin is the right call, and for many websites, it is. But that’s not always the honest answer.

A custom AI agent earns its cost when your product catalog is genuinely complex (configurable products, bundles, region-specific inventory), when you need real omnichannel consistency across web, WhatsApp, and a CRM simultaneously, or when you’re automating something beyond customer-facing chat, like internal order routing or predictive restocking alerts. In those cases, a generic widget hits its ceiling fast, and you end up paying for integrations piecemeal that a custom build handles as one coherent system from day one.

A widget or plugin is genuinely sufficient when your questions are repetitive, your catalog is simple, and your main goal is deflecting basic support volume. Don’t let anyone talk you into a six-figure custom build to answer “what are your store hours?”

The honest dividing line is complexity, not size. A small store with a genuinely complicated inventory setup might need a custom agent sooner than a mid-sized business with a simple product line. Botiqueai typically sees the clearest returns when a client is already juggling three or more disconnected tools (a CRM, a catalog, a support inbox) and wants one intelligent layer sitting across all of them.

— Botiqueai

Get Aria Running on Your Site or Store

If you’ve read this far and you’re leaning toward the deeper end, that’s exactly the gap Aria closes. Aria is Botiqueai’s AI chatbot built for websites and e-commerce specifically, meaning it comes catalog-aware out of the box instead of needing months of custom integration work to answer basic product questions.

Aria handles the buyer jobs that generic widgets tend to struggle with: live product and stock lookups, order status without a human touch, and conversation handoff to your team when a question genuinely needs a person. For WordPress or WooCommerce stores specifically, the dedicated WordPress and WooCommerce integration connects directly to your existing catalog and checkout flow. If your needs go beyond chat entirely, into internal workflow automation or backend process work, Botiqueai’s custom AI automation services cover that ground too.

Hands connecting network device for chatbot integration

Start with a discovery call. Botiqueai will walk through your current setup, flag where a widget is already enough, and scope a pilot only where a custom build genuinely pays for itself.

Sources

© 2026 BotiqueAI — Reproduction prohibited without attribution.