Last updated: 31 July 2026
BotSupport AI Chat is a fully managed AI shopping and support assistant that sits on your ecommerce website. Your customers type what they actually want, in their own words, the way they would ask a person or ask ChatGPT. The assistant understands the question, searches your real product catalogue, and replies with a short, human answer plus product cards showing live prices and stock. It also answers delivery, returns, sizing and policy questions, tracks orders securely, and captures enquiries when a human needs to pick things up.
It is powered by OpenAI. It is built, configured, tuned and maintained by us. You do not build flows, write scripts, train a bot or manage anything.
Price: £79 per month during early access, locked in forever. No contract. Cancel any time. When early access closes the price becomes £149 per month, and everyone who joined at £79 stays at £79 for as long as they stay with us.
For twenty-five years online shopping worked the same way. You landed on a homepage. You picked a category. You picked a sub-category. You applied four or five filters. You scrolled a grid. You opened six tabs. You compared. Somewhere in there, a lot of people gave up.
That model made sense when the alternative was a phone call. It does not make sense any more, because a very large number of people now have a different daily habit.
ChatGPT passed 900 million weekly active users in early 2026 and crossed one billion monthly active users in mid-2026, making it the fastest application in history to reach that mark. It is on track to pass a billion weekly users before the end of the year. Something in the order of one in eight people on earth now opens a chat box, types a messy, half-formed, entirely human question, and gets a straight answer back.
That habit does not switch off when someone lands on your shop. They arrive already trained. They already expect to be able to say "I need something machine washable for a south-facing bay window, budget about £40 a metre" and get an answer. Then they hit a category grid and a filter sidebar, and the experience feels twenty years old, because it is.
Site search is where the gap shows up first, and it is expensive.
Read those three facts together and the picture is blunt. The people using your search box are your highest-intent visitors. Your search box fails one in eight of them. Four in five of those leave, and most do not come back.
Traditional search fails because it matches strings. It does not know that "emerald" is a green, that a "settee" is a sofa, that "something for a toddler's room" implies washable, or that "the one in the photo on your Instagram" is a real query a real person will type. It cannot handle a sentence. It can only handle a keyword, and customers stopped thinking in keywords the day they started using ChatGPT.
AI assistants are now a genuine referral channel, and the visitors they send are unusually good. ChatGPT accounts for the overwhelming majority of AI referral traffic to websites, AI-referred sessions have grown several hundred percent year on year, and multiple 2026 datasets show AI-referred traffic converting substantially better than traditional organic, because the buyer arrives pre-informed and already recommendation-primed.
So the shift runs in both directions. Customers arrive from a conversation, and they expect to continue one. A category grid is not a conversation.
The conversational commerce market sits at roughly $10 to $14 billion in 2026 and is projected well past $40 billion into the 2030s. McKinsey has projected AI agents facilitating trillions of dollars of global commerce by the 2030s. In vendor surveys, 87% of companies say they will keep investing in conversational commerce over the next twelve months, and 79% report increased sales and conversions.
Those are industry numbers, not our promises. We include them because the direction of travel matters more than any single figure: the storefront is becoming a conversation, and the shops that get there early get the compounding advantage.
BotSupport AI Chat is a chat assistant that appears on your ecommerce website. It has three jobs.
1. It sells. It understands what a customer is looking for, finds the products in your catalogue that genuinely match, and presents them as clickable product cards with live prices and stock. It behaves like a good shop assistant who has read every product page, remembers every line, and never gets tired at 11pm on a Sunday.
2. It supports. It answers the questions your inbox is full of. Delivery times, returns policy, sizing, materials, care instructions, made-to-measure processes, lead times, what happens if something arrives damaged. It answers from your own published information, not from generic internet knowledge.
3. It captures. When a question genuinely needs a human, or the customer wants a quote, a callback or a sample, it collects their details and delivers them to you straight away.
It is powered by OpenAI. It runs on your site behind your branding. And, critically, it is fully managed, which is the part most of this page is about.
This is what a shopper actually sees and does.
A small chat launcher sits in the corner of every page. You choose the side (right, left, or centred), the label, the colours and the icon. It can carry a short attention message to draw the eye, for example "Ask us anything" or "Find your fabric". It is deliberately quiet. It does not fire a pop-up at someone three seconds after they land.
