How to use AI to increase online sales: the order that actually pays.
Most sellers use AI in the wrong order. They start with the exciting part, which is ads and content, while the product page still does not say what the thing is made of. AI is very good at making more of whatever you already have, so if the page cannot hold a visitor, AI helps you lose visitors faster and at a higher cost. This guide puts the work in the order that returns the most for the least effort: words and SEO first, then photos, then the people who nearly bought, then the people who already bought, and paid ads last. Every step has what to do, how to tell whether it worked, and one caution worth reading before you press go.
Why the order matters more than the tool
Each step below feeds the one after it. Better product copy gives search engines and shoppers something to hold on to. Better photos make the copy believable. Both make cart reminders worth sending, because the person who comes back lands on a page that answers them. Only then does a rupee spent on ads have somewhere useful to go, and only then do you have a retargeting audience to spend it on.
Do it the other way round and you pay for traffic that meets a page you have not fixed yet. Nobody can tell you what percentage more you will sell by working in this order, and anyone who does is guessing. What this order does is stop you spending money on the last step before the first four are done.
One rule for every step
AI drafts, you check. Every rewrite, every image and every message on this page is a draft until a human has read it against the real product. That single habit is what separates AI that helps from AI that creates returns, complaints and a store that reads like everyone else's.
Step 1: fix the words on the product pages
What to do. Take your top ten products by views and rewrite three things on each: the title, the first two lines of the description, and the meta description. The title should say what the thing is in the words a buyer would type, not a poetic name. The first two lines should answer the questions people actually ask you on WhatsApp: what is it made of, what sizes, how long does delivery take, can it be returned. The meta description is the line that shows in Google, so lead with what the product is, keep it around 140 to 160 characters, and skip the adjectives.
Do the store title and description too. They are the sentence that decides whether anyone clicks your shop at all, and most stores still have the template default sitting there.
How to judge it. Open Search Console and look at impressions and clicks for those product pages over the next several weeks, not the next several days. Search is slow. The second measure is closer to hand: count the questions arriving in your DMs. If people stop asking about size and delivery, the page has started answering.
The caution
AI will happily invent a material, a size chart or a certification that your product does not have, because the sentence reads better with one. Read every rewrite against the real product before you publish. An invented claim is not a copy problem, it is a returns problem and, for some categories, a legal one.
Step 2: photos, and only then
What to do. Every product needs three images: a clean packshot on a plain background, one photo of the thing in use or being worn, and one social image with the price or the offer written on it. Keep the framing and the background consistent across the catalogue so the shop looks like one shop rather than twelve phone cameras. AI image tools do this from the photo you already took, which is why this step is now cheap enough to sit second.
How to judge it. Compare product page views against add-to-cart for the same products before and after. Use your own store analytics, over a few weeks, and only on the products you changed. Then step back and look at the category page: does it look like a shop you would buy from?
The caution
An AI image tool will redraw your product if you let it. Check the colour, the shape, the logo and any text on the image every single time. If the picture is not exactly the thing in the box, do not use it. That is a promise you cannot keep at delivery.
Step 3: bring back the people who nearly bought
What to do. Someone typed their phone number or email at checkout and then left. That is the warmest audience you will ever have and it costs nothing to reach. Capture the contact as it is typed, send one reminder about an hour later while the intent is still alive, and one the next day. Put a link in it that restores the cart, so coming back takes one tap. Then stop. Two messages, not five.
How to judge it. This is the one step with a clean measure. Count the orders that came back through a reminder link. It is a real number, not an inference, and it either exists or it does not.
The caution
You are messaging someone who did not buy. Honour every unsubscribe and every STOP reply, immediately and permanently. On WhatsApp, marketing messages need a Meta-approved template, so this is not something you can improvise from your own phone. And do not add a discount to the first reminder by reflex; you will teach regular buyers to abandon carts on purpose.
Step 4: sell again to the people who already bought
What to do. Your past buyers and your newsletter subscribers are one list, and it is the cheapest sales channel you own. Message it when there is a real reason: a restock, a new size, a festival, a genuinely new product. Write in the language you sell in. If you also serve people at a counter, a stamp card or loyalty programme does the same job offline, tying a repeat visit to a reward the customer can see.
How to judge it. Repeat orders as a share of total orders, month over month, read from your own order list. It moves slowly and it is worth watching because it is the number that decides whether the business compounds or starts from zero every month.
The caution
This list is the most valuable thing you own and it is easy to burn. A promotion every week trains people to ignore you, and by the time you have something worth announcing nobody opens it. Never buy a list, and never message people who did not give you their contact for this.
Step 5: paid ads, last
What to do. Before you spend anything, work out your break-even return on ad spend from your own margin, so you know the number at which a campaign is losing money. Send your catalogue to Meta and Google as a product feed so you can advertise real products and retarget people who viewed them. Then run a small test with a written rule for when you stop and when you scale, decided before the money starts moving. AI is genuinely useful here for planning: the products to push from your real best sellers, the hooks, the audiences, the schedule.
How to judge it. Your own order count against your own spend, over a full week at least. Platform-reported conversions are a useful signal and a generous one; the count in your dashboard is the one that pays rent. Server-side purchase events make the platform's number less lossy, which is worth setting up, but it does not make it your number.
The caution
AI can write a campaign plan. It cannot promise a return, and neither can we. It also does not spend the money for you, which is a feature: an AI with your card details and no kill rule is a bad night. Set the stop rule first, and start smaller than feels serious.
If you only have an hour
Most sellers reading this have an hour on a Sunday, not a week. Here is the hour, in order, and then genuinely stop.
