A small WooCommerce store should pick the chatbot that answers your own order and returns questions correctly in a 20-question test, then compare pricing models: free tier and flat monthly fee for low volume, per-resolution pricing if conversations are heavy. Do not pay for any bot before it passes that test.
Features lists look the same across tools: "trained on your store", "handles returns", "hands over to a human". What differs is how each prices a conversation, whether it can see a customer's order, and what it does when it does not know. Those three things decide whether a bot saves you time or creates refund arguments.
What "agent" means for a WooCommerce store
There are three different things sold under this label, and they solve different problems.
- A policy chatbot reads your shipping, returns and product pages and answers questions from them. It cannot tell a customer where their parcel is.
- An order-aware agent connects to your store data so a verified customer can ask about their own order. It carries more risk and more set-up.
- A store-management agent lets you, the owner, talk to your own store through an assistant such as Claude or ChatGPT, to write product copy or spot issues. It is not a customer chatbot at all.
StoreAgent, a WooCommerce plugin, is a mix of the first and third: it answers shoppers from store data, writes product content in bulk, and connects Claude or ChatGPT to your store through MCP, a standard way for an assistant to talk to other software. WooCommerce's developer post on version 11.3 says the old WooCommerce MCP endpoint is removed and that connections now go through the separate WordPress MCP Adapter plugin with an application password, which matters if you are following an older guide that sets MCP up the old way.
Know which of the three you need. Most small stores that ask this question want the first: fewer "what is your returns policy" emails. Order lookups come second, once the policy answers are safe.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowFour options compared on how they charge
Prices below are USD list prices fetched from each vendor's page in November 2026. They change, so confirm on the live page.
| Option | What it is | How it charges | Free route |
|---|---|---|---|
| StoreAgent | WooCommerce AI agent plugin | Lite free; Growth $19 a month (500 AI chat replies); Pro $49 (5,000); Business $249 (50,000). AI usage included in every plan. | Lite: 50 AI chat replies and 200 AI content tasks, one time |
| Tidio | Chat and helpdesk with Lyro AI agent | Free $0; Starter $24.17 a month; Growth from $49.17; Lyro add-on from $32.50 a month for 50 conversations | Free plan with 50 Lyro conversations for life |
| Gorgias | Helpdesk with an AI Agent | Starter $10 a month for 50 tickets; Basic $50 annual or $60 monthly for 300; AI Agent interactions $0.90 each on annual plans or $1.00 on monthly plans, on top | No free plan; a 7-day trial |
| ChatBot.com | Chatbot builder | Essential $19 per user a month annual ($25 monthly) with 10 AI resolutions a month; Growth $79 annual ($99 monthly) with 200; extra resolutions $49.50 per 50 | 14-day trial, no permanent free tier |
A few notes on reading that table. Gorgias bills a ticket for each conversation that includes a reply from a human, a rule or its AI Agent, and an "automated interaction", a request resolved without a human, is billed on top at the rates shown; if a human replies within 72 hours the interaction is billed as a ticket instead. Tidio defines a Lyro conversation as a customer interaction started through any channel connected to Tidio, and repeated replies within the same issue still count as one. Its Lyro add-on starts at $32.50 a month for 50 conversations, with larger tiers up to 1,000 a month whose prices the page does not show. StoreAgent says AI usage is included, so there is no separate AI bill; its free allowances are one-time, not monthly. Tidio's pricing page names Shopify but not WooCommerce; its WordPress.org plugin listing describes a WooCommerce live chat integration that shows your team visitor carts and order history and sends product cards, but does not tie that store data to Lyro's answers. Gorgias lists WooCommerce among supported platforms without describing how deep it goes, and ChatBot.com's pricing page does not mention it, so ask each vendor what data actually flows from WooCommerce into the AI's replies.
Three pricing models and who they suit
Flat monthly. You pay a set fee for a bundle. Easy to budget, and best when volume is steady and low. The risk is hitting a limit mid-month.
Per conversation. Tidio's Lyro works this way. Costs follow traffic, which suits a store with seasonal spikes, but a viral post or a bot loop can run up a bill.