Opening the chat shows a greeting you write, plus a set of starter prompts. Starter prompts matter more than they look, because they teach the customer what the assistant can do in the two seconds they will give it. Good ones are specific to the shop:
The customer types in whatever way they like. Full sentences, fragments, typos, brand names, a colour and a room, a size code, a part number, a budget. This is the whole point. There is no query syntax to learn and no filter to apply.
The assistant streams a reply back, word by word, the way ChatGPT does, so it feels immediate rather than making the customer stare at a spinner. While it is looking something up it shows a status line, for example "Searching products", so the wait feels like work being done rather than a page hanging.
The reply is short. It leads with a direct answer, then shows up to four product cards, each with the product image, name, live price, stock state and a "View product" button in your brand colours. It does not pad, it does not write an essay, and it does not list ten near-identical options to look thorough.
Under the reply, the assistant offers up to three contextual quick replies, single-tap follow-ups relevant to what was just said, for example "Show me cheaper options", "Is it machine washable?", "What's the delivery time?". Most customers tap rather than type, and each tap is another step towards a purchase.
If the customer wants a person, a quote, a sample or a callback, the assistant offers a short form. You choose the fields. Submissions arrive by email and can be pushed straight into your CRM or workflow tool. The full conversation goes with it, so whoever picks it up already knows the context.
The customer never leaves your website, and they never lose their place.
This is the core capability and the hardest one to get right.
A customer says "something in a deep green, geometric, hard-wearing enough for a dining chair". There is no such filter on your website. The assistant expands that into the colours, patterns, materials and synonyms it needs to search for, looks through your real catalogue, reads the results, and returns the ones that genuinely match. If nothing genuinely matches, it says so and offers the nearest alternatives, clearly labelled as alternatives.
That last behaviour is a deliberate design decision, and it is worth dwelling on. A chatbot that confidently presents a beige stripe as the emerald geometric you asked for is worse than no chatbot at all, because it destroys trust in one turn and the customer leaves for good. Ours is built to admit a miss.
It also handles exact requests exactly. If someone types a size code, a part number or a precise product name, they get that product immediately, not a clarifying question and not three sizes they did not ask for.
Product cards show current prices pulled from your store, not a stale copy from whenever the assistant was set up. Where a product has several sizes or options, the card features the sensible default and mentions the others, for example "also available by the half-metre" or "samples are £2.99".
Out-of-stock items are shown, clearly marked "Out of stock", never hidden. In-stock items are listed first. Hiding an out-of-stock product just makes the customer search again, find nothing, and conclude you do not sell it.
Delivery, returns, exchanges, sizing, care, materials, guarantees, lead times, made-to-measure processes, trade accounts, sample policy, what to do about a damaged parcel. The assistant answers from your own published information: your policy pages, FAQs, product information and anything else you want it to know. Where an answer has a relevant page on your site, it links to it, so the customer can verify.
"Where is my order?" is the single most common message most ecommerce businesses receive. Industry reporting puts WISMO at 20 to 50% of all support tickets in normal trading, rising past 50% in peak season, and well-implemented AI handles 90%+ of those queries without a human.
BotSupport handles it, and it handles it safely. Order data is never handed out on the strength of an order number alone. The customer proves control of the email address on the order using a one-time code sent to that address, and only then is order information released. The verification is enforced by our systems before any order data enters the conversation, and it stays valid for a short window so follow-up questions in the same chat do not force the customer to re-verify. If you prefer a lighter flow, an order-number-plus-email match is available instead.
This is available on both WooCommerce and Shopify.
You define the fields: name, email, phone, free text, dropdowns, checkboxes, whatever your process needs. Submissions arrive by email and can be sent onward to a CRM, helpdesk or automation tool. The conversation transcript is attached, so your team is not starting cold.
For shops where every enquiry matters more than every order (made-to-measure, trade, high-ticket, B2B), this is often the feature that pays for the subscription on its own.
If someone is on a product page and asks "does this come in blue?", the assistant knows what "this" refers to. If they clearly mean something else, it ignores the page context. Small thing, large effect on how human it feels.
For businesses with calendars rather than catalogues, holiday lets, venues, classes, courses and events, the assistant can read live availability and answer "are you free the week of the 14th?" or "what's on in August?". Multiple named calendars are supported, so it can talk about specific properties, rooms or venues by name.
If a customer writes in French or German, they get a reply in French or German. There is no separate configuration or translation step.
Optional and off by default. Where it makes sense for your business, the assistant can look beyond your own information. Most shops do not want this, because the whole value is answers grounded in your store, so we normally leave it off and tell you why.