- Rewrite the five products that get the most views. Title, first two lines, meta description. Check every line against the real product.
- Fix the store title and description. One sentence that says what you sell and to whom. It is the line Google shows.
- Make one good photo of your best seller. A clean packshot. Use it everywhere.
- Switch on cart reminders. One at an hour, one the next day, with the cart restored in the link.
- Write down your break-even return on ad spend. Not to spend today; so that when you do, you know the number.
That is the hour. Nothing in it involves an ad account, and every part of it keeps working while you sleep.
What each step is on SitesPlaced, and what it costs
The SitesPlaced AI Co-founder plan is ₹1,499 a month, includes everything in the ₹499 Growth plan, and comes with 1,500 AI credits a month. One credit is ₹1 of face value, and every action shows its price before you press the button. Here is the same order of operations mapped onto it.
| Step | What you use | Credits |
|---|---|---|
| 1. Words and SEO | Beacon, the SEO agent: a store-wide audit with a score, buyer keywords, a store title and description, and a meta description per product that you tick to apply. Product copy rewrites run in bulk across the catalogue. | Audit 5, copy rewrite 1 per product, agent message 1 |
| 2. Photos | The creative studio: 12 styles and 5 aspect ratios, built from the product's own photo, with the product kept identical. | Draft 2, standard 12, premium 25, ultra 45 |
| 2b. Social posts | Instagram carousels of 3 to 7 slides, with the slide copy, caption and hashtags written for you. | 12 per slide |
| 3. Near-buyers | Abandoned-cart reminders: email at 1 hour and again at 24 hours, plus WhatsApp through a Meta-approved template once your Cloud API is connected. | Nothing per use |
| 4. Past buyers | Broadcasts from the dashboard to your newsletter subscribers and past buyers. Scheduled promotions put a coupon and an announcement bar on a timer. | Nothing per use |
| 4b. Shopper questions | An optional storefront assistant that answers from your own catalogue and policies. Off by default; you switch it on. | 1 per shopper message |
| 5. Paid ads | The campaign planner turns a goal and a daily budget into products from your real best sellers, break-even maths, hooks, audiences and a 7-day schedule with kill and scale rules. The retargeting kit gives you a Meta CSV and a Google XML product feed. | Plan 5, feeds nothing |
| Every week | A Monday digest of what actually happened, and Ledger, the numbers agent, to ask why. | Digest nothing, agent message 1 |
Failed generations are refunded automatically. The full price table and three worked example months are in AI credits explained.
You do not have to start at ₹1,499. The ladder is: Free to publish a real store with 0% commission, Starter at ₹199 a month when you want your own name on it, Growth at ₹499 a month for the selling tools, and AI Co-founder at ₹1,499 a month when you want the work above done with you. Steps 1 to 5 are the right order on any of them.
If you would rather use the assistant you already pay for, the AI Co-founder is also reachable from ChatGPT or Claude through the SitesPlaced MCP connector, where reading your numbers, rewriting your SEO, saving it and scheduling a promotion cost no credits at all. That is written up here.
What AI does not do here
- It does not reply to WhatsApp messages that arrive on your own number. That is not live on SitesPlaced. If it is the thing you need this month, we do not have it.
- It does not make video. Also not live, and also worth saying out loud rather than in a footnote.
- It does not promise a number. No honest tool can tell you how much more you will sell. What it can do is take the writing, the photographing and the remembering off your evening.
- It does not replace looking. The steps above all end in a human check, because every one of them can be wrong in a way that costs you a customer.
FAQs
What is the single highest-return AI task for a small store?
Rewriting product copy and the SEO that goes with it. It costs the least, it works on every visitor who arrives from now on, and it makes every later step better: photos, cart reminders and ads all land on a page that finally says what the product is. Start with the ten products that get the most views.
Should I use AI for ads first?
No. Ads multiply whatever your page already does, so paying for traffic before the product page answers basic questions means paying to lose visitors faster. Do the copy, the photos and the cart reminders first, then advertise, and know your break-even return on ad spend before you start.
Do abandoned-cart reminders annoy customers?
Not at two messages. Send one about an hour after they leave and one the next day, with a link that restores the cart, and then stop. Honour every unsubscribe and every STOP reply immediately. On WhatsApp, marketing messages go through a Meta-approved template, so they cannot be improvised. Adding a discount to the first reminder is usually a mistake, because it teaches people to abandon carts on purpose.
How do I know an AI rewrite is safe to publish?
Read it against the real product before it goes live. The failure to look for is an invented fact: a material, a size chart, a certification or a delivery promise that your product does not actually have. AI adds these because they make the sentence read better. Everything else in a rewrite is a matter of taste; an invented claim is a returns problem.
What does all of this cost on SitesPlaced?
Publishing a real store is free with 0% commission. Starter is ₹199 a month, Growth is ₹499 a month, and the AI Co-founder plan is ₹1,499 a month including everything in Growth plus 1,500 AI credits, where one credit is ₹1 of face value. An SEO audit is 5 credits, a product copy rewrite is 1, images are 2 to 45 by quality, a carousel slide is 12 and a campaign plan is 5. Cart reminders, broadcasts, scheduled promotions and the Monday digest cost nothing per use.
Do the first four steps this week.
SitesPlaced AI Co-founder is ₹1,499 a month with 1,500 credits and everything in the Growth plan: the SEO audit, bulk product copy, the creative studio, cart reminders, broadcasts, scheduled promotions and a campaign planner for when you are ready to advertise. Every action shows its price first, and nothing reaches your live store without your approval.
See the AI Co-founder →