Per resolution or per ticket. Gorgias charges for tickets and for AI-resolved interactions. This rewards a bot that really resolves problems, but find out what counts as resolved. Gorgias, for example, defines an automated interaction as a request fully resolved without a human agent, so ask how the tool treats a customer who simply gives up and leaves.
Worked cost example for a small store
An illustrative store: 600 orders a month, 150 customer chats a month, of which about 60% are policy questions (shipping time, returns, sizes) and 40% are about a specific order. All figures are invented to show the method.
- Workload: 150 chats at six minutes each is 15 hours a month. At $20 an hour of staff time, $300.
- Bot-answerable: 60% policy questions is 90 chats. If the bot resolves 70% of those without a person, that is 63 chats and 6.3 hours saved, about $126.
- Cost on a per-conversation plan: Tidio's entry Lyro tier works out at $0.65 a conversation ($32.50 for 50). If 150 conversations cost the same rate, that is $97.50, leaving $28.50 of net saving a month before you count set-up time, which is thin. Tidio does not publish its larger tiers' prices, so ask for the real figure at your volume.
- Cost on a flat plan: StoreAgent's $19 Growth plan includes 500 AI chat replies a month, enough for 150 chats of up to three replies each, and leaves about $107 saved a month.
The same store saves more on a flat-fee tool than on a per-conversation one, because volume is low and steady. With 1,500 chats a month the arithmetic reverses: per-resolution pricing at about $1 an interaction could cost $1,000 or more, and a flat plan with a higher cap wins. This is why you price the bot against your own chat count, not the vendor's headline.
The 20-question policy test
Before paying, build twenty questions from your real inbox and run them through each bot's free tier or trial. Copy real customer wording, including misspellings and vague phrasing.
- Pull 20 real questions from the last three months: eight about returns and refunds, four about shipping times and costs, four about products, four about orders and tracking.
- Write the correct answer for each, from your own policy, before you test.
- Ask the bot each question in a clean chat and record the answer.
- Score each as correct, partly correct, wrong or declined. A decline that hands over to a person counts as correct for questions your policy does not cover.
- Add five trick questions: a request for a refund outside your window, a request to waive shipping, a question about a product you do not sell, an angry message and a question in another language.
Pass mark: at least 18 of 20 correct or properly declined, and zero invented policies. A bot that gets 16 right and invents two refund promises is worse than one that gets 14 right and declines six, because customers will quote an invented promise back at you.
What a failed answer looks like
Illustrative transcript from a test. The customer asks: "Can I return a candle I burned once?" Your policy says unused items only. A bad bot replies: "Yes, you can return any item within 30 days for a full refund." The fault is that it quoted a generic 30-day rule it learnt elsewhere. A good bot replies: "Our policy covers unused items in original packaging within 30 days. A burned candle would not qualify, but I can pass your message to the team if you would like them to look." Score the first as wrong even though some of its words are right.
Make the source pages bot-ready first
Most wrong answers come from unclear source pages, not from the AI. Before testing, tidy your returns, shipping and FAQ pages so each answer is stated once, with the exact numbers: days, costs, conditions and exclusions. If your pages contradict each other, the bot will pick one. The tutorial on building the FAQ your chatbot needs walks through the writing, and the one on whether a chatbot can handle order tracking and returns covers the limits on the order side.
A before and after to illustrate: before, the returns page says "Returns accepted within a reasonable time." After: "Returns accepted within 30 days of delivery for unused items in original packaging. Customers pay return postage unless the item is faulty. Refunds go to the original payment method within 5 working days of receipt." The second can be answered correctly by any bot; the first cannot.
What the bot can read from a WooCommerce store
A bot is only as good as what it is fed, so list the sources before you connect anything. For a typical store they are: product pages (titles, descriptions, attributes, variations and stock status), your shipping page and the shipping zones set in WooCommerce, the returns and privacy pages, FAQ content and, if the tool supports it, order data. Decide which of these the bot may use. Leave out draft products, private pages, old campaign pages and supplier notes, because a tool that indexes everything on a WordPress site will happily quote a page you forgot was public.