We would rather be straight with you here, because a knowledgebase full of overclaims produces an assistant that lies to your customers.
If any of those five become important to you, tell us. We would rather scope it honestly than sell you something that disappoints.
AI Chat for ecommerce is a conversational interface on an online shop that lets customers ask for what they want in natural language and get real answers, real recommendations and real order information back. It replaces the sequence of category, filter, scroll, compare and give up with a single question and a useful answer.
It differs from the chatbots most people have suffered through. A traditional chatbot is a decision tree: a human wrote every question, every button and every reply, and anything outside the script produces "Sorry, I didn't understand that". An AI Chat has no script. It understands the question, consults your actual product and policy information, and composes an answer.
Three reasons.
Intent is unusually clear. Someone on a shop is not browsing an encyclopaedia. They want a thing, or they want to know where their thing is. That focus makes the assistant far more accurate than a general-purpose bot.
The answers already exist. Your product data, your policy pages, your FAQs and your order system already contain almost every answer a customer needs. The problem was never that you did not know; it was that the customer could not get at it. AI Chat is a better retrieval layer over information you already own.
The economics are immediate. Every recovered failed search is revenue. Every deflected WISMO ticket is time. Every captured enquiry is pipeline. You do not need a philosophical position on AI to see the arithmetic.
We are not going to promise you a specific uplift percentage, because anyone who does is guessing at your catalogue, your traffic and your margins. What we can set out is where the value comes from, so you can do your own maths:
You get a dashboard covering total conversations, total messages, link clicks through to products, leads captured, and, for connected stores, orders and revenue attributed to AI Chat. You can filter by date range, see conversations and leads over time, and read recent conversations and recent leads. Optional monthly summary reports can be emailed to whoever needs to see them.
Reading real conversations is the underrated part. Within a fortnight you will know exactly what your customers cannot find, what they do not understand, and which product pages are missing information. That is a merchandising and copy roadmap you did not have before, and several clients have told us it is worth as much as the chat itself.
No, and it can help. The assistant loads without blocking your page, so it does not damage Core Web Vitals or page speed. It does not alter your HTML, your headings or your internal links, so nothing that Google ranks is touched.
It helps in two indirect but real ways. Engagement signals improve, because visitors who would have bounced from a dead-end search now click through to a product. And the process of setting the assistant up surfaces every question your published content fails to answer, which is a straight list of content gaps to fill, and filling those gaps is exactly what earns citations in AI-generated answers.
WooCommerce runs a very large share of the world's online shops, and it has a particular characteristic: enormous flexibility, and consequently enormous variation. Variable products, custom attributes, per-metre and per-unit pricing, bundles, samples, trade pricing, made-to-measure, custom taxonomies, half a dozen plugins each adding their own product logic.
That flexibility is exactly what generic AI chat tools choke on. A tool that assumes every product is a simple item with one price and one stock number will get a fabric-by-the-metre shop, a coffee roaster selling 250g and 1kg, or a parts supplier with size codes badly wrong. It will feature a £2.99 sample swatch as if it were the product, or a 10-metre trade bundle as the headline price, and every recommendation it makes will look slightly stupid.
We handle this explicitly. Variants can be grouped so one card represents one product page with its sizes listed underneath, or split so each variant is its own card, whichever suits your catalogue. And we choose which size a card should feature: the standard single unit (right for fabric by the metre, so the metre is featured, never the swatch or the bulk roll) or the entry-level "from" price (right for tier-priced goods like coffee, so the 250g bag is featured, not the 1kg). That is a per-shop decision, and it is one we make with you rather than leaving you to discover it from a customer complaint.
A short, light-touch connection between your store and BotSupport. We set it up. It is read-only: we read product and order information, we never write to your store, never change products, never touch orders and never take payments. If you would rather we did not touch the site at all, there is a connection route that needs nothing installed, with a slightly reduced feature set.
Setup on your side is typically one short email exchange and about ten minutes of someone's time.
No. The chat runs on our infrastructure, not yours. Your server does not do the thinking, and your page weight barely changes. On a WooCommerce store already carrying twenty plugins, this is one of the few additions that does not cost you loading time.
Not in the sense you are probably asking. The assistant itself is added to your site with a single line of code, so it works on any WordPress theme, page builder or setup, and it survives theme changes and core updates. The optional connector we install on WooCommerce stores is what lets the assistant see live products and orders.
The Shopify App Store is not short of AI chat apps. That is the problem. Fitting one is easy, and getting one to be genuinely good at your catalogue is not, which is why so many Shopify merchants have an AI chat app installed that nobody has looked at since the trial ended.
Typical Shopify AI chat pricing in 2026 runs from around $32 to $360 per month on entry plans, with per-conversation or per-resolution charges stacked on top, and fully autonomous helpdesks reaching around $500 per month. Several well-known options start at $79 to $100 per month for a capped number of sessions, then bill overages per thousand visitors.
All of them are self-serve. You install, you configure, you write the prompts, you fix the tone, you notice six weeks later that it is recommending discontinued lines, and you go and fix that too. The software was never the hard part. Keeping it good is the hard part, and that is the part they leave with you.
Through Shopify's standard storefront access, using a token you generate in your admin and send to us. We do the rest. Read-only: we read products and prices, we do not modify your store, your orders or your customers.
Yes. It is added with a single line of code and renders above your theme without depending on it. Theme updates, redesigns and headless front ends are all fine. If you replatform your theme entirely, the assistant carries on working.
Yes. Shopify customers can ask "where is my order?" in chat and get a verified answer, without emailing you and without waiting until Monday morning.
It is handled safely. An order number on its own is never enough. The customer proves control of the email address on the order using a one-time code sent to that address, and only then is any order information released. That verification is enforced by our systems before order data ever enters the conversation, so the assistant cannot be talked into disclosing an order by a cleverly worded message.
Given WISMO is 20 to 50% of ecommerce support tickets in normal trading and more than half in peak season, this is usually the feature that gives you the most time back.
It means we do the work. All of it. Here is the honest division of labour:
| Task | Self-serve tools | BotSupport |
|---|---|---|
| Initial setup and connection | You | Us |
| Writing the assistant's instructions and tone | You | Us |
| Configuring how products are presented | You | Us |
| Deciding what the assistant should refuse to guess at | You | Us |
| Branding and styling to match your shop | You | Us |
| Testing against real customer questions | You | Us |
| Reading conversations and spotting failures | You | Us |
| Fixing the failures | You | Us |
| Keeping the catalogue current | You (or a sync you monitor) | Us, automatically |
| Keeping up with model changes and improvements | You | Us |
| Adding new capabilities as they ship | You | Us |
| Your monthly time commitment | Ongoing | Roughly zero |
The reason this matters is not laziness. It is that AI chat is not a fit-and-forget product. An assistant that was excellent in January is mediocre by June, because you added product lines, changed your delivery partner, ran a promotion, discontinued a range and started offering a service you did not offer before. Self-serve tools quietly rot. Someone has to read the conversations and notice that fifteen people this month asked about something the assistant fumbles.
With BotSupport, that someone is us.
We have been asked. The answer is that the difference between an AI chat that customers trust and one they abandon after two questions is almost entirely in the configuration, and configuration is not a thing most merchants have time to become good at. Selling a cheap DIY tier would mean selling a product that mostly fails, and then blaming the customer for not tuning it.
So there is one tier, it includes us, and the price is the price.
Realistically: an hour at the start, to tell us about your business, your tone, your policies and the things you get asked constantly. Then an occasional reply when we send you something worth looking at. That is the honest answer.
The assistant is not limited to WooCommerce and Shopify.
Any website. If your site is not on a supported ecommerce platform, or is not an ecommerce site at all, the assistant can still be built from your published pages: services, policies, guides, FAQs, about pages, anything public. It will answer questions and capture enquiries, without product cards and live pricing.
Custom and headless builds. If your catalogue lives somewhere we can read, we can usually work with it. Tell us your setup and we will tell you honestly whether it is a fit.
Hybrid businesses. Shops that also take bookings, or sell services alongside products, are common and well supported. Catalogue, calendar and enquiry capture all coexist in one assistant.
Magento, BigCommerce, Squarespace, Wix and others. Get in touch. Where we cannot connect live prices and stock, we can usually still build a strong knowledge-and-enquiry assistant, and we will tell you upfront exactly which capabilities you would and would not get.
Early access: £79 per month. Everything included. No setup fee. No contract. Cancel any time.
Locked forever means exactly that. Join during early access and £79 is your price for as long as you stay with us, regardless of what we charge later, and regardless of features we add along the way.