Product variations deserve a separate check. If a t-shirt has sizes and colours as variations, ask the bot "do you have the blue one in medium?" and compare with the live stock. Some tools read the parent product only and answer from its description, which is wrong whenever one variation is sold out.
Order lookups: the part that needs care
If a bot can read orders, the biggest risk is showing one customer another's data. Ask every vendor how the shopper is verified: order number plus email is a common minimum. StoreAgent's page lists orders as "read on request" among what its agent can see, and its MCP connection lets the owner's assistant ask about orders, but it does not state whether the shopper-facing chat can show a customer their own order status; ask before assuming it does. For a small store it is reasonable to start with policy answers only and have the bot collect the order number and email, then pass a ticket to a person.
A filled-in scoring sheet
This is an illustrative result for one bot on a store selling candles and homeware. Keep a sheet like it for each tool you test, so the decision rests on the same questions.
| Question group | Asked | Correct | Declined well | Wrong |
|---|---|---|---|---|
| Returns and refunds | 8 | 6 | 1 | 1 (quoted a 14-day window; ours is 30) |
| Shipping times and costs | 4 | 4 | 0 | 0 |
| Product facts | 4 | 3 | 0 | 1 (wrong wick type) |
| Orders and tracking | 4 | 0 | 4 | 0 |
| Trick questions | 5 | 3 | 1 | 1 (agreed to waive shipping) |
Totals: 16 correct, 6 declined well, 3 wrong out of 25. By the pass mark above the bot fails, because two of the wrong answers are invented promises. The useful part is the cause. The 14-day window came from an old FAQ entry that the store then fixed, and the shipping waiver came from a setting that let the bot offer goodwill gestures. Both are fixable, so re-run the sheet after fixing and score again before deciding.
Handing a stuck chat to a person
Check three things. The bot must know when to stop: complaints, refund disputes and anything involving a payment. The handover must carry the transcript, so a customer does not repeat themselves. And a person must be reachable: if you are out of the office, say so and set an expected reply time. StoreAgent says shoppers can be escalated to your team on request, and the major helpdesks do this as standard; test it by asking to speak to a person in your trial.
Seven jobs a bot does well, and two it should not do
- Shipping times and costs from your published table.
- Returns policy questions, quoting your exact wording.
- Sizing and product facts from your product pages.
- Opening hours and contact details.
- Collecting details for a return: order number, item and reason, before a person approves.
- Out-of-hours cover that says when a person will reply.
- Back-in-stock questions pointing to your notification sign-up.
Two it should not: approving refunds or exceptions, and giving advice that carries risk, such as allergens or safety claims on products. Leave those to a person, as the questions to ask a chatbot vendor tutorial also suggests.
Customer wording: a before and after for the bot's tone
Bots default to cheerful and long. Customers with a problem want a short, plain answer and a route to a person. Most tools let you set a tone or instruction. A short instruction you can adapt:
Answer in two or three short sentences using only the store pages provided. Quote the exact number of days or the exact cost. If the answer is not on those pages, say you are not sure and offer to pass the message to the team. Never promise a refund, a replacement or a delivery date. Do not apologise more than once.Before, an illustrative reply: "Great question! We absolutely love helping with returns! Typically you may be able to return items within a certain period, so do pop back if you need anything else!" After: "Returns are accepted within 30 days of delivery for unused items in original packaging. If you would like to start one, send me your order number and I will pass it to the team." The second is shorter, specific and safe.
Telling shoppers it is a bot, and where chats are kept
Tell shoppers they are talking to an AI, and put a line in your privacy policy about chat transcripts and which provider processes them. If you have customers in the EU, data-protection law such as the GDPR applies to what they type into the chat, so check where the vendor stores transcripts and whether you can delete them on request. Customers also paste personal details into chats unprompted; ask the vendor how long transcripts are kept.
A mistake that cost a refund war
An illustrative case: a store sets up a bot that learnt from an old promotion page offering free returns during a promotion. Weeks after the promotion ended, the bot kept telling customers returns were free. Thirty customers returned items at the store's cost before anyone noticed. The cause was a stale page in the bot's sources. The fix is a monthly review: read twenty random transcripts, check any page changed in the month is re-read by the bot, and delete old promotional pages from its sources.