After early access closes, the price is £149 per month. Early access customers are unaffected.
There are none at £79. We do not meter conversations, charge per resolution, or send you an overage bill because you had a good month. If your traffic reaches genuinely unusual volumes we will have a sensible conversation with you rather than a surprise invoice, and no such conversation has been needed yet.
We think per-conversation pricing is a bad model for retail, because it charges you more precisely when things are going well and quietly encourages you to hope fewer customers use it.
Two reasons. First, this is early access, and early customers are helping us make the product better. That has value to us and the price reflects it. Second, we are focused. We do one thing, for one kind of business, extremely well, rather than maintaining a helpdesk, a ticketing system, a social inbox, a voice product and an SMS product.
For context: comparable Shopify AI chat apps commonly start at $79 to $100 per month for capped sessions, before overages, and before you have done any of the configuration work yourself. Fully managed conversational AI for ecommerce more usually starts in the several-hundreds per month.
Better than a trial: we will build you a working demo on your own website's content, free, before you pay anything. Send us your URL. We build an assistant that knows your shop, and you talk to it and see for yourself whether it is any good with your products and your customers' questions. No card, no commitment.
We do it this way because AI chat demos on someone else's fake shop tell you nothing. The only question that matters is whether it is good at your catalogue.
Monthly by card. Cancel any time by emailing us; you are not required to give notice and there is nothing to argue about. Your assistant runs to the end of the paid month and then stops.
Day 0: you send us your URL. We build a free working demo from your website and send you a link.
Day 1 to 2: you try it. Ask it the awkward questions. Ask it the things customers actually ask. Try to catch it out; we would rather you did.
Day 3: kick-off, if you like what you see. A short call or email exchange covering your tone of voice, your policies, the questions you get constantly, the things you never want it to say, and anything specific about how your products should be presented.
Day 3 to 7: we build properly. Live catalogue connection, live prices and stock, secure order lookup, branding, starter prompts, lead capture routing, and tuning against real questions.
Day 7 to 10: you review. We show you the finished assistant on a test page. You tell us what to change. We change it.
Go live: one line of code. A single script tag, added once. If you use WordPress, Shopify, Google Tag Manager or almost anything else, this takes minutes and we will talk whoever does it through exactly where it goes.
Week two onwards: we watch it. We read conversations, spot patterns, and tune. You get told what we found, particularly the parts that are really about your product pages rather than the assistant.
You get access to a dashboard where you can read conversations, see leads, view analytics and adjust the visible settings such as greeting, starter prompts and colours. The deeper configuration stays with us on purpose, because that is what "fully managed" means and it is what stops the assistant degrading. If you want more control than that, say so and we will sort it out.
This is the objection everybody has, and they are right to have it.
A badly built AI chat will make things up. It will confidently describe a product it has not checked, invent a returns window, or dress up a vaguely related item as an exact match. One such answer in front of a customer costs you the sale and probably the customer.
The whole design of BotSupport is a response to that risk. Here is what we do about it, in terms of behaviour rather than plumbing.
It must check before it answers. The assistant is not permitted to answer product questions from general knowledge. It has to consult your actual store information first, every time. If your store does not say it, the assistant does not say it.
It cannot invent a product. Product cards can only be built from real items in your catalogue. It has no way to conjure one.
It cannot claim an attribute a product does not have. It describes each product using the information your store holds about it, and nothing else. This is the specific rule that stops the classic failure where a bot offers a beige stripe and calls it green.
A near miss is labelled a near miss. When nothing genuinely matches, it says it could not find an exact match and offers the closest options as alternatives.
Weak matches are filtered out. Loosely related material is discarded rather than passed off as an answer, because a low-confidence answer delivered confidently is the worst possible outcome.
Out-of-stock is stated, not hidden.
It is allowed to say it does not know, and to hand over to you.
We cannot promise 100% accuracy, and you should be suspicious of anyone who does. What we can promise is that we watch it, that mistakes get found and fixed, and that the assistant is built to fail safe rather than fail confident. In practice the most common cause of a wrong answer is not the AI at all, it is that the information genuinely is not on your website. When we find those, we tell you, and fixing them improves your site for everyone including Google.
Only your website. Your assistant is locked to the domains you authorise, so it cannot be lifted and embedded on someone else's site. Abuse protections and rate limits are in place to prevent scraping and misuse.
By default, as little as possible. Conversation transcripts are stored only if you want them, with a retention period you set (30 days by default), after which they are deleted automatically. We do not require customers to log in, provide personal details or identify themselves to use chat.
Lead form submissions are stored so that you can see them, and delivered to you.
Yes, and it is designed for UK and EU operation. Concretely: no personal data is required to use the assistant, transcript storage is optional and time-limited with retention under your control, visitor IP addresses are not stored in raw form, order lookup requires the customer to prove control of their own email before any order data is disclosed, and we will sign a data processing agreement. We are a UK company and we operate to UK GDPR.
Yes, and it is the part we are strictest about. An order number alone is never enough. The customer must prove control of the email address on the order using a one-time code sent to that address. Verification is enforced by our systems before any order information enters the conversation, so the assistant cannot be talked into revealing an order by an inventive prompt. Proof lasts a limited window and then expires.
No. Your catalogue, your content and your customers' conversations are used to run your assistant. They are not used to train models, and they are not shared with other customers.
The chat is powered by OpenAI, so conversation content is processed by OpenAI in order to generate replies. Everything else runs on our own infrastructure. We can provide full detail for your DPA or security review, including sub-processors, on request.
If the underlying service has a problem, the chat degrades gracefully rather than breaking your site: the widget still loads, and nothing about your storefront, checkout or product pages is affected. We monitor availability continuously, including the health of the underlying models, so if something changes upstream we act on it before it reaches your customers.
Page speed. The assistant loads asynchronously and after your page content, so it does not block rendering and does not damage Core Web Vitals. Your server does no extra work, because none of the processing happens on it.
Mobile. Fully responsive, and designed mobile-first, because that is where most conversational shopping happens. On a phone, chat is a genuinely better interface than a filter sidebar, which is roughly unusable on a small screen.
Design. Your brand colour, your button colours, your text colours, your launcher icon and label, your widget title and greeting, plus your position on the page. Custom CSS is available where you need pixel-level control. It should look like part of your shop, not like a bought-in widget, and if it does not, we will keep working until it does.
Accessibility. Keyboard navigable, screen-reader labelled, and built to standard accessibility practice.
It does not break on update. Because it is one script tag, theme updates, plugin updates, core updates and redesigns do not affect it.
Rule-based bots follow a script a human wrote. They handle the questions someone anticipated and fail on everything else, which is most things. They cannot search your catalogue, cannot understand a description, and produce the "Sorry, I didn't understand" reply that trained a generation of shoppers to close chat windows on sight. BotSupport has no script.
Site search matches strings against product titles and tags. It cannot handle a sentence, does not know synonyms, and returns nothing 10 to 15% of the time. BotSupport reads the request, expands it, searches properly, reads the results, and explains its recommendations. It complements search rather than replacing it; keep your search box.
Humans are better at complex, emotional and high-value conversations. They are also expensive, finite and asleep at 11pm. BotSupport handles the high-volume, repetitive 80% instantly and around the clock, and passes the rest to you with the context attached. Most shops find it makes their humans more valuable, not less, because those humans stop answering "where is my order?" forty times a day.
We do not offer a real-time transfer to a human agent, and that is on purpose. Live chat sounds appealing until you realise what it actually commits you to: someone watching a queue all day, and a widget that quietly makes you look worse every time nobody is there to answer it. AI Chat is meant to give store owners their time back, not to become one more inbox to monitor. So enquiries that need a person come to you as enquiries, with the whole conversation attached, and you deal with them when it suits you.
Helpdesk AI is built for ticket deflection and is generally good at it. It is usually weaker at selling, because its heritage is support, not merchandising, and product recommendation is an afterthought. BotSupport is built for selling first and supporting second. If you already run a helpdesk, the two coexist perfectly well.
Same underlying technology class, very different product. With a self-serve app, you are the implementation team, the QA team and the maintenance team, forever. Most such apps end up neglected within two months. With BotSupport, that work is ours, and the price is comparable to or below the app-store options once you count what you would otherwise pay yourself to run it.
Doing nothing is a real option, and for some businesses it is the right one for now. The cost of it is the failed searches you already have, the tickets you already answer, and the growing gap between what customers expect after using ChatGPT every day and what your website gives them.
We would rather say so now than take your money and disappoint you.
What is BotSupport AI Chat?
A fully managed AI shopping and support assistant for ecommerce websites. It answers customer questions in natural language, recommends real products from your catalogue with live prices and stock, tracks orders securely, and captures enquiries. It is powered by OpenAI and built and maintained for you.