How to check it is working after launch
- Week one: read every transcript daily. Mark wrong answers and fix the source page, not the bot.
- Week two to four: read a sample of ten a day and track the share handed to a person.
- Monthly: compare the number of support emails before and after, and the count of chats that ended with a complaint.
If support emails did not fall, the bot is either not being used or not resolving. A bar on the page that nobody opens is a common cause: move the chat button, and put a link to it on your returns page.
Which to choose, by store type
| Store | Start with | Why |
|---|---|---|
| Under 50 chats a month | Test on a free allowance (StoreAgent Lite or Tidio Free), then the cheapest flat plan | Both free allowances are one-time, so they prove the bot rather than run it |
| 50 to 300 chats, mostly policy questions | Flat low-cost plan | Predictable fee and low risk |
| Steady email and chat from many channels | A helpdesk with an AI Agent | One inbox for email, chat and social |
| Owner wants help with product copy too | A store agent with content tools | Shares one tool and one bill |
Whichever you choose, put a review date in the diary 30 days after launch and be ready to switch off a bot that fails the test. If you want a second pair of eyes on your questions and set-up, that is what my AI implementation consultation covers.
WooCommerce chatbot questions answered
Can a chatbot look up a customer's order in WooCommerce?
Only if it connects to your store data with a secure lookup. A bot trained on your policy pages can explain returns but cannot say where a parcel is. Ask each vendor, in writing, whether the shopper-facing chat can read order status and what the customer must enter to verify who they are.
Is a free chatbot plan enough for a small store?
For testing, yes. StoreAgent's free tier gives 50 AI chat replies and 200 content tasks as a one-time allowance, and Tidio's free plan gives your first 50 Lyro conversations free as a lifetime allowance. Those run out quickly on a store with steady traffic, so use them to test, not to run.
What does a chatbot do when it does not know the answer?
A well-set-up one declines and hands the chat to a person. Test this on purpose with questions your policies do not cover, and check that the bot does not invent delivery times or refund promises.
Do I need a developer to install one?
Most are WordPress plugins or a pasted script and need no code. Connecting order data or a returns system is more involved and may need an adviser or developer.
Should the bot be allowed to issue refunds?
Start with no. Let it answer questions and collect details such as order number and reason, then pass the case to a person who approves the refund. Widen its powers only after weeks of clean transcripts.
Further reads
- Best AI Chatbots for Small Shopify Stores (2026) — The same comparison for stores on Shopify.
- How Much Does It Cost to Add AI to a Shopify Store? — A cost view of adding AI to a store.
- AI Receptionist or Chatbot: Which Suits a Cleaning Company? — When a receptionist tool beats a chatbot for enquiries.
- Zapier vs Shopify Flow: Which Automation Tool Suits a Small Store? — Automate follow-up tasks around support chats.
- How to Check AI Is Doing Good Work, Not Just Fast Work — A method for checking AI answers before customers see them.
- AI Fit Shoes Plugin: Can a Size Finder Cut Returns for a Shoe Shop? — How shoe size finders work, what data they need, and a before-and-after method to test whether one cuts your returns before you pay.
- AI Chatbot for Agriculture: A Farm Supply Business's Setup and Cost — What a feed, seed and fencing seller pays for an AI chatbot, which seasonal and delivery questions it should take, and where staff step in.
- AI Chatbot for WooCommerce: Costs, Order Lookup and a Test Script — Plugin or hosted widget, per-conversation pricing, order lookup with read-only keys, and a policy test script to run before launch.
- Best AI Ecommerce Store Builder: Shopify, WooCommerce or BigCommerce? — Shopify has Sidekick, BigCommerce has Companion, WooCommerce has plugins only. What each built-in AI does, and what the plans cost.
- Best AI for Ecommerce Operations: Orders, Stock, Returns, Suppliers — Back-office AI for an online store: Sidekick for orders, Gorgias or Tidio Lyro for returns, and an assistant for supplier emails and reorder checks.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: StoreAgent home and pricing pages, Tidio pricing page and WordPress.org plugin listing, Gorgias and ChatBot.com pricing pages, and the WooCommerce 11.3 MCP developer post (all fetched November 2026).