Which AI does it use?
It is powered by OpenAI. We select and manage the configuration for each shop, and we keep it current as OpenAI's models improve, at no extra cost and with nothing for you to do.
Is this just ChatGPT on my website?
No. ChatGPT knows the general internet. Your assistant knows your shop: your products, your prices, your stock, your policies and your orders. It is built to answer from your information rather than from general knowledge, which is exactly what makes it useful and safe on a shop.
Do my customers need an account or a login?
No. They open the chat and type. Nothing to install, nothing to sign up for.
Will it work on mobile?
Yes, and it is designed mobile-first. Chat is a much better interface than filters on a phone.
Can it speak other languages?
Yes. If a customer writes in another language, the assistant replies in that language.
How many conversations can it handle at once?
Unlimited in practice. It does not queue, and it does not slow down at 9pm on Black Friday.
Can it handle rude or abusive messages?
Yes. It stays polite, does not escalate, and does not take the bait. Protections are in place against people trying to misuse it.
Can a customer be put through to a real person in the chat?
Not in real time, and that is a deliberate choice rather than a gap. AI Chat exists to free store owners up, not to add another window somebody has to sit and watch all day. A live-agent button is only as good as whoever is free to answer it, and an unanswered one makes you look worse than having none at all. Instead, when a customer needs a human, the assistant takes their details and sends you the enquiry with the full conversation attached, so you can pick it up properly when it suits you.
So my customers can still reach me?
Yes, always. The assistant offers your contact routes, captures enquiries, and passes them straight to your inbox or CRM. Nobody gets trapped talking to a robot.
How does it know about my products?
We connect it to your store and keep it current automatically. You do not upload anything, and you do not maintain anything.
What happens when I add a new product?
It appears in chat automatically. There is no re-training step for you to remember.
What if I change a price or something sells out?
Prices and stock shown to customers come from your store, so they are current. Out-of-stock items are shown and clearly marked rather than hidden.
Can it handle variable products, sizes and options?
Yes, and this is one of the things we tune per shop. Variants can be grouped onto one card with the options listed, or split into separate cards. We also choose which size a card should feature, so a fabric shop features the metre rather than the sample, and a coffee roaster features the 250g bag rather than the 1kg.
How many products can it show at once?
Up to four or five per reply, chosen as the best matches. Showing twenty options is not helpful; that is what a category page is for.
Can it recommend related products or upsell?
Yes. It can present its direct recommendation first, then a compact "you might also like" group for genuinely different complementary options. This is a per-shop setting and we will advise on whether it suits your catalogue.
What if it cannot find what the customer wants?
It says so plainly and offers the nearest alternatives, clearly labelled as alternatives. It does not pretend a near miss is a match.
Can it link to specific pages on my site?
Yes. Product cards link to product pages, and it links to relevant policy, information and contact pages when they answer the question.
Does it know what page the customer is on?
Yes. If someone on a product page asks "does this come in blue?", it knows what "this" means, and it ignores that context when the customer clearly means something else.
Can customers track their orders in chat?
Yes, on both WooCommerce and Shopify. It is included at £79, not a paid add-on.
Is that secure?
Yes. An order number alone is never enough. The customer must prove control of the email address on the order with a one-time code sent to that address, and only then is any order data released. Verification is enforced by our systems, not left to the AI's judgement.
Can someone guess an order number and see someone else's order?
No, not on the default setting. Without control of the email address on the order, no order data is disclosed.
How long does the verification last?
A short window within the same conversation, so follow-up questions do not force the customer to re-verify, after which it expires. The window is configurable.
Can it collect customer details?
Yes. You choose the fields, including text, email, phone, longer messages, dropdowns and checkboxes.
Where do enquiries go?
To an email address you choose, and onward to a CRM, helpdesk or automation tool if you use one.
Does the conversation come with the enquiry?
Yes. Whoever picks it up sees the full context.
Can I turn lead capture off?
Yes, if you would rather the assistant only ever answered questions.
How long does setup take?
Typically 7 to 10 days from kick-off to live, including the review round. The free demo comes back much faster than that.
How much work is it for me?
About an hour at the start, then occasional replies. That is the point of a managed service.
Do I need a developer?
Only to paste one line of code into your site, which takes minutes. We will talk whoever does it through it, and on most platforms it is a single field in the admin.
Will it break when I update my theme or plugins?
No. It is one script tag and it does not depend on your theme.
Can I change the greeting, colours or starter prompts myself?
Yes, from your dashboard. Deeper configuration stays with us, which is what keeps it accurate.
What if I do not like something it says?
Tell us and we will change it. That is included, unlimited, for as long as you are a customer.
Do I get to see the conversations?
Yes, in your dashboard, if transcript storage is enabled. We strongly recommend it for at least the first few months, because it is the fastest product-and-copy feedback you will ever get.
Can I have more than one shop on it?
Yes. Each shop gets its own assistant. Talk to us about pricing for multiple sites.
How much does it cost?
£79 per month during early access, locked in forever. £149 per month once early access closes. Early access customers keep £79.
Is there a setup fee?
No.
Is there a contract?
No. It is monthly, and you can cancel any time.
What does "locked in forever" mean?
If you join at £79, you pay £79 for as long as you stay with us, including as we add features and even after the standard price rises to £149.
Are there per-conversation or per-resolution charges?
No. No metering, no overages, no surprise invoices for having a busy month.
Is there a free trial?
We do something better. We build you a free, working demo on your own website's content before you pay anything. Send us your URL.
What happens if I cancel?
The assistant runs to the end of your paid month and then stops appearing on your site. Remove the line of code at your leisure. No exit fees, no data held hostage.
Why is it cheaper than other managed AI chat?
Early access pricing, and focus. We do one thing for one kind of business rather than maintaining a helpdesk, a social inbox, a voice product and an SMS product alongside it.
Will it make things up?
It is specifically built not to. It has to consult your real store information before answering, it cannot invent a product, it cannot claim an attribute a product does not have, and it is allowed to say it does not know. No one can honestly promise 100% accuracy, but it is designed to fail safe rather than fail confident, and we monitor and fix.
What if it gives a customer wrong information?
Tell us and we fix it, usually the same day. In our experience most wrong answers trace back to information genuinely missing from the website, and fixing that helps every visitor, not just chat users.
Can it promise a discount or agree to a refund?
Not unless you tell us it should. By default it explains your policies rather than making commitments on your behalf.
Can someone trick it into saying something damaging?
People try. It is built to resist prompt manipulation, it will not disclose order data without verification, and it is scoped tightly to your business rather than being an open-ended chatbot.
Will it hurt my SEO?
No. It does not change your HTML, headings or links, and it does not block page rendering, so it does not affect Core Web Vitals. It tends to improve engagement signals, because visitors who would have bounced from a dead-end search now click through to a product.
Does it help me appear in ChatGPT and AI search?
Indirectly, and meaningfully. Setting it up surfaces every question your published content fails to answer. Filling those gaps with clear, factual, well-structured answers is precisely what earns citations in AI-generated results. Several of our customers have used their chat transcripts as a content roadmap and it is one of the better sources of one we have seen.
Can I see how much revenue it generates?
Yes, for connected stores. Your dashboard shows orders and revenue attributed to AI Chat alongside conversations, messages, product clicks and leads.
Can I get a monthly report?
Yes, emailed to whoever you choose.
AI Chat: a chat interface powered by a large language model that understands natural language rather than following a script.
Conversational commerce: buying and selling through conversation rather than through category pages and filters.
AI shopping assistant: an AI Chat focused on helping customers find and choose products.
WISMO: "Where Is My Order?", the single most common ecommerce support question, typically 20 to 50% of all tickets.
Ticket deflection: a customer question resolved without a human, and therefore without a support ticket.
Zero-results rate: the share of site searches returning nothing. Industry average 10 to 15%; best practice under 5%.
Product card: the rich product panel the assistant shows in chat, with image, name, live price, stock state and a link.
Starter prompts: the suggested questions shown when the chat opens, which teach customers what the assistant can do.
Quick replies: the contextual one-tap follow-up suggestions shown under a reply.
Fully managed: the provider builds, configures, monitors, tunes and maintains the service, rather than handing you a toolkit.
AEO / GEO: Answer Engine Optimisation / Generative Engine Optimisation: structuring content so AI assistants cite it when answering questions.
Hallucination: an AI stating something confidently that is not true. The central risk in retail AI, and the thing our design is built around.
Send us your website address and we will build you a free, working AI Chat demo trained on your own products and pages. No card, no commitment.
Get my free demo & analysisAn expert salesperson for your store.
Launched in 24 hours, fully managed by us.